|Ben Trevett||Home | GitHub | Twitter|
What is intelligence? The stereotypical example is someone who is good at math. What does being “good at math” even mean? Someone who can multiply two 10-digit numbers in their head in under a second seems to be “good at math”. What about someone who isn’t quick at mental math but can solve a complex mathematical problem after methodically working through the correct steps? I’d say they were also intelligent. Of course, intelligence isn’t just tied to mathematical ability what about someone who can write a great novel? Or can compose a beautiful piece of music? Surely they’re intelligent too, and there’s no guarantee they’re good at math.
What about someone who is pretty good at math and composing music and writing and a few other things, but they’re not outstanding in any single one of those? These are people who are more generalist (a jack-of-all-trades) rather than specialist. People with broad skill sets, where their knowledge goes more than surface level, are usually seen as intelligent.
OK, so we’ve decided that intelligence isn’t tied to a specific field. It can’t be measured by ability in one subject. It should be a vector rather than a scalar. What does this mean for IQ tests which measure intelligence as a scalar? I’ve only ever done online IQ tests and pretty much all of them have questions in the form of “given a sequence of patterns, find the next glyph”. Are IQ tests actually measuring intelligence, or are they measuring pattern matching ability? If they’re just measuring pattern matching ability, how well does the ability to pattern match correlate with intelligence? In Thinking, Fast and Slow Kahneman talks about how our System 2 (the slower, logical process of thinking) is used to learn to see patterns through experience which our System 1 (the faster, instinctive process of thinking) uses. For example, chess experts can see obviously good move quicker than me because they’re played a lot more chess and so are able to quickly recognize patterns from past games. So, it’s all pattern matching? Well, what about creative subjects like writing and music? Most authors read a lot and most musicians probably listen to a lot of music. Are authors and musicians recognizing patterns when they make their own patterns? Authors and musicians create their art in a certain style which is usually influenced by past authors and musicians, so they’re re-applying patterns they’ve recognized.
If it’s all about pattern matching learned through experience then with enough time, say 10,000 hours, can’t we all become intelligent? Gladwell’s 10,000 hours theory only works when it’s applied to a single skill, and we already said that generalists are intelligent too. What if we split the 10,000 hours into 2,500 hours in four different skills? David Epstein’s Range argues that specialists are more desired in today’s world, but does demand correlate with intelligence? You would prefer to hire intelligent people over unintelligent ones, that’s for sure. But do you need to be intelligent, or do you just need to be “good enough” generalist? Is specialization, becoming a “one-trick-pony”, actually an unintelligent thing to do? However, most people who are specialized, and thus we would say intelligent, are usually pretty good at other things too, so their pattern matching ability learned from one field is being applied to another field. What about people who are “gifted” and are solving complex mathematical equations from a young age? They haven’t put in 10,000 hours of studying, are they just naturally more intelligent than the rest of us? Are they born with a 10,000-hour head start? You don’t need to be a child prodigy to be intelligent, but I’m sure it helps.
So, we can all be intelligent with enough time spent on learning a skill that requires pattern matching. But isn’t this a biased way to look at the world? What if I’m poor and have to spend most of my day travelling to and from my job where I have to work long hours at a physical task? When I get home I’m going to be too exhausted to study mathematics for a few hours. A quick and efficient bricklayer it definitely skilled at their job, but most people wouldn’t rush to call them intelligent. Is this a biased view of intelligence? Or is the world biased against these people, so we just assume they aren’t intelligent? Is it because we assume anyone can be a skilled bricklayer but only the best can be a skilled knowledge worker? But didn’t we just say anyone could put in 10,000 hours to become intelligent? So it turns out the only thing stopping us is enforced poverty.
If intelligence is pattern matching then machine learning should be the shining example of intelligence as machine learning is all about pattern matching. But it turns out that machine learning algorithms, specifically deep neural networks, are too good at pattern matching, they match patterns that we don’t want to match. They can be fooled by adding noise to images, changing a single pixel in an image, or by the presence of weird looking stickers. They learn textures when we want them to learn shapes, and now they can be fooled to think an apple is a library by simply writing the word “library” on a piece of paper and sticking it on the apple. Can we blame neural networks for this? Surely not. We asked them to be the best pattern matches, and they did, it’s not their fault we did such a poor job. But are they intelligent? I think you’d have difficulty convincing someone that they are.
What is intelligence? Intelligence is not pattern matching, but “intelligent” pattern matching And what makes “intelligent” pattern matching actually “intelligent”? Pattern matching, of course. And not just any pattern matching, but intelligent pattern matching. It’s intelligent pattern matching all the way down!
Note: This article massively anthropomorphizes intelligence, but considering we want artificial intelligence to do things humans currently do, shouldn’t we be comparing it directly to human intelligence?