LAWS2075 M1 Learning Plan 2024 PDF

LAWS2075 ======== AI Regulation and Society ========================= Module 1** **The Complexities of AI Regulation =============================================== https://309w5s255371fs4df2vftgbl-wpengine.netdna-ssl.com/wp-content/uploads/2017/09/placeholder.jpg**\ \ ** Module Overview --------------- Theme: Understanding the Foundations This module introduces students to the fundamental complexities of AI technologies, including their technical intricacies, security challenges, and potential biases. By establishing a foundational understanding of AI, students will be prepared to explore the regulatory landscape and its implications in subsequent modules. Module Purpose -------------- By the end of this module, you will be able to: 1\. Analyse the technological complexities of AI systems, including machine learning algorithms, neural networks, and foundational models, and explain how these complexities pose challenges for effective regulation. 2\. Appraise the security risks associated with AI systems, such as adversarial attacks, data poisoning, and model inversion, and discuss the regulatory approaches needed to mitigate these risks. 3\. Assess the potential biases in AI systems, including algorithmic, data, and human biases, and propose regulatory strategies to ensure fairness and non-discrimination in AI applications. 4\. Evaluate the challenges of AI governance frameworks that can balance AI innovation with current and future societal concerns, such as job displacement, hyper-surveillance, and technocratic governance. Introduction ------------ This module lays the foundation for understanding how AI systems function and the variety of challenges these technologies present for society. We start by exploring key concepts like machine learning, the distinctions between narrow and general AI, and the essential components of AI systems. By understanding these core ideas, students will be better equipped to navigate the complexities of AI regulation that arise in subsequent modules. This fundamental knowledge provides a technical understanding and frames the discussion around why regulating AI poses unique challenges, from addressing biases to ensuring accountability in complex and opaque systems. As AI continues to evolve, its impact on society will deepen, making it essential for future legal professionals, policymakers, and technologists to grasp these foundational aspects to advocate for responsible AI development and regulation. 1. Introduction to AI Technologies ---------------------------------- Artificial Intelligence (AI) has become a transformative force in our lives, revolutionising how we approach complex problems and interact with technology. To understand the complexities of AI regulation, it\'s crucial first to grasp AI technologies\' fundamental concepts and components. Lawyers need to understand technology, particularly AI, for several key reasons that impact their practice and their ability to serve clients effectively: **1. Legal Implications of AI** AI technology raises numerous legal issues, such as data privacy, intellectual property rights, liability concerns, compliance with regulations like the Privacy Act 1988 (Cth), anti-discrimination laws, and the raft of AI-focused laws domestically and internationally. Lawyers must grasp the underlying technology to offer sound legal advice on these emerging issues, ensuring clients navigate the regulatory landscape without exposure to legal risks. **2. Changing Legal Practice** AI is transforming the legal industry, automating routine tasks like document review and enhancing legal research with AI-powered tools. Understanding AI enables lawyers to leverage these technologies to increase efficiency, reduce costs, and improve accuracy in their practice. It also helps them stay competitive as clients increasingly expect modern, tech-savvy legal services. **3. Client Expectations and Risk Management** Many clients, particularly in the technology sector, expect their legal advisors to understand the technologies that are central to their business. A lawyer who can engage meaningfully with AI-driven issues is better positioned to anticipate and manage the risks clients face, especially in areas such as cybersecurity, data protection, and the ethics of AI deployment. **4. Ethical and Professional Responsibility** Lawyers have a duty to maintain competence, including understanding how technological advances affect their work. In jurisdictions such as Australia and the U.S., legal ethics rules increasingly emphasise technology competence as part of a lawyer's duty. Knowledge of AI helps lawyers meet these professional standards and avoid potential malpractice claims or disciplinary actions related to inadequate technological understanding. **5. AI in Litigation and Dispute Resolution** AI tools are increasingly used in litigation for tasks like predictive coding in eDiscovery, pattern recognition, and even legal decision modelling. Lawyers who understand how these tools work can better advise their clients, challenge opposing parties\' use of AI, and present more persuasive cases when AI is involved. AI is also raising questions about evidence admissibility, algorithmic transparency, and bias, all of which are critical to litigation. **6. AI and Fundamental Rights** The use of AI impacts fundamental rights such as privacy, freedom of expression, and non-discrimination. Lawyers must understand how AI technologies can infringe on these rights to protect individuals and organisations from potential harm. For example, biased algorithms in decision-making processes can lead to unfair treatment in areas such as employment or lending, requiring legal redress. **7. Cybersecurity and Data Protection** As AI becomes more embedded in business processes, it presents new cybersecurity challenges, including vulnerabilities to cyber-attacks and the potential for AI systems to be exploited or manipulated. Lawyers must understand these risks to advise clients on best practices for AI security and compliance with cybersecurity standards. Understanding AI enables lawyers to provide better client service, navigate the evolving legal landscape, maintain professional competence, and mitigate legal and ethical risks. ### Activity: A timeline of key developments in the history of AI \[Insert Text\] \[H5P -- Interactive Timeline\] 1940s-1950s: Foundations of AI 1943 -- McCulloch & Pitts Neural Network Model: Warren McCulloch and Walter Pitts propose the first artificial neuron model, laying the groundwork for neural networks. 1950 -- Turing Test: Alan Turing publishes \"Computing Machinery and Intelligence\" and introduces the Turing Test, proposing a test for a machine's ability to exhibit intelligent behaviour indistinguishable from humans. 1956: The Birth of AI 1956 -- Dartmouth Conference: John McCarthy, Marvin Minsky, and others coin the term \"artificial intelligence\" at a conference, marking the official beginning of AI as a field of study. 1960s: Early AI Programs 1961 -- Unimate Robot: The first industrial robot, Unimate, is deployed in a General Motors assembly line, signalling the first use of AI in manufacturing. 1966 -- ELIZA: Joseph Weizenbaum develops ELIZA, an early natural language processing program that simulates conversation. 1970s: AI Winter Begins 1972 -- PROLOG: The programming language PROLOG, designed for AI applications like expert systems, is developed. 1973 -- AI Winter: Funding for AI research is drastically reduced due to limited progress, beginning the first \"AI Winter\" period of reduced interest and investment. 1980s: Expert Systems and Renewed Interest 1980 -- Expert Systems Boom: AI regains momentum with the rise of expert systems, programs that emulate decision-making processes of human experts. 1986 -- Backpropagation Algorithm: Geoffrey Hinton and colleagues rediscover backpropagation, a key algorithm for training neural networks, which revitalises interest in machine learning. 1990s: Machine Learning and Deep Learning Foundations 1997 -- Deep Blue vs. Kasparov: IBM's Deep Blue defeats world chess champion Garry Kasparov, demonstrating AI's potential in complex decision-making. 1998 -- LeNet Convolutional Neural Network: Yann LeCun develops LeNet, a convolutional neural network for image recognition, laying the foundation for modern deep learning. 2000s: Rise of Big Data and Modern AI 2006 -- Deep Learning Emergence: Geoffrey Hinton and others pioneer deep learning techniques, using large datasets and powerful computing to improve AI performance significantly. 2009 -- Google Brain: Google begins the \"Google Brain\" project, applying large-scale deep learning to data-driven tasks like image and speech recognition. 2010s: AI in Everyday Life 2012 -- AlexNet: Alex Krizhevsky et al. develop AlexNet, a deep convolutional neural network that wins the ImageNet competition and propels AI into the mainstream. 2016 -- AlphaGo Defeats Lee Sedol: DeepMind's AlphaGo defeats Go champion Lee Sedol, marking a milestone in AI's ability to tackle highly complex, intuitive tasks. 2017 -- Transformer Model: Google researchers introduce the Transformer model, revolutionising natural language processing and paving the way for large language models like GPT. 2020s: Generative AI and AI Ethics 2020 -- GPT-3: OpenAI releases GPT-3, a powerful language model capable of generating human-like text, setting new standards for natural language processing. 2023 -- ChatGPT & Generative AI: The rise of generative AI models, including OpenAI's ChatGPT, brings AI-powered content generation and automation to widespread use in various sectors, prompting discussions on AI ethics and regulation. ### Types of Artificial Intelligence Artificial Intelligence refers to the simulation of human intelligence in machines that are designed to think, learn, and solve problems autonomously. It encompasses a wide range of computational techniques that allow machines to perform tasks traditionally requiring human cognition. These tasks include visual perception, speech recognition, decision-making, and language translation. AI is not a single technology but a collection of subfields (specialised domains), such as: - **Generative AI**: Creates new content---such as text, images, music, or code---by learning patterns from existing data and generating outputs. - **Machine Learning**: Systems that learn from data, identify patterns, and make decisions without being explicitly programmed for specific tasks. - **Natural Language Processing** (NLP): Enables machines to understand, interpret, and generate human language. - **Computer Vision**: Allows machines to interpret and make decisions based on visual data. - **Expert Systems**: Simulates human decision-making through rule-based models. - **Robotics**: Embeds AI systems in physical machines to interact with the environment. ### Categories of AI: Capabilities and Functionality Capabilities and Functionalities are two important ways to categorise Artificial Intelligence (AI), providing insight into how advanced an AI system is and how it works. In short, Capabilities describe the breadth and depth of intelligence AI can possess---ranging from performing singular tasks to hypothetically surpassing human intelligence. Functionalities, on the other hand, describe how AI operates based on whether it can learn from experience, interact socially, or even exhibit self-awareness. Both dimensions help us understand the current landscape of AI technology and envision future possibilities. Let's dive deeper into each of these categories. ### AI Capabilities (Degree of Intelligence) Capabilities refer to the level of intelligence an AI system has and how broadly it can apply that intelligence. There are three key categories: - **Narrow AI (Weak AI):** AI systems designed to perform a specific task, such as playing chess or recommending products on an e-commerce platform. Most AI applications in use today are narrow AI systems. - **General AI (Strong AI):** Hypothetical AI systems that possess generalised cognitive abilities similar to a human. - **Super AI**: This theoretical AI would surpass human intelligence in