belief poll #2

[spoonless]

This time I made sure that the lower end of the range is 0 rather than 1, to make it symmetric with the 10. I don't know why lj defaults to 1-10... I was lazy last time and just left it how they set it up.

I know different people mean different things by choosing different numbers, so to standardize try and do it this way: pick 10 if you are 95%-100% confident that the statement is true. Pick 0 if you are 0-5% confident (in other words, 95%-100% confident it's false). Pick 9 if you are 85%-95% confident it's true. Pick 5 if you are 45-55% confident it's true (in other words, you don't know). I'm going to take browascension's suggestion this time and say that if you're unfamiliar with the topic, just skip it rather than picking 5.

I tried to pick questions that I was a little more agnostic on this time... last time I had too many extreme responses, both from myself and from everyone, so hopefully this one will be more mixed.

[Poll #1438874]

Comments

General Intelligence:

[rws1st.livejournal.com]

General Intelligence: Take the set of all possible problems (with all possible tool aids, and constraints (time), environments etc.). The GI of an agent is the subset of the set of all problems that the Agent can solve.

The does not easily turn into a numeric measure.

Re: General Intelligence:

[spoonless.livejournal.com]

Except that if you put a brand new problem they've never seen before in front of someone who scores high on g-loaded tests, they are more likely to be able to solve it than someone who scores low on such tests. I'm not saying IQ tests are all that accurate or that the resolution they claim to have is meaningful... but if you look at the extreme examples, like someone who's a 60 versus someone who's a 140, there is a pretty striking difference. That's why I chose an "8" for that question... I'm reasonably sure that it's measuring something that deserves to be called general intelligence, but not completely sure.

Re: General Intelligence:

[rws1st.livejournal.com]

We don't disagree about the abilities of standard IQ tests.

In the framework I presented think about it this way:

The set of all problems can be loosely thought about as existing in some space of some dimensionality.

So say some problems take a certain amount of short term memory, or speed of computation, or background knowledge or some such. The IQ tests try and find some boundaries in this space that predict the sub-set of problems that a agent can do.

Now there are limitations. There are certain important real world problems that might required persistent work for months to solve where the person doesn't lose focus or resolve. An IQ test may do poorly at predicting success in these types of problems. And so on for other dimensions that IQ tests don't measure.

But they do measure some important dimensions and so can be useful!