Artificial Intelligence Overview

Overview of Artificial Intelligence

• AI is a transformative technology significantly changing daily life and work practices.
• Three types of AI:
  • Artificial Narrow Intelligence (ANI): Weak AI focused on specific tasks.
  • Artificial General Intelligence (AGI): Strong AI that can self-teach and operate like a human brain.
  • Artificial Super Intelligence (ASI): Intelligence that surpasses human abilities, capable of emotions and relationships.

Historical Evolution of AI

• The concept of self-operating entities dates back to ancient times; the robo-bird was created around 400-350 BCE by Archytas of Tarentum.
• Alan Turing proposed the Turing Machine in 1936 as a universal computation model.
• Turing introduced the Turing Test in 1950 to evaluate machine intelligence.
• The term "artificial intelligence" was coined at The Dartmouth Conference in 1956.
• AI faced a downturn, referred to as AI Winter between 1970 and 1980, due to unmet expectations and reduced funding.
• Machine Learning gained traction in the 1980s, followed by significant advancements like the backpropagation algorithm in 1986 and Deep Learning around 2010.

Core Elements of AI

• Data: Essential for insights and strategies; must be relevant for effective AI outcomes; sources include images, audio, video, and text.
• Model: Mathematical framework enabling learning from data; advanced models like Deep Learning rely on Artificial Neural Networks (ANNs).
• Compute: Computational resources like GPUs and TPUs are required to train models and process data, with the necessary resources varying by system complexity.

Machine Learning (ML) Overview

• ML emerged in the 1980s, integrating algorithms that imitate human learning to improve accuracy.
• Three types of ML:
  • Supervised Learning: Uses labeled datasets to predict outcomes; includes classification and regression algorithms.
  • Unsupervised Learning: Analyzes unlabeled datasets to discover patterns; includes clustering and dimensionality reduction techniques.
  • Reinforcement Learning: Employs trial-and-error; agents learn from their actions to maximize rewards in an interactive environment.

Supervised Machine Learning

• Utilized for enterprise AI solutions, training models on labeled datasets.
• Advantages include high predictive accuracy and applicability in various domains (e.g., healthcare, finance).
• Disadvantages include potential bias from labeled data and the need for constant retraining due to changes in input-output relationships.

Unsupervised Machine Learning

• Analyzes unlabeled data to find patterns; major tasks include clustering and dimensionality reduction.
• Advantages encompass less manual preparation and discovery of unknown patterns.
• Challenges include the absence of labeled data for quality evaluation and unpredictable results.

Reinforcement Learning

• Key elements include agent, environment, state, actions, and rewards.
• Learning through trial-and-error allows agents to adapt and improve.
• Examples of use include robot navigation and real-time decisions in various applications.

Artificial Neural Networks (ANNs) and Deep Learning (DL)

• ANNs consist of input, processing, and output layers; Deep Learning involves multiple layers to process data more effectively.
• DL excels in applications like speech and image recognition.

Comparison of Machine Learning and Deep Learning

• ML: Works with structured labeled data; trained through manually selected features.
• DL: Capable of self-learning from complex and unstructured data; requires large datasets and powerful systems.

Current and Emerging Use Cases

• ML applications: fraud detection, customer experience (e.g., chatbots), dynamic pricing.
• DL applications: autonomous vehicles, image recognition, claims adjudication, and real-time language translations.

Generative AI

• Broader category that includes AI technologies capable of generating text, images, audio, and video.
• Large Language Models (LLMs) enable human-like content generation and understanding context.
• ChatGPT: A specific LLM application from OpenAI that creates conversational dialogue and generates varied written content.
• DALL-E 3: Text-to-image generation tool that produces images from user prompts.### Dall-E 3 and Generative AI in the Workplace
• Dall-E 3 leverages ChatGPT for prompt refinement and generation.
• Generative AI transforms workplace functions such as search, policy creation, and regulatory compliance.

Search and Recommendations

• AI-powered search engines utilize Natural Language Processing (NLP) and Machine Learning (ML) to interpret user needs.
• Enhanced search capabilities provide quick insights, promote efficient task completion, and reveal new perspectives.
• Industry-specific AI research tools, like Westlaw Precision, allow natural language inquiries with reliable content responses.

Policy Creation AI Assistance

• AI assistants can aggregate and summarize relevant data to address policy themes, gaps, and conflicts.
• Interactive platforms and chatbots foster public participation in policymaking, enhancing citizen feedback mechanisms.

Automated Regulatory Compliance

• AI reduces the complexity and costs of compliance processes through automated tasks such as document analysis and content moderation.
• Generative AI assists in understanding regulations, assessing their impacts, and facilitating necessary adaptations.

AI-Related Risks

• Notable risks in AI deployment include:
  • Fabricated and inaccurate responses.
  • Data privacy concerns and confidentiality issues.
  • Bias in models and outputs.
  • Risks to intellectual property and copyright.
  • Cyber fraud vulnerabilities.
  • Consumer protection challenges.
• Automation bias can lead users to uncritically accept AI outputs, highlighting the need for AI training for interpreters.

Governance and Leadership in AI

• President Biden issued Executive Order 14110 in October 2023 to guide the development and use of AI.
• Agencies must appoint a Chief AI Officer (CAIO) to align AI initiatives with business objectives, a trend also seen in private sector organizations.

