AI Ethics in Academia

Ethics of Artificial Intelligence in Academia

• Artificial intelligence (AI) has become widely used in daily life, from workplaces to social media.
• AI now enjoys the status due to increased computing power and large datasets available to train such programs.
• A major question is how to uphold the integrity of assessments and research with the rise of AI.
• Ethical discourse on AI application is not limited to education but extends to other fields like social sciences and business.

Rise and Rise of AI

• Science fiction has long explored artificially intelligent beings with human-like attributes.
• Jonathan Swift's "Gulliver's Travels" featured a machine capable of creating prose and poetry which came out in 1726.
• Karel Capek's play "Rossum's Universal Robots" was the first known reference to "robot" in 1921, influencing research.
• Popular entertainment and news media can shape perceptions of AI, sometimes negatively.
• Alan Turing proposed the Turing Test in 1950 to assess AI by mimicking human conversation.
• The aim was, observers would have no more than a 70% chance of correctly identifying the machine after five minutes of questioning.
• John McCarthy coined the term "artificial intelligence" in the 1950s.
• With increased computing power, AI has become widespread across various industries.
• AI could increase global GDP by $15.7 trillion by 2030.
• Artificial Narrow Intelligence (ANI) has infiltrated daily life, but many are unaware of its extent.
• ANI is designed to solve specific tasks like facial recognition.
• Artificial General Intelligence (AGI) can reason and think like humans.

AI Generated Content

• AI content generators use NLP and natural language generation (NLG) methods to produce content.
• Large Language Models (LLM) simulate human learning by collecting information from training datasets.
• Third-generation Generative Pretrained Transformer (GPT-3) is a powerful language model using Transformer processing.
• AI-generated content includes text, images, art, music, and videos.
• Generated content can be partially supervised or wholly unsupervised.
• The media industry has been consuming AI-generated content, like the Washington Post's robot reporter Haliograf in 2016.
• Concerns exist regarding copyright infringement and the future of human creativity and workforce.

AI Ethics

• Ethics involves the rational study of morality and behavior evaluation in moral problems.
• The first ethics course was recorded in The Journal of Philosophy in 1912.
• The biomedical sector initially led ethics discussions.
• Academics are concerned with ethical and social responsibilities in various fields.
• Immanuel Kant's theory of goodwill, Utilitarian theories, and virtue ethics are philosophical theories related.
• Associations like IEEE and ACM have established ethical principles for IT professionals.
• Performance, robustness, bias, interpretability, and privacy are key aspects of algorithmic auditing.
• All discussions on AI and ethics start with Isaac Asimov's Three Laws of Robotics.
• The laws are that a robot may not harm a human, must obey orders, and must protect itself.
• AI in science fiction has influenced ethical discussions on AI development and implementation.
• Bias in system development and dataset collection remains a concern.
• From biased skin tones to gender bias, concerns about biased algorithms came out when using proctoring software tools
• Harvard University and MIT introduced AI Ethics courses in 2018.
• GPAI was formed in 2019, observed by UNESCO, to address AI ethics concerns.
• Frameworks and guidelines often lack binding agreements and enforcement mechanisms.
• The European Union and UNESCO published guidelines on AI in education and policymaking.
• Academic integrity is not the main focus to be taken from the EU and UNESCO guidelines which focus on data protection.

AI in Education

• AI in education (AIED) involves the application of AI in academia.
• Early trends focused on intelligent tutoring services for students.
• Developers have innovated AIED with twenty-first-century learning skills and the fourth industrial revolution drivers.
• Data analytics, powered by machine learning, enables personalized learning and classroom management.
• Object-detection and image recognition enhance safety and learning in schools.
• UNESCO predicts that expenditure in AIED will reach $6 billion by 2024.

AIED Concerns

• Science fiction shapes perceptions of AI, influencing expectations of AIED.
• Constant data capture raises privacy and security concerns.
• One-third of undergraduate students are concerned about privacy issues.
• AIED has been noted to give students to engage more while overcoming disadvantages
• Inequality in AIED access due to wealth disparities is a concern.
• Dataset biases and developer bias lead to issues like racial bias in AIED systems.

AIED and Academic Integrity

• Academic integrity involves ethical principles and standards in education and research.
• Academics strive to ensure knowledge development and skills in the workplace.
• Student cheating, plagiarism, and contract cheating are historical concerns.
• Paraphrasing tools use AI to generate original text, raising concerns about plagiarism.
• Using tools have shown a direct link to poor comprehension of text and that causes inability to demonstrate understanding.
• Tool use leads to a "reductionist perception of language among students" which transforms customer-oriented translation and foreign language education.

Concerns Over Generative Al Content Use by Students

• OpenAI's ChatGPT has spurred discussion in academia due to its conversational way usage.
• Concerns have been raised over student use of generative AI tools.

Cognitive Offloading

• Cognitive offloading reduces cognitive demand through physical actions or tools.
• The use of alarms, calendars, and autosaves are good examples.
• Arguments exist against using technology to reduce arithmetic and writing skills.
• Such calculators, spell checkers, grammar checkers, and even machine translation tools, could be a disadvantage to students when learning
• Offloading tools can improve task performance, speed, and accuracy.
• It will be hard to spot what is Al, for example even any sentences which were completed by the built-in features of Microsoft Word
• Increased "cost of externalizations” can counter negative effects and push people to use "internal strategies"
• Assessments may be redesigned as well to use transparently cognitive offloading.
• Transparency will let an assessor know when students are using tools and content generated, this also encourages academic integrity.

Acknowledgment

• Students should be asked to use generative AI tools and the content that is generated by AI
• If students are allowed to use such tools, then the question becomes how they should acknowledge such use. Students are generally expected to
• It's crucial that the sources are verifiable and the students are aware of that
• AI tools cannot be cited as a source due (a) not being sentient beings who can take responsibility for the output nor be held as credible, and (b) prompts cannot reproduce the same output for verification.
• Generative Al tools cannot be cited as co-authors meaning they should rather be under the acknowledgement or methodology
• If the tool is used only as a support, then the tool's use can be acknowledged
• If the tool is used to generate actual results that are used in the academic paper, then it needs to be part of the methodology with appropriate justification of use, just like tools.

Plagiarism

• AI text generators may involve plagiarism, which is literary theft.
• In fact, the bigger concern may well be the factual nature of the content produced.
• While there may be concern over copyright infringement depending on the source of the dataset used to train the models, the actual content produced may not be a 'copy-paste'.
• Chat GPT can generate factually inaccurate and outdated texts and known what is known as artificial hallucination.

Al Hallucination

• Generative AI may provide artificial hallucination, which means the mode provides output that may not be true/accurate all the time.
• The models are “prone to hallucinate unintended text, which degrades the system performance
• It is very key that the knowledgeable academic, researcher or tool user knows that the tool can make up information to fill in any gaps within knowledge.
• This is a major concern for academics because this means students are presented with accurate and inaccurate information in a variety of disciplines and manners, and
• Such arguments on Al are to be expected and should be addressed.
• As it is a tool for academics, it should be verifiable for accuracy, that will enable users to better understand the concepts they research and cannot distinguish between accurate and inaccurate information