[orenshved]

Hey everyone
In a panel talk about the future of AI, Bart Selman said that advancements in AI would be uneven, with some areas advancing much sooner and others much later. Could you help shed some light on this and give your thoughts about which fields do you think would advance sooner and which later?
Thanks.

[Don Patrick]

They already are uneven. Current mainstream AI is heavily preoccupied with associative processes, in the form of digital Neural Networks. In the 90's, the situation was reversed: The associative branch of development was trashed while inference-driven AI flourished in the form of Expert Systems. Whenever a new approach is invented, that approach becomes the new hope and attracts all the funding. Currently AI is behind in reasoning, planning, problem solving, inventing, and understanding, while it is further developed in associating, predicting, guessing, translation, and free-form art generation.

When we look at specific applications, the more commercially viable areas will develop before others. e.g. the only aspect of humour that has seen much research is sarcasm detection, because there is a lot of money in mining opinions about commercial products. Conversely, art generation has developed sooner than expected, but mostly as a byproduct. The AI field's history is too erratic for me to make predictions with any accuracy.

[MikeB]

DeepMinds new AlphaFold demonstrates that you can develop a system that solves a problem, but still no one else can solve it except for them, so now everyone has to process their protein folding data via DeepMind. Lots of $$$ in it... so that's one future. The other is people who've had enough of the monopolys and hand write the solution in algorithms or patterns.