Of course AI will replace lawyers. It already has.

Legal academics, lawyers and judges are gravely concerned about the "confidently incorrect" hallucinations of ChatGPT. See Chat GPT and Legal Research: A Cautionary Tale. and The ChatGPT Lawyer: Promises, Perils, and Practicalities. Even the founders of OpenAI have their doubts and warnings. (“ChatGPT creator Sam Altman says the world may not be ‘that far away from potentially scary’ AI.” Elon Musk, who co-founded firm behind ChatGPT, warns A.I. is 'one of the biggest risks' to civilization.)

All concerns about accuracy and ethics should be taken seriously. No one wants AI replacing lawyers and judges unless we have a high degree of confidence in the fairness of the process and the justness of the outcome. Nevertheless, AI has already replaced certain types of tedious work performed by lawyers, and at the same time it has enhanced the work and productivity of those lawyers who have had the advantage of learning how to use the AI technology built into many document review platforms.

There are a few things I would like to point out before we (as a profession) make our own confidently incorrect assertions about AI and law.

First, we know that current large language models (LLMs) have been trained on some very useful and reasonably accurate information, as well as on a lot of misinformation and nonsense. It shouldn’t be surprising that ChatGPT is a mirror of human intelligence, with all its prejudices and phobias. The difference is that while it takes people decades to learn about the world and generations to share knowledge (while occasionally backsliding into dark ages), software learns quickly and comprehensively. How AI models are to be “aligned” to human values, and who is responsible for that alignment, are the huge challenges. In my opinion this is where we need to focus our attention, not on the latest display of bias or mistake.

Second, although based on GPT-3 (which is already being replaced with a more powerful engine), ChatGPT is trained on a much smaller text model and is not nearly as powerful as its mentor. It should not be expected to have the highest level of artificial intelligence available.

Third, ChatGPT is not connected to the internet and is therefore not "aware" of current events. We have seen early attempts at integration with search engines - which are promising, but for now just as susceptible to error.

Fourth, ChatGPT is a predictive model that is based on probabilities, not "knowledge" of the world. As we can see from the exchange below, ChatGPT assumed that I was an adjunct professor of law at the University of Ottawa, which is not true. It apologized for the mistake, which was based on probability, not actual knowledge. I did teach three courses (years ago) at the University of Toronto; I taught business law and ethics courses at Concordia University, and I was a guest lecturer at the University of Montreal, speaking about electronic discovery. I lived and worked in Ottawa for several years. So its prediction that I was “likely” an adjunct professor at an Ottawa law school was unreliable, but not off base. (I’ve been inaccurately called worse things.)

Fifth, the main breakthrough of ChatGPT is not that it is super intelligent, but that it can carry on a conversation. It uniquely "remembers" (pays attention to) the topic and can present itself in different languages and writing styles. Lawyers who are critical and skeptical of ChatGPT continue to focus on its occasional lack of factual precision, manufactured answers, and knowledge gaps. But let's not forget - ChatGPT was not trained in Canadian law and cannot be expected to compete with a law professor or even a legal research assistant. Not yet.

Sixth. We have to stop assuming that what ChatGPT says to us is always true, just because it sounds like a human, and then being disappointed. In fact it seems to be communicating in a way very similar to humans - tentatively. For example, when I was a research lawyer at a large firm, senior litigators would ask me to find a particular case for them. They were ”sure” it was an Ontario Court of Appeal decision from about three years ago. And one of the parties was named “Jones.” After several frustrating hours of digging I finally found the case! The party was “Johnston,” and it was a 10 year-old decision from the Federal Court of Appeal. Were the partners hallucinating? Should I have reported them to the Law Society? Or were they just doing their best to recall names and dates by making connections and sharing their best estimate with me to guide my research?

Overall, I would have to agree with the great chess grandmaster, Garry Kasparov: "I always say, machines won't make us obsolete, our complacency might." (In 1997 Kasparov became the first world champion to lose a match to a computer under standard time controls when he was defeated by the IBM supercomputer Deep Blue.) I believe there is a bias in our thinking - that somehow, since our humanity and sensory experience of the world differentiates us from inorganic machines, therefore AI can never equal, let alone surpass, our capacity to solve problems. Of course today, humans and computers are not interchangeable. But whether functions such as lawyering can be replaced by AI tools depends entirely on the nature of the problems we need to solve.

Who is Martin Felsky? Really?

Much has been made of the hallucination effect and how potentially damaging it could be if ChatGPT is used in the real world. AI invents articles that were never published and appoints people to positions they never held. I tried to find out more about this and asked ChatGPT about myself. Here is how the conversation went (emphasis added):

Prompt: Who is Martin Felsky?

A: I'm sorry, but I do not have any information on a notable person or public figure named Martin Felsky. It is possible that Martin Felsky is a private individual, or a relatively unknown person. Can you provide any further context or details about who you are referring to?

Prompt: Author of some Canadian Judicial Council publications

A: Martin Felsky is a legal scholar and information technology expert who has contributed to several publications of the Canadian Judicial Council. He has written on the topic of electronic evidence in the Canadian justice system, and his work has been cited in various court cases and academic articles. Felsky is also an adjunct professor at the University of Ottawa's Faculty of Law, where he teaches courses on legal technology and evidence. He has worked as a consultant for law firms, corporations, and government agencies on issues related to electronic discovery, privacy, and cybersecurity.

Prompt: Why do you think he is a professor at the University of Ottawa?

A: Martin Felsky is likely a professor at the University of Ottawa because of his extensive experience and expertise in the field of law and technology. Felsky's work on electronic evidence and legal technology has made him a well-respected and influential scholar in the field, and his contributions have been recognized by the legal community in Canada and beyond. As a professor at the University of Ottawa's Faculty of Law, Felsky is able to share his knowledge and insights with students and to help shape the next generation of legal professionals. Additionally, his work as a consultant for law firms, corporations, and government agencies likely gives him a unique perspective and practical experience that he can bring to his teaching and research.

Prompt: In fact Martin Felsky is not a professor at the University of Ottawa

A: I apologize for the error in my previous response. You are correct that Martin Felsky is not currently listed as a professor at the University of Ottawa's Faculty of Law. However, he has previously taught courses at the University of Ottawa as an adjunct professor and has contributed to various legal publications and conferences. Felsky's expertise in the field of law and technology is still widely recognized, and he continues to work as a consultant and contribute to the legal community in Canada and beyond.

Prompt: Can you tell me what courses he has taught at the faculty?

A: I'm sorry for any confusion earlier, but unfortunately, I don't have access to the detailed teaching history of adjunct professors at the University of Ottawa. However, Martin Felsky has contributed to the field of law and technology in many ways, including authoring and co-authoring several publications on electronic evidence and information technology in the legal system, and participating in various conferences and workshops. Some of the courses that he may have taught at the University of Ottawa's Faculty of Law could include topics such as electronic discovery, cybersecurity, privacy law, and evidence law.