ASK HN: AI was always a probability problem?
The article discusses the historical perspective on AI as a probability problem, particularly referencing the early medical diagnosis program MYCIN. It highlights the challenges of reasoning with uncertainty in both medical diagnosis and language understanding. The author notes that the lack of a robust method to integrate probabilities with logic contributed to the decline of symbolic AI.
- ▪MYCIN was an early medical diagnosis program that dealt with reasoning under uncertainty.
- ▪Language understanding also involves complex probability issues, as many sentences can be parsed in multiple ways.
- ▪The decline of symbolic AI was partly due to the absence of a useful approach to combine probabilities with logical reasoning.
Opening excerpt (first ~120 words) tap to expand
This was my opinion circa 2011 or so when I was recapping the old AI.Consider, for instance, the successful early medical diagnosis program MYCINhttps://en.wikipedia.org/wiki/Mycinwhich like any kind of diagnosis process is a problem of reasoning with probability. Language understanding has the same issue, like if you wrote a grammar for English you'd find that common sentences have 1000s of possible ways to parse and you will need to either make a guess or keep your options open.MYCIN had a half-baked approach to reasoning about uncertainty that worked, one of the reasons why symbolic AI fell out of favor was that nobody developed a generally useful approach to bolt probabilities onto logic.
Excerpt limited to ~120 words for fair-use compliance. The full article is at Ycombinator.