all aspects. **Narrow AI** (also called Weak AI) refers to artificial intelligence systems designed to perform a specific or limited range of tasks. Unlike general AI, which aspires to mimic human intelligence across various activities, narrow AI is highly specialised, functioning within predefined boundaries. Examples of narrow AI include virtual assistants like Siri or Alexa, recommendation algorithms on platforms like Netflix or Amazon, and facial recognition technology. These systems can excel at their specific tasks---often outperforming humans in terms of speed or accuracy---but they cannot transfer their \"intelligence\" to other domains or tasks outside of what they were designed for. Narrow AI operates based on algorithms and data-driven models, learning patterns within a limited scope but lacking true understanding or consciousness. For instance, a language translation AI can process and translate millions of sentences accurately, but it doesn't "understand" the nuances of language the way a human does. The term Weak AI doesn't imply that these systems are ineffective; rather, it highlights their limited scope of functionality. Despite its limitations, narrow AI is widespread and integral to many industries, including healthcare (for diagnostic tools), finance (for fraud detection), and legal practices (for contract analysis and document review). **Strong AI**, also known as **General AI** (Artificial General Intelligence, AGI), is an advanced form of artificial intelligence that can perform any intellectual task a human can with the same level of understanding, reasoning, and learning capacity. Unlike Narrow AI, which is designed to execute specific tasks, General AI can adapt to new tasks and situations without needing pre-programming or task-specific training. A key characteristic of Strong AI is its ability to exhibit consciousness, self-awareness, and autonomous reasoning. It would be capable of learning and understanding concepts across various domains, making independent decisions and solving problems in any field---similar to human cognitive abilities. This form of AI remains largely theoretical and has not yet been achieved. The development of General AI raises profound questions about ethics, societal impact, and control. If achieved, General AI could revolutionise industries, economies, and even daily life. Still, it poses potential risks, such as job displacement, loss of privacy, and the challenge of ensuring such systems align with human values. Understanding the difference between narrow and general AI is crucial for lawyers and policymakers, as it informs legal frameworks around accountability, data privacy, and ethical AI deployment. While narrow AI presents fewer existential risks than general AI, it still raises significant regulatory and ethical concerns, especially as it becomes more integrated into decision-making processes. And, because of its far-reaching implications, General AI is a key focus of ongoing research and debate in AI ethics, law, and governance. **Super AI**, also referred to as ASI, is a hypothetical level of AI that surpasses human intelligence in all areas. As imagined by fantastical science fiction stories, ASI is AI, such as HAL from 2001: Space Odyssey and Skynet from The Terminator. Some consider the development of such advanced AI the ultimate goal of AI research, while others view it with caution due to its potential implications. ### AI Functionalities (How It Operates) **Functionalities** describe the **way an AI system functions or behaves**, based on the complexity of its operation and interaction with the environment. There are four main categories: 1. **Reactive Machines**: These are the simplest form of AI. **Reactive AI** systems respond to specific inputs with predefined outputs. They do not have memory or the ability to use past experiences to influence future decisions. For instance, **IBM\'s Deep Blue** chess-playing computer is a reactive machine---capable of evaluating the chessboard's current state but not learning from previous games. 2. **Limited Memory AI**: These AI systems can learn from past experiences to a limited extent. They can retain historical data for a short period, which informs decision-making. For example, self-driving cars use Limited Memory AI to observe other vehicles and pedestrians and adjust accordingly based on stored past data. 3. **Theory of Mind AI**: Still largely theoretical, this type of AI would be able to understand human emotions, beliefs, and intentions. It would recognise that humans have thoughts and emotions that influence their actions. This level of AI would be crucial for human-like interaction, but it has not yet been realised. 4. **Self-aware AI**: This is the most advanced and speculative stage. **Self-aware AI** would possess **consciousness and self-awareness**, allowing it to make autonomous decisions, reflect on its own thoughts, and potentially have emotions. This kind of AI is still a far-off concept and exists more in the realm of theory and speculation than reality. ### Key Components of AI Systems: Algorithms, Data, and Computational Power AI systems rely on three foundational components: **algorithms**, **data**, and **computational power**. Each of these play a critical role in determining the efficiency, accuracy, and scalability of AI applications. **1. Algorithms** Algorithms form the backbone of AI systems. They are the mathematical models and rules that define how an AI system will process data, learn from it, and make decisions. Some of the most common AI algorithms include Supervised Learning, Unsupervised Learning, and Reinforcement Learning. **2. Data** The effectiveness of AI systems hinges on the quality and quantity of the data they process. AI models are typically trained on vast amounts of data, and the availability of large datasets is one of the key factors driving the success of modern AI technologies. However, data can also introduce challenges such as bias, privacy concerns, and data scarcity in certain domains. - **Data quality** is paramount because poor-quality data can lead to inaccurate predictions and unintended biases. For instance, a facial recognition system trained predominantly on images of light-skinned individuals may perform poorly on people with darker skin tones. - **Big Data** has become synonymous with AI, as the rise of social media, e-commerce, and IoT devices has led to the generation of massive datasets. These datasets provide AI systems with the diverse input required to function effectively, though managing and processing such large volumes of data poses its own challenges. **3. Computational Power** The third key component of AI systems is computational power. Advances in hardware, particularly Graphics Processing Units (GPUs) and Tensor Processing Units (TPUs), have enabled AI systems to process vast amounts of data in real-time. Computational power is critical for training deep learning models, which require enormous processing capability due to their complexity and size. - **Parallel Processing**: Modern AI systems often rely on parallel processing, where multiple calculations are performed simultaneously to speed up the training process. This is essential for deep learning models that involve billions of parameters. - **Cloud Computing**: Cloud platforms like Google Cloud, AWS, and Microsoft Azure have democratised access to high-performance computing resources, enabling even small organisations to develop and deploy sophisticated AI models without needing to invest in costly hardware. ### Watch: IBM Technology This video is from IBM Technology. It reviews the different types of AI based on capabilities and functionality. IBM Technology: The 7 Types of AI - And Why We Talk (Mostly) About 3 of Them https://www.youtube.com/watch?v=XFZ-rQ8eeR8 (6:49 min) ### Defining Artificial Intelligence There are many definitions of AI, and they usually reflect the primary purpose for which the author requires the definition. Two commonly cited and used definitions of AI are from the Organisation for Economic Co-operation and Development (OECD) (Council on Artificial Intelligence) and the International Organisation for Standardization (ISO). Other definitions have been developed by the (National Institute of Standards and Technology (USA) and through various regulations worldwide. The current working definition that the Australian government has adopted in its 'Policy for the Responsible Use of AI in Government' (see \) and many other world governments, including the EU, is the OECD definition of an 'AI System', current as of 2023; *An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it received, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment.* (See OECD (2024) 'Explanatory Memorandum on the Updated OECD definition of an AI System, OECD Artificial Intelligence Papers, no 8. See https://www.oecd.org/en/publications/explanatory-memorandum-on-the-updated-oecd-definition-of-an-ai-system\_623da898-en.html) This definition 'describes characteristics of machines considered to be AI, including what they do/how they are used but also how they are built.' It is a broad definition that covers AI systems that range from simple to complex. (See Grobelnik, Perset, Russell 'What is AI? Can you make a clear distinction between AI and non-AI systems?' (6 March 2024) https://oecd.ai/en/wonk/definition) The International Organization for Standardization (ISO) provides standardised terms and definitions for AI technology in ISO/IEC 22989:2022. In Section 3.1 - Terms related to artificial intelligence, it states that 'Artificial Intelligence' is an: *Engineered system set of methods or automated entities that together build, optimize and apply a model so that the system can, for a given set of predefined tasks, compute predictions, recommendations, or decisions.* See: ISO/IEC DIS 22989(en) Information technology --- Artificial intelligence --- Artificial intelligence concepts and terminology \. Note that ISO/IEC DIS 22989(en) is a broad technical document with many terms and definitions, including a specific definition for 'AI systems'. ### Key Differences Between ISO and OECD Definitions The specific differences in the approaches to defining AI are noted here: **ISO** **OECD** ------------------------ ---------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------ **Context of Use** The ISO definition is used mainly to standardise AI\'s technical components and operations, aiding developers and engineers in creating interoperable systems. The OECD definition guides policymakers and governments in shaping laws, regulations, and ethical guidelines around AI deployment. **Scope of Focus** The ISO definition is technical, focusing on how AI systems function (interpreting data, learning, and adapting to achieve goals). The OECD definition is policy-oriented, concentrating on AI\'s effects (making predictions, recommendations, or decisions that impact environments). **Learning Aspect** The ISO explicitly mentions that AI systems learn from data and adapt flexibly. The OECD does not emphasise learning but instead focuses on the ability to influence decisions in human-defined contexts. **Application Domain** The ISO focuses on the functional design of AI systems and their adaptability in performing tasks. The OECD provides a more societal and governance-oriented view, acknowledging how AI systems affect real-world environments and decisions. See Gulley and Hilliard, *Lost in Transl(A)t(I)on: Differing Definitions of AI,* (19 February 2024) \. These differences reflect the distinct roles the two organisations play in influencing AI. While ISO focuses on technical standardisation, the OECD emphasises AI development\'s societal and ethical implications. ### Why Lawyers Should Know Definitions of Artificial Intelligence It is important for lawyers to understand the different definitions of AI because these definitions shape how AI technologies are regulated, interpreted, and applied across various legal contexts. The distinctions between technical and policy-oriented definitions of AI can have significant implications for legal practice, particularly in areas