In the 1980s, Richard Susskind helped build the world’s first fully operational AI system that actually worked for lawyers. It ran on what he calls knowledge engineering: sitting down with a deep expert in an area of law, translating what they knew into rules, and mapping the whole thing out as an enormous decision tree. The one he worked on had over 2 million paths through it. The clever part, he says, was never the software. It was the humans doing the analysis that went into it.
Nobody builds systems that way anymore, which is roughly his point about the systems everyone is building today. He thinks large language models are simply the current enabling technology, and that in 10 years we’ll find it remarkable that we ever thought this was how you did AI. What hasn’t changed, he says, is the moral and social questions. He expands on the argument in a 30th anniversary edition of The Future of Law, published by Oxford University Press.
