I get news updates relating to artificial intelligence, and I thought it would be nice to record some information about them.
There has been a lot of talk recently, and not even that recently, about the inadequacies of the Turing Test. The problem with the test is of course that it’s pretty easy to write a program which produces output that looks a little bit human-like, and if you make things easy on yourself by pretending that it doesn’t speak English particularly well and is a child then you can sometimes fool people into thinking that it is human.
Nonetheless, I kind of like the Turing Test. I feel that nobody who was permitted extended, or repeated, interactions with a chatbot would be fooled, and, although I realise that changing the rules post hoc is a little no-true-Scotsman, the test is ultimately about distinguishing intelligence from nonintelligence, something which programs which succeed by taking advantage of the strictness (or laxity) of the rules, like the above-linked Eugene Goostman, clearly do not exhibit.
The Winograd Schema Challenge attempts to address rule-benders by imposing stricter rules. Specifically, participating programs and humans would have to give answers to common-sense questions such as "The town councilors refused to give the demonstrators a permit because they feared violence. Who feared violence?”. Any answer other than “the councilors” would be wrong. By contrast, Goostman’s reply to the question "Which is bigger, a shoebox or Mount Everest?” was "I can’t make a choice right now. I should think it out later. And I forgot to ask you where you are from…”
So it seems like a nice idea, but ultimately this represents a subset of human interaction as represented by the Turing test, specifically the subset dealing with factual questions related to semantic understanding. It seems a bit like a big data problem. Not that I'm suggesting it's easy.
Articles about the Winograd Schemas Challenge: Livescience; Vox; Commonsense Reasoning. Winograd Schemas Challenge.
The end of law firms.
There are plenty of AI-related predictions which start with “The end of”, but this is a particularly weird one as the prediction hinges on viable general-purpose AIs, essentially, which would do much more than render lawyers obsolete.
Legal Futures, The End Of Law Firms
Movie: The Imitation Game
I haven’t seen this Turing biopic yet as I instinctively doubt it will do the man justice, though I did like the NY Times piece which argued that movies must portray genius as insensitive and cold because it gives us an otherness against which we can contrast ourselves.