such as compliance, liability, intellectual property, and data privacy. Here are several reasons why lawyers need a nuanced understanding of AI definitions: **1. Regulatory Compliance and Legal Interpretation** AI technologies are subject to evolving regulatory frameworks at national and international levels. Different definitions of AI, such as those provided by the ISO and OECD, influence how these regulations are crafted and enforced. For instance: **A technical definition**, like that from the ISO, helps lawyers understand the operational aspects of AI systems, which is crucial when advising clients on compliance with technical standards or industry-specific regulations. **A policy-driven definition**, like the OECD's, focuses on AI\'s societal impact, guiding lawyers in understanding broader regulatory frameworks governing issues such as ethical AI use, fairness, and transparency. Understanding how AI is defined in various contexts helps lawyers advise clients on navigating legal risks, complying with relevant standards, and interpreting legislation correctly. **2. Liability and Risk Management** AI systems introduce new legal risks, especially in terms of accountability and liability for decisions made by AI. Different definitions of AI may influence who is held responsible when AI systems cause harm or when outcomes are unfair or discriminatory: ISO's technical definition, which emphasises a system's ability to adapt and learn, may inform legal arguments about AI\'s inherent unpredictability. This could be relevant in determining product liability or negligence when AI systems operate autonomously. OECD's focus on human-defined objectives suggests that responsibility for AI decisions may lie with the entity that set those goals or deployed the AI system rather than the system itself. This is key for understanding corporate liability or regulatory breaches. Lawyers need to grasp how the definitions of AI affect responsibility and risk distribution when something goes wrong. **3. Contracts and Intellectual Property** The definition of AI also plays a significant role in contract drafting, intellectual property (IP) law, and software licensing. For example: Contracts involving AI technologies may need to specify which type of AI is being developed, licensed, or utilised. This includes whether the AI involves machine learning (with self-improving capabilities) or rule-based systems. In IP law, AI-generated outputs raise questions about ownership and authorship, particularly when AI learns and adapts (as in the ISO definition) versus when it follows pre-programmed human objectives (as in the OECD definition). Understanding the nuances of AI definitions can ensure that legal agreements adequately address ownership rights, liability, and usage terms for AI technologies. **4. Data Privacy and Security** The deployment of AI systems raises critical data privacy and security concerns, particularly under frameworks like the General Data Protection Regulation (GDPR) or Australia's Privacy Act. Different definitions of AI impact how privacy laws apply to AI systems: ISO's focus on machine learning and data-driven adaptation highlights issues of data usage, storage, and consent---key areas of concern in privacy law. OECD's emphasis on decision-making and predictions underscores the potential for AI systems to make sensitive inferences about individuals, which could trigger legal protections under privacy laws. Lawyers need to understand how different AI systems interact with personal data to help clients navigate privacy regulations and mitigate security risks. **5. Bias and Fairness in AI** AI systems can perpetuate biases based on the data they are trained on, which can result in discriminatory outcomes in fields such as hiring, lending, and law enforcement. Different definitions of AI shape how bias and fairness are understood and addressed: The ISO definition, with its focus on system adaptability and learning from data, suggests that AI systems could evolve in ways that reflect historical biases in the data. The OECD definition emphasises human-defined objectives, which suggests that bias may arise from the goals or parameters set by humans who deploy the AI. Lawyers dealing with cases involving discrimination, fairness, or ethics in AI must understand how these definitions relate to algorithmic bias and how to argue for fairness in the use of AI technologies. **6. AI Governance and Ethical Considerations** AI governance frameworks are being developed globally to ensure that AI technologies are used responsibly. Different definitions of AI influence how these frameworks are structured: ISO's technical definition informs the creation of governance structures around safe deployment, system integrity, and compliance with operational standards. OECD's policy definition shapes governance frameworks emphasising accountability, transparency, and ethical concerns such as job displacement and societal impacts. Understanding these definitions allows lawyers working in fields such as AI ethics, public policy, or corporate governance to advise on the creation of AI governance structures that align with ethical principles and legal standards. **7. Adapting to Future Legal Frameworks** AI regulation is still in its formative stages, with many jurisdictions crafting AI-specific laws. Legal definitions of AI will shape future litigation, compliance, and regulatory strategies. By understanding different AI definitions, lawyers will be better positioned to anticipate future legal challenges, such as regulating autonomous decision-making systems or handling AI-generated content. Lawyers can help shape AI policy debates by advocating for definitions that protect human rights, data privacy, and ethical standards while promoting technological innovation. **A Better Understanding Makes for Better Lawyering** For lawyers, understanding the different definitions of AI is not merely academic---it is foundational to advising clients, mitigating risks, drafting contracts, and ensuring compliance with evolving legal standards. As AI becomes more integrated into daily life and business, its definitions will significantly impact legal practice across sectors, requiring lawyers to stay informed about both technical and policy-oriented perspectives. ### Activity: Title? \[H5P -- Multiple Choice Questions\] **Question 1: Which of the following best reflects the key difference between the ISO and OECD definitions of AI?** A. The ISO definition focuses on AI\'s ability to influence real or virtual environments, while the OECD definition emphasises AI\'s technical ability to interpret and learn from data. B. The ISO definition focuses on the technical nature of the AI and AI systems, while the OECD definition focus is policy-oriented, concentrating on AI\'s effects. C. The ISO definition is primarily concerned with AI's ethical implications, while the OECD definition focuses on its economic impact. D. The OECD definition emphasises machine learning, while the ISO definition stresses AI\'s ability to operate autonomously. Correct Answer: B. The ISO definition focuses on the technical nature of the AI and AI systems, while the OECD definition focus is policy-oriented, concentrating on AI\'s effects. **Question 2: Why is it important for lawyers to understand the different definitions of AI?** A. To improve their programming skills and better understand how to build AI systems. B. To assess which AI tools can be used to increase their personal productivity in legal practice. C. To determine the potential legal liabilities, compliance obligations, and ethical considerations associated with AI use. D. To decide which AI systems will be the most profitable investments for their law firm. Correct Answer: C. To determine the potential legal liabilities, compliance obligations, and ethical considerations associated with AI use. #### Subsets of AI Artificial Intelligence encompasses a broad field of computer science with the goal of creating systems that can perform tasks traditionally requiring human intelligence. As with most things, there are many ways of achieving a desired goal. Within AI, specific subsets form a hierarchical structure, each representing a more specialised form of AI. At the top level is AI, the overarching domain that includes various approaches and technologies. A key subset (method) of AI is Machine Learning (ML), where systems learn from data and improve over time without being explicitly programmed for every task. Within ML lies a further specialised subset called Deep Learning, which uses artificial neural networks to process vast amounts of data and recognise patterns at multiple levels of abstraction. The Australian Signals Directorate: Convoluted Layers: An Artificial Intelligence (Ai) Primer (https://www.cyber.gov.au/sites/default/files/2023-11/%28OFFICIAL%29\_Convoluted\_layers\_-\_an\_artificial\_intelligence\_%28AI%29\_primer.pdf) These subsets represent not just categories but a progression in complexity and capability. Numerous approaches and models---such as decision trees, support vector machines, and neural networks---are employed within these subsets to solve specific problems, highlighting AI\'s versatility and layered nature. #### Specialised domains of Artificial Intelligence AI encompasses a range of specialised domains, each focusing on distinct capabilities and applications of intelligent systems. **Generative AI**, for instance, is concerned with creating new content---such as text, images, or music---by learning patterns from existing data. **Machine Learning** (ML), another core domain, enables AI systems to identify patterns in data and make predictions or decisions without being explicitly programmed for each task. **Natural Language Processing** (NLP) focuses on enabling AI to understand, interpret, and generate human language, facilitating applications like chatbots and language translation. On the other hand, **Computer Vision** allows AI to interpret and process visual data, such as recognising objects or analysing images and videos. Each of these domains employs a variety of approaches and techniques---ranging from neural networks to probabilistic models---tailored to the specific tasks and challenges they aim to address. **Specialised domains of Artificial Intelligence** ![](media/image3.png) (See: https://www.digital.nsw.gov.au/policy/artificial-intelligence/a-common-understanding-simplified-ai-definitions-from-leading) ### Generative AI Generative AI is a category of artificial intelligence systems designed to generate new content based on patterns learned from existing data, such as text, images, music, or video. Unlike traditional AI, which primarily focuses on recognising patterns or making predictions, generative AI can create novel outputs that mimic human-like creativity. These (**foundational**) models have revolutionised content creation, design, and natural language processing by producing outputs resembling human-generated work. **Large Language Models (LLMs):** A prominent approach in generative AI is the use of Large Language Models (LLMs), such as OpenAI's GPT (Generative Pre-trained Transformer). LLMs are trained on vast amounts of text data and use deep learning, specifically transformer architecture, to understand and generate coherent, contextually relevant text. By processing millions of sentences from books, websites, and other sources, LLMs learn patterns in grammar, meaning, and structure, allowing them to produce human-like responses, complete sentences, or even generate entire documents. LLMs are particularly useful in applications such as chatbots, automated writing, and summarisation tools, where fluency and understanding of context are critical. **Multimodal Foundation Models:** Another advanced approach in generative AI is the Multimodal Foundation Model, which integrates multiple types of data---such as text, images, and audio---into a single framework. These models, like OpenAI's DALL-E and CLIP, allow AI to understand and generate content across different formats. For instance, DALL-E can create images from textual descriptions, while CLIP understands the relationship