Doug Lenat Died Yesterday

May his common sense inspirations flourish forever!

Indeed, common sense will carry on despite most attention being focused on machine learning models, those statistical exercises that have grown substantially in stature over the last dozen years.

Study enough AI history and you’ll find many different, competing schools of thought, most of which rise and fall, only to reemerge and steal the show, then fall from the limelight once again. Browse through Wikipedia’s take on the History of artificial intelligence for more detail.

Doug Lenat was father to a large family of AI projects based on symbolic logic, the most famous of which was Cyc, which he created. An attempt to capture and encode a vast knowledge base of common sense, it’s been going for over forty years! Between 1984 and 2017, he and his team invested something on the order of 2,000 person-years building Cyc’s 24 million rules and assertions.

Cyc has been used inside various commercial software applications but it has never caught on, perhaps because it’s not open enough to encourage everyone to exploit it — an issue that Gary Marcus raised in his tribute to Doug. I encourage you to read the tribute to learn more about Lenat.

Symbolic logic is what’s missing in statistical AI. Stat-AI consists of machine learning, artificial neural networks, large language models and the flood of extremely large scale statistically based AIs that Google researchers foreshadowed when they wrote about statistical transformers six years ago. With 60 or more intervening inventions, these statistical models, coupled with huge datasets on which to train them, using enormous collections of “GPU” hardware, we’re now in a world of statistical chatbots (like ChatGPT.)

Lenat (and Gary Markus) released a paper this summer exploring the limitations of statistical AI and the benefits (and limits) of integrating symbolic knowledge into statistical systems.