Recently, a representative from a Mauritius-based investment fund approached me for legal advice on obtaining recognition for their foreign investment scheme in Uganda under the Collective Investment Scheme Act.
The client was frustrated with previous legal options that relied on Artificial Intelligence (AI). They specified in the engagement terms that my advice should be based solely on my judgment and not generated by any language model, as they sought my expertise in the capital markets, not generic content.
Another Ugandan client, who exercised Employee Stock Options ‘ESOPs’ in an American Delaware-based startup, also reached out at the same time to seek my legal opinion on whether the recently publicised Uganda Revenue Authority (URA) crackdown on foreign assets held by Ugandans for taxation purposes on worldwide income-would affect him. He showed me some of the legal advice he had generated from these AI Large Language Models. After reviewing it, I had to advise him not to rely on that legal advice because it was incorrect.
I believe certain aspects of the legal profession will not be replaced by GPT Large Language Models. In fact, 65 percent of what lawyers do still remains stubbornly human, despite the fact that the consulting firm, McKinsey, estimates that 23 percent of tasks performed by a highly experienced lawyer can be automated.
I was speaking with a colleague at an international law firm based in London, UK, and he told me his firm developed a custom internal generative AI tool used by the firm’s lawyers to enhance document review, summarise legal documents, and compare inconsistencies in witness statements. This custom-made AI illustrates how various industries are reporting the risks they face that could lead to litigation.
Junior lawyers at that firm have evolved from being drafters and reviewers of documents to becoming checkers of AI output. However, verifying the soundness of contracts requires judgment that AI cannot provide.Unfortunately, fresh graduates rarely possess this skill, as it is developed through years of accumulated experience.
I keep getting amused by people who say current AI models will replace judges entirely. Don’t dream about that. When a judge or magistrate sits down to make a judgment on your case, they go beyond just using syllogistical or inferential logic that these AI models excel at; they also use more of abductivee reasoning, which is much harder to programme and relies on instincts and cultural nuances.
All AI models excel at “reckoning,” not “judgment.” Trust me, if current AI models were to perform judicial functions, it would occasion even more injustice. For now, A I models are just “speed reading plagiarists” that can synthesise large volumes of data sets in less time than a human being ever can, but can’t replace the function of human judges in courts.
Even multi-layered deep neural network AI models with machine learning capabilities that utilise ‘Back propagation’ and ‘Gradient descent’ aren’t a substitute for human judicial officers. Judges and lawyers do much more than being ‘speed reading plagiarists.’
I will conclude with the important lessons from former Soviet Lt Col Stanislav Petrov, which emphasise the significance of abductive logic in the everyday work of a lawyer.
On September 26, 1983, Petrov averted World War III when the “Oko” early warning system, a primitive AI, mistakenly detected five US intercontinental ballistic missiles due to sunlight reflecting off clouds.
Using abductive logic, Petrov deduced that a genuine US strike would involve hundreds of missiles to breach Soviet defenses. He correctly identified this as a false alarm and chose to override the warning system. This incident underscores the critical role of abductive logic in legal judgement and advice.