3 ms·

Are you talking about an IQ score? An IQ score is a measure of intelligence, it's not intelligence itself. Just as the temperature you read off a thermometer i

Are you talking about an IQ score? An IQ score is a measure of intelligence, it's not intelligence itself. Just as the temperature you read off a thermometer is only a measure of the heat in the room. The thermometer tells you very little about the position and momentum of the gas particles in the room, only that higher temperatures correlate with more momentum. The value of IQ scores for measuring human intelligence is that they correlate with all sorts of life outcomes. The hypothesis behind the research is that both the life outcomes and the IQ scores are caused by a single factor, known as g, for general intelligence. Obviously, if we train some model on IQ tests we can probably get a machine to ridiculously high IQ scores. Does that mean we have created intelligence? No. The machine is not going to show any of the life outcomes we expect from a human with high IQ. It's not going to do anything! That's why none of these machines are intelligent.

mlyle 4y ago
Yes, I understand all that (and have talked about and mentioned g in other comments in this thread). The point is, what's "improving intelligence"? It's something that we have a hard time quantifying on a linear, agreed axis. As you point out, IQ scores are a somewhat flawed measure of g, which itself is a somewhat handwavy observation of a underlying factor that is a strong correlate to human performance in a broad set of areas. And even if you could put that human performance in a linear measure, there's still a question of how you value different changes in that distribution. Which is better-- raising everyone by .1 standard deviations, or boosting a few people near the top by 1 standard deviation?