Role of RIM in AI Governance

• AI governance ensures the ethical and safe use of AI through established frameworks and standards.
• Records and Information Management (RIM) professionals play a crucial role in documenting high-risk AI processes.

AI Policies and Ethical Considerations

• Comprehensive AI policies clarify employee rights and obligations, focusing on privacy, bias, and accountability.
• Ethical principles for RIM professionals emphasize integrity, transparency, compliance, and the protection of sensitive data.

Employee Training and ROT Removal

• Effective AI implementation requires employee training on ethical usage, legal compliance, and societal impacts.
• RIM practices should eliminate Redundant, Obsolete, or Trivial (ROT) data to secure AI processes and minimize vulnerabilities.

Legal and Regulatory Compliance

• RIM professionals excel in creating records retention and disposal schedules, crucial for compliance with varied state regulations.
• Specific state laws address issues such as facial recognition use and require regular assessments of AI impacts.

Sensitivity Classification and Data Protection

• Implementing sensitivity labels (e.g., confidential, private) controls data access and enhances security for AI applications.
• RIM professionals ensure the protection of sensitive personal data through expertise in legal definitions and encryption practices.

AI Standards and Guidance

• ISO standards provide frameworks for AI governance, addressing bias and risk management.
• The EU AI Act establishes a risk-based approach to regulate AI, categorizing systems by risk levels to enhance transparency and compliance.

Additional Legislative Frameworks

• Canada’s Artificial Intelligence and Data Act aims for accountability and safety in government AI systems.
• The NIST AI Risk Management Framework encourages organizations to address trustworthiness in AI, with resources for self-assessment.

State Legislation on AI Management

• Various states have enacted legislation focusing on AI use, ensuring bias evaluations and transparency in automated decision-making.
• Examples include California's comprehensive assessment requirements and Illinois's notification mandates for AI applicants.

Paradata and AI Transparency

• Governance and regulations necessitate documentation of AI activities, aiming for accountability and transparency in AI use.
• Research efforts advocate for collecting and preserving paradata as a measure of responsible AI deployment.### Paradata vs. Metadata
• Paradata includes information about the procedures and tools used to create and process information resources, as well as details about the individuals involved.
• It documents the full scope of application and context, beyond just the algorithm.
• Explainable AI (XAI) refers to the understanding of why a tool produced a specific output from given inputs, whereas paradata encompasses the why, how, and effect of a tool's usage in a specific context.
• Metadata documents and preserves the resource itself, while paradata provides insights into the AI process, ensuring transparency and accountability.

Categories of Paradata

• Two main categories: technical paradata and organizational paradata.
• Technical Paradata includes:
  • AI model selection and testing
  • Evaluation and performance metrics
  • Logs generated
  • Model training datasets
  • Training parameters and versioning information
  • Vendor documentation
• Organizational Paradata includes:
  • AI policy, design plans, employee training
  • Ethical considerations and impact assessments
  • Implementation processes and regulatory requirements

Importance of Paradata

• Organizations must document AI usage to ensure responsible implementation and be prepared for accountability.
• Archivists and Records Information Management (RIM) professionals should engage from the planning stage to address new record-keeping requirements resulting from AI deployment.

AI Enhancements in Records Management

• AI can automate content creation and management, reducing sorting and tagging time.
• Capable of processing large data sets and repetitive tasks.
• Provides analytical tagging of images and enhances keyword metadata.
• Functions continuously, assisting differently-abled individuals, and improves decision-making efficiency.

User Insights on AI in Records Management

• A 2023 survey revealed only 16% of organizations utilized AI for records management.
• Among those, common uses included:
  • Content analysis and data extraction (48%)
  • Auto-classification (42%)
  • Metadata extraction (32%)
  • AI usage for retention (19%) and disposition (23%)

AI-Enabled Products Analysis

• In 2022, an examination of 25 AI-integrated products showed:
  • All employed machine learning features.
  • Use of natural language processing in 11 products and deep learning in 5 products.
  • Common AI-enabled features included classification (24 products), extraction (20), and sentiment analysis (8).

Examples of AI-Enabled Features for RIM

• Capture: AI-powered OCR assists with document classification and data capture; Rossum's solution boasts a 96% accuracy rate.
• Analysis: Hyland Alfresco integrates Amazon tools like Textract for data extraction and Rekognition for image analysis.
• AI Integrated Workflow: Microsoft Syntex offers intelligent document processing and features like content assembly and retention linking.
• In-place Records Management: RecordPoint automates records management directly with AI-driven classification intelligence.

Best Practices for AI Readiness

• Keep informed about AI advancements and integration into RIM.
• Understand potential AI risks and governing laws.
• Explore AI standards such as the NIST AI Risk Management Framework.
• Collaborate effectively and adapt successful AI strategies from other domains.

Summary

• In the context of data governance, high-quality data is essential for authoritative records.
• RIM professionals can mitigate redundancy and inefficiency through automation tools.
• Different types of automation like workflow automation or intelligent process automation can streamline operations.
• Recognize AI's dual role: as a tool to enhance records management and as a framework for documenting responsible AI usage.