between images and text to improve image generation or search. The \"multimodal\" aspect refers to the AI's ability to operate across various data types, making these models versatile for tasks that require an understanding of both language and visual elements, such as generating images, videos, or even 3D models based on descriptive prompts. ### Machine Learning Machine Learning (ML) is a subfield of AI that enables systems to learn from data without being explicitly programmed for specific tasks. It focuses on developing algorithms and statistical models that enable computer systems to improve their performance on a specific task through experience without being explicitly programmed. The key feature distinguishing ML from traditional programming is the model\'s ability to learn patterns and make decisions based on those patterns. These algorithms analyse and learn from large amounts of data, identifying patterns and making decisions with minimal human intervention. ML models improve their performance and accuracy over time as they are exposed to more data. There are three primary types of machine learning: - **Supervised Learning**: In this model, the algorithm is trained on labelled data, meaning that each input is paired with the correct output. Supervised learning is widely used for tasks like image classification, fraud detection, and predictive modelling. - **Unsupervised Learning**: This model works with unlabelled data, meaning the system must find hidden patterns or relationships in the data. Clustering algorithms and association rules are common unsupervised learning techniques often used in market basket analysis and customer segmentation. - **Reinforcement Learning**: This model involves training an agent to act in an environment to maximise cumulative reward. It is commonly used in robotics, gaming, and autonomous vehicle development. **Approaches to Machine Learning** There are several approaches to ML, including **Decision Trees**, which split data into branches to make decisions; **Support Vector Machines** (SVMs), which find the optimal boundary between classes; **and K-Nearest Neighbours** (KNN), which classify data points based on their proximity to others. Ensemble methods, like **Random Forests** or **Gradient Boosting**, combine multiple models to enhance predictive accuracy. Each of these techniques has strengths depending on the problem, data type, and computational complexity. **Neural Networks** are a foundational approach in Machine Learning, particularly in tasks involving pattern recognition and complex decision-making. A neural network mimics the structure of the human brain, consisting of interconnected layers of nodes (neurons) that process input data. In **supervised learning**, these networks are trained on labelled datasets to identify patterns, make predictions, or classify information. The network learns to minimise errors and improve performance by adjusting the weights between nodes during training. **Deep Learning**, a subset of neural networks, uses multiple hidden layers to learn hierarchical features from raw data, making it especially effective in fields like image recognition, natural language processing, and autonomous systems. Neural networks excel in processing unstructured data, such as images or text, because they can automatically extract features and adapt to complex tasks. **Unsupervised learning** with neural networks focuses on finding hidden patterns or structures in data without labelled outputs. Neural networks can learn to compress and reconstruct data, identify clusters, or generate new data resembling the input set. These models are particularly useful in tasks like anomaly detection, data compression, or generating realistic images. **Reinforcement learning** uses neural networks to help an agent learn optimal actions through trial and error in an environment. The network acts as a function approximator, mapping states and actions to rewards. **Deep reinforcement learning** combines neural networks with reinforcement learning, allowing the system to solve complex, high-dimensional tasks, such as playing video games or autonomous driving, by learning from interactions with the environment and improving through feedback over time. ### The role of big data in AI development Big data plays a crucial role in the development and advancement of Artificial Intelligence (AI). AI systems, particularly machine learning and deep learning models rely on vast amounts of data to learn patterns, make decisions, and improve their accuracy. The more data these systems are trained on, the better they can recognise subtle patterns, correlations, and anomalies, leading to more accurate predictions and insights. Big data refers to the massive and complex datasets that are generated by a variety of sources, such as social media platforms, IoT devices, e-commerce transactions, and digital interactions. These datasets are typically too large or complex to be processed by traditional data management systems. In AI development, big data provides the raw material that enables models to train and improve. For instance, deep learning models, which use neural networks, depend on large datasets to tune their layers of interconnected nodes. Without access to big data, the accuracy of these models would be limited, and their capacity to generalise across different tasks would be constrained. Moreover, AI-driven applications like natural language processing, computer vision, and autonomous systems require enormous datasets to ensure their systems can perform reliably in real-world conditions. In essence, big data fuels AI, allowing it to continuously evolve and become more intelligent and effective across a wide range of applications. ### Read: What Is Machine Learning (ML) The Berkeley School of Information go into the details of what machine learning is and how it fits into the domain of AI. What Is Machine Learning (ML)? \| Berkeley School of Information \ ### Watch: Title The following three videos are from IBM Technology. The first video reviews Machine Learning and considers the difference between Supervised, Unsupervised, and Reinforcement Learning. The second video reviews what Machine Learning and Deep Learning mean. The third video explains how Large Language Models (LLMs) work. IBM Technology: What is Machine Learning? \ (8:22 min) IBM Technology: Machine Learning vs Deep Learning \ (7:49 min) IBM Technology: How Large Language Models Work \ (5:33 min) ### Activity: Title \[H5P Multiple-choice\] **Question 1:** **Which of the following best describes General AI (Strong AI)?** A. An AI system that can perform any intellectual task a human can, with the ability to reason, learn, and adapt across domains. B. An AI system that specialises in a single task, such as facial recognition or playing chess. C. A theoretical AI system that surpasses human intelligence in all aspects, including creativity and decision-making. D. An AI system that reacts to inputs without learning from past experiences. Correct Answer: A. An AI system that can perform any intellectual task a human can, with the ability to reason, learn, and adapt across domains. **Question 2: How is deep learning related to neural networks?** A. Deep learning is a type of neural network that uses decision trees to process data. B. Neural networks are used in deep learning to create multiple layers for analysing complex patterns in data. C. Neural networks are a form of unsupervised learning, while deep learning focuses only on supervised tasks. D. Deep learning uses reinforcement learning instead of neural networks to solve problems. Correct Answer: B. Neural networks are used in deep learning to create multiple layers for analysing complex patterns in data. **\ ** 2. Technical Complexities and Their Regulatory Implications ----------------------------------------------------------- ### The \"Black Box\" Problem in AI Decision-Making *The \"black box\" problem refers to the difficulty in understanding how AI systems, particularly those based on machine learning algorithms, arrive at their decisions. This opacity poses significant challenges for regulation and accountability.* ***Understanding the Black Box*** *Understanding the \"black box\" problem in AI is crucial as it highlights the challenges of transparency and interpretability in complex AI systems. The term \"black box\" metaphorically describes AI models, particularly those based on deep learning, where the internal decision-making processes are obscured from users and even developers. While these models can produce highly accurate predictions, the pathways leading to these outputs remain largely unknown, raising concerns about trust, accountability, and ethical implications.* *In practical terms, users can observe the inputs fed into an AI system and the outputs it generates, but the intricate calculations and logic that occur in between are hidden. This opacity complicates efforts to diagnose errors or biases in AI decisions, especially in high-stakes applications such as healthcare or finance. For instance, if an AI system used for medical diagnosis makes an incorrect recommendation, understanding why it failed becomes nearly impossible without insight into its decision-making process.* ***Key Points:*** - ***Complex Algorithms**: Complex algorithms are at the heart of many AI systems, enabling them to process vast amounts of data and generate accurate predictions. However, their complexity often leads to significant challenges, particularly in terms of interpretability and transparency. An example of this is neural networks. These are inherently complex structures composed of interconnected layers of artificial neurons that process data through numerous parameters. As neural networks grow deeper--- (deep learning) their ability to capture and model intricate patterns in data improves, but so does their opacity. This non-linear mapping of input features to outputs makes it challenging to decipher how specific decisions are made.* - ***Data Dependency**: The performance of AI systems heavily relies on the quality and quantity of data used for training. Poorly curated datasets can lead to flawed decision-making.* ***Regulatory Implications*** *The black box nature of AI has profound implications for regulatory frameworks. Regulators face challenges in ensuring that AI systems operate fairly and transparently. Without clear insights into how decisions are made, it becomes difficult to hold organisations accountable for adverse outcomes.* ***Challenges Include:*** - ***Accountability**: If an AI system makes a harmful decision, determining liability becomes complicated when the decision-making process is not transparent.* - ***Bias Detection**: Identifying and mitigating biases in AI systems is challenging without understanding how these biases are introduced during the decision-making process.* ***Approaches to Mitigate the Black Box Problem*** *To address the challenges posed by the black box problem, various strategies can be implemented:* - ***Explainable AI (XAI):** Developing models that provide explanations for their decisions can enhance transparency. Techniques such as LIME (Local Interpretable Model-agnostic Explanations) allow users to understand why a model made a particular decision. (See next section for more information.)* - ***Regulatory Frameworks**: Governments and organisations are beginning to draft regulations that require transparency in AI systems. For example, the European Union\'s AI Act* emphasises *the need for explainability in high-risk AI applications.* - ***Audit Mechanisms**: Implementing regular audits of AI systems can help ensure compliance with ethical standards and regulatory requirements. These audits can assess both the algorithms used and the datasets they rely on.* ***Ethical Considerations*** *The ethical implications of the black box problem extend beyond regulatory compliance. The inability to explain AI decisions raises concerns about fairness, justice, and discrimination:* - ***Fairness**: If an AI system\'s decision-making process is opaque, it may inadvertently perpetuate biases in training data.