I think this may turn out to be a very interesting test of AI ability for abstract thought. That's funny -- but AI models don't get the joke Date:July 31, 2023 Source: Cornell University https://www.sciencedaily.com/releases/2023/07/230731122233.htm To understand a joke you need a general understanding of what is a normal experience and what about abnormalcy can be funny. This is usually associated with biology . But can AI learn to make a distinction between normal and different in a "funny" way?
No, AI is more than that. With complex systems, there are emergent abilities that are over and above the sum of the parts. Intelligence (biological or artificial) is such an emergent ability. The GPT series AI can think via a feedback system that kinda resembles human thought processes. GPT AI can think! Check out Leta and the GPT4 who probes her record.
Even many humans might not be able to conceptually explain why some events or stories are funny. In those cases, the laughter or whatever response is instinctive and any knowledge or understanding of "why" is tacit. AI may be generating jokes from one of its emulative, statistically-guided block boxes -- without the definitive rules/principles, templates, etc that humans use for demystification. There are categories of humor, theories of humor, philosophy of humor, etc. That is, if we can covert humor to explicit knowledge -- precisely articulate why something is funny and classify it and explain why it belongs to _X_ type, and codify such as instructions... Then AI has what it needs to explain or understand why _X_ was funny. Actual laughter or an inner feeling of mirth is unnecessary. Since physical description as well as the procedural operations of computer hardware do not capture and carry the qualitative characteristics and manifestness of human experiences, anyway: feelings, visual images, odors, etc. Even if machines did have private states, they can't be accessed and vetted. The same would be the case with "other humans" if each individual didn't have their own inner proof and be forced to morally and via consistency attribute the same capacity to the rest of the bipedal herd. _
I believe it has something to do with the listener's expectation built by an elaborate story that leads to a completely unexpected closing "punch line" and shocks the listener's sense of logical conclusion. The laughter is the cognition and admission of having been fooled (trapped) into a totally unexpected turn of events. I wonder what an Chat AI would make of this little skit by Victor Borge, that contains many subtle variations of humor. Can AI watch TV?
