![\"\"](\"https:\/\/world.edu\/wp-content\/uploads\/2023\/11\/collective-learning.jpg\")
<\/p>\n<\/p>\n
Generative artificial intelligence (GenAI) tools like ChatGPT based on Large Language Models (LLMs) are revolutionizing the ways we think, learn and work.<\/p>\n
But, like some other forms of AI, GenAI technologies have a\u00a0black box nature<\/a>\u00a0\u2014 meaning it\u2019s hard to explain and understand how mathematical models compute their output.<\/p>\n If we as a society are to employ this new technology on a broad scale, we will need to engage in a collective discovery process to better understand how it works and what it is capable of.<\/p>\n As AI experts work on\u00a0making AI systems more comprehensible to end users<\/a>, and\u00a0as OpenAI, the maker of ChatGPT<\/a>, navigates leadership shakeups and questions\u00a0about its strategic direction<\/a>, post-secondary institutions have a critical role to play in enabling collective learning about GenAI.<\/p>\n For AI systems based on large neural networks with a black box nature, like GenAI, a lack of transparency makes it\u00a0difficult for people to trust<\/a>\u00a0the AI and to rely on it for sensitive applications.<\/p>\n Carnegie Mellon University professor Elizabeth A. Holm\u00a0has argued<\/a>\u00a0that black box AIs can still be valuable if they produce better results than alternatives, if the cost of wrong answers is low or if they inspire new ideas.<\/p>\n Still, cases of matters gone horribly wrong erode trust, such as when ChatGPT\u00a0got tricked into giving instructions to make a bomb<\/a>, or when it\u00a0accused a law professor of a serious crime he didn\u2019t commit<\/a>.<\/p>\n This is why researchers working on\u00a0AI explainability<\/a>\u00a0have tried to devise techniques to see into the black box of neural networks. However, the LLMs behind many GenAI tools are just too large and too complex for these methods to work.<\/p>\n Fortunately, LLMs like ChatGPT have an interesting feature that previous black box neural networks did not have: they are interactive. Think of it this way: we cannot understand what a person is thinking by looking at a map of the neurons in their brain, but we can talk to them.<\/p>\n A new field of science is emerging under the label of \u201cmachine psychology\u201d to understand how LLMs actually \u201cthink.\u201d<\/p>\n New research, yet to be peer reviewed, is examining how these models can surprise us with their emergent capabilities. For example,\u00a0researchers surmised<\/a>\u00a0that because with LLMs every new word generated depends on the sequence of words that came before it, asking an LLM to work through a problem step by step may produce better results.<\/p>\n New studies<\/a>, not yet peer reviewed, on this \u201cchain of thought\u201d technique and variations of it have shown they improve outcomes. Others suggest LLMs\u00a0can be \u201cemotionally manipulated\u201d<\/a>\u00a0by including phrases like \u201care you sure?\u201d or \u201cbelieve in your abilities\u201d in prompting.<\/p>\n In an interesting combination of these two methods, Google DeepMind researchers\u00a0recently found<\/a>\u00a0that for a series of math problems, one LLM improved its accuracy significantly when it was prompted with \u201ctake a deep breath and work on this problem step-by-step.\u201d<\/p>\n Understanding GenAI is not something only researchers are doing, and that\u2019s a good thing. New discoveries that users have made have surprised even the makers of those tools, in both delightful and alarming ways.<\/p>\n Users are sharing their discoveries and prompts in online communities such as Reddit, Discord and dedicated platforms such as\u00a0FlowGPT<\/a>.<\/p>\n These prompts often include \u201cjailbreak\u201d prompts that succeed in getting GenAI tools to behave in ways they are not supposed to. People can trick AI\u00a0into bypassing built-in rules<\/a>\u00a0\u2014 for example, producing hateful content \u2014 or\u00a0creating malware<\/a>.<\/p>\n These rapid advances and surprising outcomes are why some AI leaders\u00a0called for a six-month moratorium<\/a>\u00a0on AI development earlier this year.<\/p>\n In higher education, an overly defensive approach emphasizing flaws and weaknesses of GenAI or how it allows students to cheat is ill-advised.<\/p>\n On the contrary, as\u00a0workplaces start to see the benefits of GenAI-powered employees or workplace productivity<\/a>, they will expect higher education to prepare students. Students\u2019 education needs to be relevant.<\/p>\n Universities are ideal spaces to forge co-operation across research fields, an imperative of developing responsible AI. Universities, in contrast to the private sector, are best positioned to embed their GenAI practices and content within a framework of ethical and responsible practice.<\/p>\n One thing this entails is understanding of GenAI as\u00a0an augment, not a substitute, for human judgement<\/a>\u00a0and discerning when relying on this is permissable and acceptable.<\/p>\n Educating for GenAI involves developing critical thinking and fact-checking skills, and ethical prompt engineering. It also involves understanding that GenAI tools do not just repeat their training data, and can\u00a0generate new, and high-quality ideas<\/a>\u00a0based on patterns in that data.<\/p>\n The\u00a0ChatGPT and AI for Higher Education UNESCO Quick Guide<\/a>\u00a0is a helpful starting point.<\/p>\n <\/p>\n Quick Start Guide for ChatGPT and AI in higher education @UNESCO<\/a> https:\/\/t.co\/Z2ZPasJHG1<\/a> #aiedu<\/a> #ai<\/a> pic.twitter.com\/e9vDfJl5tL<\/a><\/p>\n \u2014 Lidija Kralj (@LidijaKralj) April 21, 2023<\/a><\/p><\/blockquote>\nDifficult to understand<\/strong><\/h4>\n
Machine psychology\u2019<\/strong><\/h4>\n
Collective discovery<\/strong><\/h4>\n
AI and learning<\/strong><\/h4>\n
\n