* - ***Trust**: Users\' trust in AI technologies diminishes when they cannot understand how decisions affecting their lives are made.* ### Watch: Title This amusing and informative video gives and explanation of what the black box problem is. While it doesn't mention the phrase, it does explain how AI learns and where humans fail to understand what AI is doing while learning. CGP Grey: How AIs, like ChatGPT, Learn https://www.youtube.com/watch?v=R9OHn5ZF4Uo (8:54 min) ### Read: Title *In this piece, the 'black box' problem and approach to dealing with it are explained.* *Lou Blouin: AI\'s mysterious 'black box' problem, explained* *https://umdearborn.edu/news/ais-mysterious-black-box-problem-explained* ### Explore and Reflect *The black box problem represents a significant hurdle in effectively regulating AI technologies. As these systems increasingly influence various aspects of society---from healthcare to finance to law---addressing their opacity is crucial for fostering accountability, fairness, and trust. By understanding the complexities associated with AI decision-making processes, lawyers, policy makers, and the public will be better equipped to advocate for robust regulatory frameworks that balance innovation with societal concerns.\ * Questions: To what extent do you believe the black box problem is an obstacle to the wide-scale adoption of AI? Do you believe that the black box problem poses a threat to the safety of the use of AI? How do you perceive its impact on sectors that are vital to our society, such as healthcare, finance, and law? Does the black box problem influence the way that you use or interact with AI? ### Explainable AI (XAI) *Explainable Artificial Intelligence (XAI) is increasingly recognised as a critical component in the responsible deployment of AI technologies. As AI systems become more integrated into various sectors, the need for transparency and accountability in their decision-making processes has never been more pressing.* ***Definition and Purpose*** *Explainable AI refers to methods and techniques that make the outputs of AI systems understandable to humans. (See: IBM What is explainable AI? \) The primary goal of XAI is to provide insights into how AI models make decisions, thereby allowing stakeholders to comprehend the rationale behind these outcomes. This understanding is essential for fostering trust among users and ensuring that AI systems align with ethical standards and legal requirements.* ***Key Features of XAI:*** - ***Transparency**: XAI aims to clarify the decision-making processes of AI systems, making it easier for users to understand how inputs are transformed into outputs.* - ***Interpretability**: It focuses on creating models that can be interpreted by humans, allowing them to grasp the underlying logic of AI decisions.* - ***Accountability**: By providing explanations for decisions, XAI helps establish accountability for both developers, organisations, and all those deploying these systems.* ***XAI: Three critical concepts---Prediction Accuracy, Traceability, and Decision Understanding*** *In the context of Explainable AI (XAI), three critical concepts---Prediction Accuracy, Traceability, and Decision Understanding---play vital roles in addressing both technological requirements and human needs.* ***Prediction Accuracy** refers to the ability of an AI system to make correct predictions based on input data. High prediction accuracy is essential for ensuring that AI applications function effectively, particularly in high-stakes environments like healthcare or finance where incorrect predictions can have serious consequences. This technical requirement emphasises the need for robust algorithms that can analyse complex datasets reliably.* ***Traceability**, on the other hand, pertains to the ability to track and verify the decision-making processes within AI systems. It ensures that every decision made by an AI can be traced back to its underlying data and algorithms, which is crucial for regulatory compliance and accountability. This addresses technological requirements by providing a clear audit trail that can be reviewed and assessed.* *In contrast, **Decision Understanding** focuses on human needs by ensuring that users can comprehend how and why specific decisions were made by the AI system. This understanding fosters trust and confidence among users, as it allows them to engage with AI technologies more meaningfully.* *By bridging the gap between complex algorithmic processes and human interpretation, XAI enhances user experience and promotes responsible AI deployment, ultimately aligning technological capabilities with societal expectations.* ***Enhancing Compliance with Regulations*** *The integration of XAI into regulatory frameworks is becoming increasingly necessary as governments worldwide recognise the potential risks associated with opaque AI systems. For instance, the European Union\'s Artificial Intelligence Act (EU AIA) mandates that high-risk AI applications must provide clear and comprehensible explanations of their decision-making processes. This requirement underscores the importance of XAI in ensuring compliance with emerging regulations.* ***Key Regulatory Aspects:*** - ***Adverse Actions**: Regulations may require organisations to explain adverse actions taken based on AI decisions, such as loan denials or job applications.* - ***Data Protection Compliance**: XAI can aid organisations in complying with data protection laws by providing transparency about automated decision-making processes, thus fulfilling obligations under regulations like GDPR.* ***Building Trust Through Explainability*** *One of the most significant benefits of XAI is its potential to build trust among users. When individuals understand how an AI system arrives at its conclusions, they are more likely to accept its recommendations and decisions. This trust is crucial in sectors such as healthcare, finance, and law, where AI systems can significantly impact people\'s lives.* ***Benefits of Trust:*** - ***Informed Decision-Making**: Users can make better-informed decisions when they understand the reasoning behind an AI\'s output.* - ***Reduced Bias**: Transparency in decision-making can help identify and mitigate biases present in training data or algorithms.* ***The Technical Complexity Challenge in Implementing XAI*** *Despite its benefits, implementing XAI poses several challenges. Many advanced AI models, particularly those based on deep learning techniques, are inherently complex and difficult to interpret. Striking a balance between model accuracy and interpretability remains a significant hurdle for developers.* ***Key Challenges Include:*** - ***Trade-offs Between Accuracy and Interpretability**: More complex models often yield better performance but at the cost of being less interpretable.* - ***Resource Requirements**: Developing explainable models requires specialised skills in both machine learning and domain knowledge, which may not always be available within organisations.* ***Future Directions for XAI in Regulation*** *As awareness of the importance of explainability grows, regulatory bodies are likely to continue developing frameworks that emphasise XAI. This evolution will require organisations to adopt best practices for implementing explainable models while ensuring compliance with legal standards.* ***Potential Developments:*** - ***Standardisation of Explainability Metrics**: Establishing standardised metrics for evaluating explainability could help organisations assess their compliance with regulatory requirements.* - ***Increased Scrutiny from Regulators**: As regulations become more stringent, organisations may face increased scrutiny regarding their use of opaque models without adequate explanations.* *Explainable Artificial Intelligence plays a vital role in addressing the complexities associated with regulating AI technologies. By enhancing transparency and accountability, XAI helps organisations comply with emerging regulations while fostering trust among users. However, challenges remain in implementing effective explainability measures within complex models. As regulatory frameworks continue to evolve, organisations must prioritise XAI to navigate the intricacies of responsible AI deployment successfully.* ### Watch *For more information on what XAI is, watch the following video.* *IBM Technology: What is Explainable AI?* *https://www.youtube.com/watch?v=jFHPEQi55Ko* *(7:29 mins)* ### Read *Inam, F et al. (2021) 'Explainable AI -- How Humans Can Trust AI', Ericsson White Paper \* 3. Trust as a Critical Issue in AI ---------------------------------- *Trust is a critical issue in the realm of AI, as it underpins the acceptance and effective implementation of these technologies across various sectors. As AI systems increasingly influence decision-making processes, concerns about fairness and bias have come to the forefront. Stakeholders must grapple with the potential for algorithmic bias to perpetuate existing inequalities, making it essential to ensure that AI models are developed and deployed with fairness in mind. The robustness of AI systems is crucial for maintaining reliability in dynamic environments, while AI security is paramount to protect sensitive data from breaches and malicious attacks.* ### Fairness: Protecting Against Bias and Drift *Fairness is a fundamental pillar of trustworthy AI, emphasising the need for AI systems to operate without bias against any individual or group. This concept is crucial to ensuring that algorithms do not systematically disadvantage specific populations based on sensitive attributes like age, gender, or ethnicity. To achieve fairness, the creators of AI models must assess and mitigate biases present in the training data and model outputs. This involves employing diverse datasets that accurately represent the populations affected by AI decisions, thus enhancing the model\'s ability to make equitable assessments.* *Fairness is closely linked to explainability; if users cannot understand how an AI system makes decisions, they cannot ascertain whether those decisions are fair. By fostering an environment where fairness is prioritised, organisations can build trust with users and stakeholders, ultimately leading to broader acceptance of AI technologies.* *To maintain fairness in an AI system, there are two related concepts that need to be considered: Bias and Model Drift* ***Bias** in AI systems is a critical concern that arises from various stages of the AI development process, particularly in data collection and model training. When the data used to train AI algorithms is not representative of the diverse populations it aims to serve, it can lead to biased outcomes that reinforce existing societal inequalities. For example, if an AI model is trained predominantly on data from one demographic group, it may perform poorly when applied to individuals outside that group, perpetuating stereotypes and discrimination. This selection bias can manifest in numerous applications, including hiring algorithms that favour certain gender or ethnic groups based on historical hiring patterns. Thus, addressing bias requires a comprehensive approach to ensure that datasets are diverse and inclusive, as well as continuous monitoring of AI outputs to identify and rectify any emerging biases.* ***Model drift** is another significant factor in maintaining fairness in AI systems. As the environment in which an AI system operates changes over time---whether due to shifts in societal norms, user behaviour, or external conditions---the model\'s performance can degrade if it is not updated accordingly. This phenomenon can lead to outdated predictions that no longer reflect current realities, further exacerbating issues of bias and unfairness. For instance, an AI model trained on historical data may not accurately predict outcomes for new populations or under changing circumstances, resulting in unfair treatment of individuals who do not conform to the original training data\'s characteristics. To combat model drift, there must be regular evaluations and updates of AI models, to ensure that they remain relevant and equitable over time.