4 ms·
It is super-impressive, but peoples expectations are also massively higher because more and more people have become accustomed to things that were "crazy" 15-30
by vidarh 10y ago
It is super-impressive, but peoples expectations are also massively higher because more and more people have become accustomed to things that were "crazy" 15-30 years ago.
Even so, to paraphrase William Gibson, the future is not very well distributed.
On one hand people expect miracles and are not very impressed with systems that are incredibly impressive. On the other hand I've been showing some a side-project that mostly just uses a basic Bayesian model that people get impressed by seemingly because it's in an area where they are used to poor results (I'm using it for a very basic text categorisation problem).
People in general appear to be very bad at extrapolating possibilities and mostly judge these things based on what they're used to rather than any assessment of what "ought" to be possible.
- dasboth 10y agoThis is also true in research in the sense that the moment a problem is solved that previously felt unattainable, it gets somewhat dismissed as "weak AI". The goalposts are constantly moved and the progress to date is underestimated.
- ajuc 10y agoMy friend did a simple python script that asked "give me a few words, with one being in some ways different than the others", and the script returned the word that was different. So, for "dog, snake, sheep, tree, owl" it would respond with "tree". For "blue, green, sky, red, black" it would respond with "sky". It worked by posting a query to google with all the words except the first one, then all the words except the second, etc. And then it compressed the results, and the word that resulted with the smallest compressed result when removed - was the odd one. Very nice use of Kolmogorov complexity estimation. It still blows people' minds few years later, when I describe what it could do, but then I explain how it does it - and they react with "meh, a simple trick, not an AI". That's how people react to AI in general - it's only AI if you don't understand it.
- csydas 10y agoI think the point on this story is a bit different than you are interpreting it. What you're basially describing is the same issue that magic tricks have, where it's fantastic until you know the secret which is often simple. The difference is that with magic, no (good) magician is claiming they actually have magical powers, and that's why some people get really skeptical and dismissive of AI claims. (Not to say you or your friend are claiming you had strong AI, but more to the modern dsicussion of AI in general) The problem really is that there are a lot of people commenting on AI who exist in the "know enough to be dangerous" category of technology. They don't fully understand things about computers or programming, but they know enough to sound like they do. This, combined with the fact that extraordinary claims and articles about AI is great publicity for businesses, there is a tendency towards exaggeration when it comes to the idea of AI. For example, I personally wouldn't call what Google, et. al., do as "AI", but as a subset of the field, focusing a lot on speech analysis. This isn't nearly as evocative as saying "AI Research", but it's much more accurate and I think would prompt more intelligent commentary and questions. When I read your story, I thought it was cool and was curious how you did it; I didn't ascribe AI attributes to it, however, and still found the technique clever as a result. When I hear a claim about "AI" from news currently, I find myself having to distinguish what they're actually talking about first. When there is an inevitable article about AI and people claim that AGI is right around the corner, that is when I get skeptical, and when I think that those who do know a bit about computing get very skeptical and dismissive, since they expected a bit more than more stuff related to just keyword searches. The simple fact is that, like magic tricks, our imagination for AI right now is a lot more impressive than the reality behind current AI research, and a lot of what is presented as AI is more focusing on a specific subset of AI with an impressive front-end to it. This isn't to dimish the work that the researchers are doing, but clarity of language would be preferable to just the umbrella "AI" category.
- bbctol 10y agoI wonder if, as neuroscience research improves, intelligence itself will seem less "magic." As you say, good AI researchers shouldn't claim to have magical powers--as far as we can tell, only these lumps of meat in our head are magical!
- squeaky-clean 10y agoI'd consider this a trick because initially you'd believe the script is what's doing the intelligent part. I think this is an example of impressive AI, but your friend's script isn't the AI, Google is. Imagine using Twilio to text the question to a real human and await their response. The script isn't the "intelligent" entity in either of those cases, just a broker.