* ***Types of bias: algorithmic, data, and human bias*** *Bias in artificial intelligence (AI) can be categorised into three main types: algorithmic bias, data bias, and human bias. Understanding these biases is crucial for developing fair and effective AI systems.* ***Algorithmic bias** occurs when the algorithms themselves produce systematically skewed results. This can arise from the design of the algorithm or the way it processes data. For instance, if an algorithm is optimised for certain outcomes without considering broader contexts, it may inadvertently favour specific groups over others. A notable example is facial recognition technology, which has been shown to misidentify individuals from minority groups due to its training on predominantly datasets of light skinned individuals.* ***Data bias** refers to inaccuracies in the data used to train AI models. This can stem from unrepresentative samples, leading to outcomes that do not reflect the real-world population. For example, if a dataset used for training an AI hiring tool predominantly features male candidates, the model may develop a preference for male applicants, perpetuating gender inequality in hiring practices. Data bias can manifest in various forms, such as selection bias and measurement bias, and often requires careful auditing and diverse data sourcing to mitigate.* ***Human bias** is introduced during the data collection and interpretation processes. There are close to 200 documented different types of cognitive biases, in most instances, people are rarely aware of their cognitive biases. (Cirillo, Rementeria, Chapter 3 - Bias and fairness in machine learning and artificial intelligence (2022) Sex and Gender Bias in Technology and Artificial Intelligence, Academic Press, 57,) Researchers\' preconceived notions or societal stereotypes can influence how data is gathered and analysed. This type of bias can lead to flawed conclusions that reinforce existing prejudices. For instance, if survey questions are framed in a way that reflects a particular viewpoint, they may skew results toward that perspective. Addressing human bias involves training personnel to recognise their biases and implementing structured methodologies for data collection.* *By understanding and addressing these three types of bias, developers can work towards creating more equitable AI systems.* ***Consequences of biased AI systems in various domains*** *Biased AI systems can have severe consequences across various domains, impacting decision-making processes and perpetuating existing inequalities.* *In **healthcare**, biased AI can lead to misdiagnoses or inappropriate treatment recommendations. For instance, if an AI system is trained on data that underrepresents certain demographics, it may produce recommendations that are less effective for those groups. Research indicates that healthcare professionals influenced by biased AI recommendations may replicate these biases in their own decisions, leading to long-term adverse effects on patient care and outcomes. (See Vicente, L., Matute, H. Humans inherit artificial intelligence biases. Sci Rep 13, 15737 (2023). https://doi.org/10.1038/s41598-023-42384-8)* *In **finance**, algorithmic bias can result in discriminatory lending practices. AI tools used for credit scoring may inadvertently assign higher risk scores to applicants from minority backgrounds based on historical data that reflects societal biases. This can restrict access to loans and perpetuate economic disparities, significantly costing marginalised communities in terms of financial opportunities. (See Australian Human Right Commission, Using artificial intelligence to make decisions: Addressing the problem of algorithmic bias-Technical Paper (2020) p41. https://humanrights.gov.au/sites/default/files/document/publication/final\_version\_technical\_paper\_addressing\_the\_problem\_of\_algorithmic\_bias.pdf)* *In the **employment** sector, AI hiring algorithms can favour candidates based on biased training data, disadvantaging women and minority applicants. For example, a job search algorithm might prioritise male candidates for high-paying positions due to historical hiring patterns, thus reinforcing gender disparities in the workplace (See* *Chen, Z. Ethics and discrimination in artificial intelligence-enabled recruitment practices. Humanit Soc Sci Commun 10, 567 (2023). https://doi.org/10.1057/s41599-023-02079-x).* *Moreover, the **legal domain** faces challenges with biased AI tools used in risk assessments or predictive policing, which can lead to wrongful accusations or sentencing based on flawed data interpretations. The implications of these biases extend beyond individual cases, affecting public trust in institutions and exacerbating systemic inequalities across society. (See* *Gültekin-Várkonyi, Predictive Policing and Bias in a Nutshell: Technical and Practical Aspects of Personal Data Processing for Law Enforcement Purposes Digital Criminal Justice: A Studybook Selected Topics for Learners and Researchers. (2022). https://ssrn.com/abstract=4238774)* *Addressing these biases is essential to ensure fairness and equity in AI applications across all sectors.* ### Robustness: AI Security Risks and Vulnerabilities *As artificial intelligence (AI) systems become increasingly integrated into various sectors, they introduce a range of security risks and vulnerabilities. These risks not only threaten the integrity of AI applications but also pose significant regulatory challenges.* ***Understanding AI Security Risks*** *AI security risks can be broadly categorised into several types, each with unique implications for organisations deploying AI technologies.* ***1. Data Poisoning*** ***Definition and Impact**: Data poisoning occurs when malicious actors manipulate the training datasets used to develop AI models. By injecting erroneous or biased data, attackers can skew the model's outputs, leading to incorrect predictions or harmful decisions. This type of attack is particularly concerning in critical sectors such as healthcare and finance, where compromised data can have severe consequences.* ***Regulatory Implications**: Organisations must ensure that their data handling practices comply with data protection regulations like GDPR, which mandates data integrity and accuracy. Failure to maintain the quality of training data can lead to significant legal repercussions, including fines and reputational damage.* ***2. Model Theft*** ***Definition and Risks**: Model theft involves unauthorised access to an organisation's proprietary AI models, which can then be exploited for malicious purposes. Attackers may use techniques such as reverse engineering or exploiting vulnerabilities in the system to gain access to these models. This risk is particularly acute for organisations that rely heavily on their intellectual property (IP) for competitive advantage.* ***Regulatory Considerations:** The theft of AI models raises serious concerns about compliance with intellectual property laws and trade secrets regulations. Organisations must implement robust security measures to protect their models from unauthorised access, as failure to do so could result in legal action from competitors or regulatory bodies.* ***3. Adversarial Attacks*** ***Definition and Mechanisms:** Adversarial attacks involve manipulating input data to deceive AI systems into making incorrect predictions or classifications. These attacks exploit the inherent vulnerabilities in machine learning algorithms, often leading to significant operational disruptions. For example, slight alterations in images can cause an image recognition system to misidentify objects entirely.* ***Regulatory Impact:** The potential for adversarial attacks necessitates stringent regulatory oversight concerning the robustness of AI systems. Organisations must demonstrate that they have taken adequate measures to protect against these vulnerabilities, which may include routine testing and validation of AI models.* *For more information on adversarial attacks watch the following video.* *HarrietHacks: What Is Adversarial Machine Learning?* *https://www.youtube.com/watch?v=DA8m54CDBoE* *(7:08 mins)* ***Vulnerabilities Throughout the AI Lifecycle*** *AI systems are susceptible to vulnerabilities at various stages of their lifecycle---from design and development to deployment and maintenance.* ***1. Design Phase Vulnerabilities*** *During the design phase, inadequate security protocols can lead to fundamental flaws in AI systems. Poorly designed algorithms may lack necessary safeguards against unauthorised access or manipulation.* *Mitigation Strategies include the implementation of secure coding practices and the undertaking of threat modelling to identify potential vulnerabilities.* ***2. Development Phase Vulnerabilities*** *In the development phase, reliance on third-party libraries or open-source components can introduce vulnerabilities if these resources are not adequately vetted for security risks.* *Mitigation strategies include regularly updating all software dependencies and conducting security audits of third-party components before integration.* ***3. Deployment Phase Vulnerabilities*** *Once deployed, AI systems may face threats from external actors attempting to exploit known vulnerabilities or misconfigurations in the system.* *Mitigation strategies include employing robust monitoring tools to detect unusual activity and regular updates to security protocols based on emerging threats.* ***Regulatory Frameworks Addressing AI Security Risks*** *Given the potential risks associated with AI technologies, regulatory bodies are beginning to establish frameworks aimed at mitigating these vulnerabilities:* ***1. Risk-Based Approaches*** *Regulations like the EU AI Act advocate for a risk-based approach that categorises AI applications based on their potential impact on society. High-risk applications must adhere to stricter compliance requirements, including robust security measures against identified vulnerabilities.* ***2. Mandatory Reporting Requirements*** *Some jurisdictions are considering mandatory reporting requirements for organisations that experience security breaches involving AI systems. This transparency will help regulators assess systemic risks and develop appropriate responses.* *AI security risks and vulnerabilities present significant challenges for organisations deploying these technologies. By understanding the various types of risks---such as data poisoning, model theft, and adversarial attacks---organisations can better prepare themselves for compliance with emerging regulatory frameworks. Implementing robust security measures throughout the AI lifecycle is essential for mitigating these risks and ensuring responsible AI deployment.* *For further exploration of these topics, students are encouraged to read articles such as \"How AI Creates Cybersecurity Vulnerabilities\" by WTW and \"Cyber Security Risks to Artificial Intelligence\" by GOV.UK. These resources provide valuable insights into practical strategies for addressing AI security challenges while navigating regulatory landscapes effectively.* ### Watch *AI models can be "hacked" with poisoned data. Watch the below video to see how this can be done and its associated consequences.* *IBM Technology: How Chatbots Could Be \'Hacked\' by Corpus Poisoning* *https://www.youtube.com/watch?v=RTCaGwxD2uU* *(8:30 mins)* ### Read *For insights into model theft and its implications, students can read \"* *Trend Micro: The Top 10 AI Security Risks Every Business Should Know* *https://www.trendmicro.com/en\_au/research/24/g/top-ai-security-risks.html* 4. AI Applications and Their Regulatory Challenges -------------------------------------------------- *The rapid advancement of artificial intelligence (AI) technologies has introduced a myriad of applications that significantly impact various sectors, from healthcare to finance and beyond. However, this swift evolution presents substantial regulatory challenges that lawmakers struggle to address effectively. As we have seen earlier, as AI systems become more complex and integrated into critical decision-making processes, the opacity of these technologies raises concerns about accountability, fairness, and ethical implications. The \"black box\" nature of many AI algorithms makes it difficult to understand how decisions are made, complicating compliance with existing regulations and creating potential risks for users and society at large.* *As organisations, business, individuals, and governments deploy AI systems, problems can arise that require the intervention of regulation. Understanding how these problems manifest in different sectors of society is critical to understanding why regulations are required.* ### Watch IE Insights: The Challenges of Governing AI \ (4:42 mins) ### Read Wheeler, T, 'The Three Challenges of AI Regulations' (15 June 2023), *Brookings* \< https://www.brookings.edu/articles/the-three-challenges-of-ai-regulation/\> ### Reflect 1. *What do you think is the major regulatory challenge regarding AI applications?* 2. *How can lawmakers strike a balance between fostering AI innovation and protecting individual rights and societal interests?* ### AI in healthcare: diagnostic tools and personalised medicine *The integration of artificial intelligence (AI) in healthcare is transforming diagnostic tools and personalised medicine, leading to more effective and tailored patient care. Traditional medical practices often follow a one-size-fits-all approach, but AI enables a shift towards precision medicine by analysing vast amounts of patient data, including genetic information, medical histories, and lifestyle factors. This data-driven approach allows healthcare providers to identify patterns and correlations that inform more accurate diagnoses and treatment plans. In addition to improving diagnostics, AI plays a crucial role in real-time monitoring and preventive care. Wearable devices equipped with AI can continuously track vital signs and health metrics, alerting healthcare professionals to any abnormalities that may require intervention. This proactive approach not only enhances patient outcomes but also empowers individuals to take charge of their health through personalised recommendations.* *AI applications in healthcare, while promising transformative advancements, face several significant challenges, including bias, model drift, and robustness.* - ***Bias** can emerge from various sources, such as skewed training data that does not adequately represent diverse patient populations. For instance, if an AI system is trained primarily on data from one demographic group, it may produce less accurate or even harmful outcomes for individuals outside that group.* - ***Model drift** can occur when a model trained on historical data becomes less effective as new treatments emerge or as patient populations evolve. For example, an AI diagnostic tool may become less accurate if it is not updated to reflect new medical knowledge or shifts in disease prevalence. To combat model drift, continuous monitoring and retraining of AI systems are essential to maintain their relevance and effectiveness.* - ***Robustness** is critical in healthcare settings where decisions based on AI outputs can have life-altering consequences. For example, if an AI system falls victim to an adversarial attack, this can have devastating consequences on the patients relying on the system's outputs. Developing resilient AI systems that can withstand uncertainties and provide consistent performance is vital for building trust among healthcare professionals and patients alike.* ### AI in finance: algorithmic trading and credit scoring *The finance sector is being transformed through the use of AI applications, particularly in algorithmic trading and credit scoring. In algorithmic trading, AI systems analyse vast amounts of market data in real time to identify patterns and execute trades at speeds unattainable by human traders. These algorithms can adapt to market fluctuations, optimising investment strategies based on predictive analytics. This capability not only enhances trading efficiency but also allows for more informed decision-making, potentially leading to higher returns on investment. In credit scoring, AI offers an approach to evaluating a borrower\'s creditworthiness that is more comprehensive than the traditional methods that rely heavily on historical data and predefined criteria. AI-based systems incorporate diverse data sources, including online behaviour and transaction history and provide a holistic assessment that enables lenders to make more informed lending decisions and identify potential risks more effectively.* *AI applications in finance, particularly in algorithmic trading and credit scoring, face several significant risks that can impact market stability and fairness.* ***Algorithmic Trading Risks*** - ***Market Manipulation**: AI-driven trading systems can inadvertently or deliberately create false market trends. Sophisticated algorithms might manipulate prices, leading to artificial volatility, particularly in less liquid markets. This manipulation can undermine market integrity and investor confidence.* - ***Over-Reliance on Historical Data**: AI models depend heavily on historical data for predictions. If market conditions change unexpectedly, these models may fail to adapt, resulting in poor trading decisions. This over-reliance can create a false sense of security among traders.* - ***Errant Algorithms**: The speed at which algorithmic trading occurs means that a single faulty algorithm can lead to substantial financial losses within minutes. Notable incidents, such as the Knight Capital fiasco, (Watch: A Hedge Fund Accidentally Used The Wrong Code, Lost \$400 Million \) illustrate how errant algorithms can cause widespread disruption.* - ***Cybersecurity Threats**: AI trading systems are vulnerable to hacking and other cyber threats. A breach could lead to unauthorised access to sensitive market information or manipulation of trading algorithms, potentially destabilising financial markets.* ***Credit Scoring Risks*** - ***Bias in Data**: AI systems used for credit scoring can perpetuate existing biases if trained on historical data that reflects societal inequalities. This can lead to unfair lending practices that disproportionately affect marginalised groups.* - ***Lack of Transparency**: The complexity of AI algorithms often results in a \"black box\" effect, where the decision-making process is opaque. This lack of transparency can hinder accountability and make it difficult for consumers to understand why they were denied credit.* - ***Model Drift**: As economic conditions change, models that were once accurate may become outdated, leading to incorrect assessments of creditworthiness. Continuous monitoring and updating of these models are essential to maintain their effectiveness.* - ***Regulatory Challenges**: The rapid development of AI technologies often outpaces regulatory frameworks, creating gaps in oversight that can lead to unethical practices and consumer harm.* *Addressing these risks requires a careful balance between innovation and ethical considerations, ensuring that AI technologies enhance financial services without compromising fairness or security.* ### AI in the public sector: predictive policing and social services *AI is increasingly being utilised in the public sector, particularly in predictive policing and social services, to enhance efficiency and improve outcomes. In predictive policing, AI algorithms analyse vast amounts of historical crime data to forecast potential criminal activity, enabling law enforcement agencies to allocate resources more effectively. By identifying high-risk areas and times for crime, these systems aim to prevent incidents before they occur. In social services, AI applications help streamline processes and enhance service delivery. For instance, predictive analytics can identify individuals at risk of homelessness or substance abuse, allowing social workers to intervene proactively.* *The use of AI in the public sector, particularly in predictive policing and social services, presents several significant risks that can impact both individuals and communities.* ***Predictive Policing Risks*** - ***Bias and Discrimination**: AI algorithms often rely on historical crime data, which can reflect systemic biases present in law enforcement practices. This can lead to discriminatory outcomes where certain communities, particularly marginalised groups, are disproportionately targeted for surveillance and policing. For instance, predictive policing tools may flag areas with higher historical arrest rates as high-risk areas, perpetuating a cycle of over-policing.* - ***Lack of Transparency**: Many predictive policing systems operate as \"black boxes,\" making it difficult for the public and even law enforcement officers to understand how decisions are made. This opacity can undermine accountability and trust in law enforcement agencies, as civilians may feel subject to arbitrary and unexplainable policing practices.* - ***Erosion of Public Trust**: The reliance on AI for policing can exacerbate existing tensions between law enforcement and communities, particularly if these technologies lead to increased surveillance and perceived injustices. This erosion of trust can hinder community cooperation with police efforts and diminish overall public safety.* ***Social Services Risks*** - ***Data Privacy Concerns**: The use of AI in social services often involves collecting sensitive personal information to assess needs or risks. This raises significant privacy concerns, especially if data is mismanaged or accessed by unauthorised individuals.* - ***Algorithmic Bias**: Like predictive policing, AI systems in social services can inherit biases from the data they are trained on. If these systems are not carefully monitored and audited, they may perpetuate existing inequalities in service provision, affecting vulnerable populations adversely.* - ***Dependence on Technology**: Over-reliance on AI tools can lead to diminished human oversight in social services, which is critical for understanding complex individual circumstances. Automated decisions may lack the nuance required for effective intervention, potentially leading to inadequate support for those in need.* ### AI in transportation: autonomous vehicles and traffic management *Artificial intelligence (AI) is significantly transforming the transportation sector, particularly through applications in autonomous vehicles and traffic management. Autonomous vehicles (AVs) utilise AI to navigate and operate without human intervention, relying on a combination of sensors, cameras, and machine learning algorithms. These technologies enable AVs to interpret their surroundings, make real-time decisions, and improve safety by reducing human error---a leading cause of traffic accidents. In addition, AI plays a crucial role in traffic management systems. By analysing data from various sources---such as traffic cameras, sensors, and GPS---AI can optimise traffic flow, reduce congestion, and enhance overall urban mobility. Smart traffic signals can adjust in real time based on current traffic conditions, improving travel efficiency and minimising delays.* *(For more information see: Myers, How AI Is Making Autonomous Vehicles Safer. (2022) \)* *AI in transportation, particularly in autonomous vehicles and traffic management systems, presents several significant risks that must be carefully managed to ensure safety and efficiency.* - ***Ethical Decision-Making**: One of the primary challenges in autonomous vehicles is programming ethical decision-making algorithms. In unavoidable crash scenarios, AVs must be designed to make split-second decisions that prioritise the safety of passengers, pedestrians, or other road users. This raises complex moral questions about how these decisions are made and who is responsible for the outcomes.* - ***Bias and Discrimination**: Bias in AI systems can lead to discriminatory practices, particularly in navigation and pedestrian recognition. If the training data used to develop these algorithms is skewed or unrepresentative, it may result in AVs that perform poorly in certain environments or with specific demographics, which can exacerbate existing inequalities.* - ***Data Privacy and Surveillance**: The deployment of AI technologies in traffic management often involves extensive data collection through cameras and sensors. This raises concerns about constant surveillance and potential misuse of personal data, leading to public distrust in these systems.* - ***Cybersecurity Threats**: Both AVs and traffic management systems are vulnerable to cyberattacks. Breaches could lead to unauthorised access to sensitive information or manipulation of traffic signals, causing chaos on the roads.* - ***Model Drift**: As traffic patterns and urban environments evolve, AI models may experience drift, where their performance deteriorates over time if not regularly updated. This can lead to inaccurate predictions and unsafe driving conditions.* *Addressing these risks requires robust regulatory frameworks, continuous monitoring for biases, and ethical considerations in the design and implementation of AI systems in transportation.* *(For a comprehensive overall of the risk associated with AI and transportation, see:* *Yu, Hulse, and Irshad, Understanding AI Risks in Transportation (2024) \).* 5. Emerging Challenges Requiring AI Regulation ---------------------------------------------- *The rapid development of AI technologies presents emerging challenges that necessitate effective regulation. As AI systems become increasingly sophisticated, their application raises significant ethical and safety concerns. Regulators must collaborate closely with tech companies to create frameworks that not only foster innovation but also safeguard public interests. This partnership is essential for developing guidelines that address potential biases, ensure transparency, and mitigate risks associated with AI deployment.* *Moreover, as high-risk AI tools are integrated into daily life, it is crucial to keep the public informed about their implications. Many individuals may not fully understand the risks associated with advanced AI tools. Public awareness campaigns can educate people on how these technologies operate and the potential consequences of their use, fostering a more informed dialogue about AI\'s role in society. By prioritising transparency and accountability, regulators can build trust among the public while ensuring that AI technologies are developed and implemented responsibly. Ultimately, a proactive approach that emphasises collaboration between stakeholders will be vital in addressing the challenges posed by rapidly evolving AI systems and ensuring they serve the greater good.* **AI and intellectual property rights** *If 'content is king', data harvesting and scaping is the artificial king maker, and that's where artificial intelligence (AI) and intellectual property (IP) rights presents complex challenges that regulators must navigate. As AI technologies, particularly generative AI, advance rapidly, they raise significant risks regarding authorship, ownership, and potential copyright infringement. One primary concern is that AI-generated outputs often closely resemble the training data they were fed, which frequently includes copyrighted materials. This blurring of lines between original creation and reproduction can lead to legal disputes over IP rights, as the original creators may claim that their works have been unlawfully used without permission.* *Regulators face technical hurdles in establishing clear guidelines for IP protection in the context of AI. Existing IP laws were primarily designed for human-created works and may not adequately account for the nuances of AI-generated content. For instance, determining whether an AI output is sufficiently original to qualify for copyright protection poses a significant challenge. Additionally, the lack of harmonisation between jurisdictions complicates enforcement and compliance efforts, as different countries may interpret IP rights differently concerning AI. To effectively regulate this evolving landscape, regulators must develop nuanced frameworks that balance innovation with the need to protect intellectual property rights, ensuring that both creators and users of AI technologies are adequately safeguarded against infringement risks.* *For more information see:* *Pope. NYT v. OpenAI: The Times's About-Face (2024) \* **AI-generated content and deepfakes** *AI-generated content and deepfakes present significant risks that challenge existing regulatory frameworks. One major concern is the potential for misinformation. AI systems can create highly convincing but false narratives, images, or videos, which can be disseminated rapidly across digital platforms. This capability raises alarms about the erosion of trust in media and the difficulty of discerning truth from fabrication, particularly in politically sensitive contexts.* *The technology has also introduced significant risks associated with fraud and illegal activities, posing challenges for law enforcement and regulators. Deepfakes can be used to create convincing but false representations of individuals, which can facilitate a range of criminal behaviours, including identity theft, financial fraud, and extortion. For instance, perpetrators can manipulate videos or audio recordings to impersonate someone in a position of authority, leading to fraudulent transactions or the dissemination of false information that can disrupt markets or influence public opinion. Europol\'s report highlights that deepfake technology is increasingly employed in \"crime as a service\" models, where individuals can purchase deepfake capabilities for illicit purposes, further complicating law enforcement efforts.* *Regulators face technical hurdles in combating these threats, as the rapid advancement of deepfake technology outpaces existing legal frameworks. Detecting deepfakes requires sophisticated tools and techniques, which are still in development. Additionally, establishing clear legal definitions and frameworks for accountability presents challenges, especially in jurisdictions with varying laws regarding digital content, privacy and free speech. As deepfakes become more accessible and sophisticated, there is an urgent need for comprehensive regulations that address both the creation and distribution of such content while ensuring that law enforcement agencies are equipped to respond effectively to these emerging threats.* *For more information see: Riehle, Europol Report Criminal Use of Deepfake Technology (2022) \* *For an interacting case study see: Chen and Magramo, Finance worker pays out \$25 million after video call with deepfake 'chief financial officer' (2024) CNN World \)* **AI in warfare and autonomous weapons systems** *The integration of artificial intelligence (AI) into warfare, particularly through autonomous weapons systems, raises profound ethical and operational risks. Autonomous weapons, capable of making decisions without human intervention, can enhance military efficiency but also introduce significant dangers. One major concern is the potential for automation bias, where operators may overly rely on AI recommendations without sufficient scrutiny, leading to unintended consequences such as civilian casualties. Additionally, the unpredictability of AI systems can result in unintended engagements, where targets are misidentified due to faulty data or algorithmic errors.* *Regulators face substantial technical hurdles in overseeing these technologies. Establishing clear definitions of accountability is complex; if an autonomous weapon malfunctions or causes harm, determining liability---whether it lies with the manufacturer, the military, or the AI itself---poses significant challenges. Furthermore, the rapid pace of AI development often outstrips existing legal frameworks, necessitating new regulations that can adapt to evolving technologies. Ensuring transparency in how these systems operate is also crucial; many AI models function as \"black boxes,\" making it difficult to understand their decision-making processes. As nations invest heavily in AI for military applications, a coordinated international effort will be essential to mitigate risks and establish ethical guidelines for the use of autonomous weapons in warfare.* For more information see: Csernatoni, Governing Military AI Amid a Geopolitical Minefield. (2024) \. ***Case Study: US Drone Operations and the Transition to Autonomous Systems*** *The Obama administration\'s use of remotely piloted aircraft (drones) in military operations provides valuable insights into the challenges and controversies surrounding AI in warfare. While not employing fully autonomous lethal weapons, this period marked a significant step towards increased automation in military technology. During 2009-2017, US drone strikes resulted in an estimated 380 to 801 civilian casualties in Pakistan, Yemen, and Somalia, according to the Bureau of Investigative Journalism. The US government\'s own estimate was lower, ranging from 64 to 116. These operations, while involving autonomous systems, still required human operators for targeting and firing decisions. This case highlights the difficulty in accurately assessing civilian casualties and the ethical concerns surrounding remote warfare. It also demonstrates the gradual progression towards more autonomous systems, as evidenced by the military\'s exploration of AI and machine learning in drone swarm attacks and targeting capabilities. As nations continue to develop autonomous weapons systems, this case underscores the critical need for clear international regulations and ethical frameworks to govern their use and prevent unintended consequences.* #### Regulating artificial general intelligence (AGI) *Regulating Artificial General Intelligence (AGI) is a pressing concern as its potential to surpass human intelligence poses significant risks. Unlike narrow AI, which is designed for specific tasks, AGI possesses the ability to understand, learn, and apply knowledge across a wide range of domains. This capability raises existential risks, including the possibility of AGI acting autonomously in ways that could be harmful to humanity.* *Moreover, the rapid advancement of AI technologies complicates regulatory efforts. Current frameworks, such as the EU AI Act, primarily focus on narrow AI applications and may not adequately address the unique challenges posed by AGI. This necessitates a collaborative approach between regulators, technologists, and ethicists to establish guidelines that ensure safety and ethical considerations are prioritised in AGI development. Public awareness is also crucial; informing society about the risks associated with high-risk AI tools can foster informed discussions and promote accountability. As we move toward a future where AGI may become a reality, proactive regulation will be essential to mitigate risks while harnessing the potential benefits of this transformative technology.* **Read** *The AGI Race: Meta Joins Google and Microsoft/OpenAI Alliance \* ### Activity ***Reflection Question:*** *Considering the unique capabilities of Artificial General Intelligence (AGI) and the potential risks it poses to society, what do you believe are the most significant challenges in regulating this technology?* How might the unpredictability of AGI behaviour complicate efforts to establish effective regulatory frameworks? Final activity: Reflect on your learning ---------------------------------------- As you conclude Module 1, reflect on the various content sections and how they interconnect to enhance your understanding of AI regulation. This activity encourages you to synthesise the knowledge gained throughout the module, linking key concepts with the Module Learning Outcomes. **Technological Complexities**: Consider how the foundational concepts of AI technologies, such as machine learning and neural networks, pose unique challenges for effective regulation. How do these complexities influence your understanding of the legal implications surrounding AI? **Security Risks**: Reflect o

LAWS2075 M1 Learning Plan 2024 PDF

Document Details

Tags

Related

Summary

Full Transcript

Upgrade to continue