Can You Beat a Machine in IQ Test-Like Games?

[EPitter]

In this page you can play against a machine in IQ test-like games. Very funny.

www.theiqchallenge.com

 

[Curious3141]

WTF ? I tried the Numeric Solver game where I took turns with the machine. I could solve all of its trivially easy sequences nearly instantly, but it couldn't solve any of mine. Of course, mine are tough, but that machine is certainly not very bright !

All jokes aside, the numeric solver simply asks you to complete your sequence correctly and archives the info. I think this "intelligent" machine is just exhaustively searching known sequences for the answer. What a disappointment.

 

[davee123]

All jokes aside, the numeric solver simply asks you to complete your sequence correctly and archives the info. I think this "intelligent" machine is just exhaustively searching known sequences for the answer. What a disappointment.

Sounds like it's a neural network, and they're trying to feed in data by getting people to throw problems at it. Far easier to let 10,000 people enter a few problems each rather than to sit there yourself feeding it information. Essentially, it's like you said insofar as it catalogs sequences it encounters, but the difference is that it tries to look for relevant patterns within the solutions, and apply them elsewhere.

Once they get it to start recognizing primes, I'll be impressed. And if the program gets "smart enough", they can use it on REAL problems. But I'd bet it's a long way off yet.

DaveE

 

[Curious3141]

Essentially, it's like you said insofar as it catalogs sequences it encounters, but the difference is that it tries to look for relevant patterns within the solutions, and apply them elsewhere

Well, I wish it joy of effort. The two sequences I gave it were hard, I would consider one of them very hard.

The first one was 2, 6, 33, 280, 3245, ?

The machine didn't get it. I was going for (n^n + n!) I would consider this one to be a "hard" sequence, but probably solvable within a few minutes of thought by a person with an IQ of over 150.

The second one was a sequence I'd consider very hard :

1, 5, 3, 7, 2, 7, 27, 8, ?

This is a nice sequence representing the sum of the distinct digits in the decimal expansions of the reciprocals of the natural numbers before they repeat for the first time (if that). I gave this once to a person with a measured IQ of over 180, and he got it after some difficulty.

Both those sequences are in Sloane's database (the latter is my contribution to it).

Of course, I realize I was aiming very high, but if the computer had gotten those, it would've knocked my socks off.

 

[NateTG]

Well, you can do things like:
Number of letters in the German word for the number
4, 4, 4, 4, 4, 5, 6, 4, 4 ,4, 3, 5, 8, 8...

But any sane person would expect to see:
4, 4, 4, 4, 4 -> 4

 

[Curious3141]

Well, you can do things like:
Number of letters in the German word for the number
4, 4, 4, 4, 4, 5, 6, 4, 4 ,4, 3, 5, 8, 8...

But any sane person would expect to see:
4, 4, 4, 4, 4 -> 4

That would be too open ended to be a "nice" (or fair) sequence. It is a tough one though !

 

[davee123]

Well, you can do things like:
Number of letters in the German word for the number
4, 4, 4, 4, 4, 5, 6, 4, 4 ,4, 3, 5, 8, 8...

But any sane person would expect to see:
4, 4, 4, 4, 4 -> 4

Well, I believe they're trying to expressly forbid that type of problem on the "numeric" version. Effectively, it should be mathematical in nature, and not something bizarre. By the same token I could list 2,3,5,6,8,9,10,12,13,... as numbers where the digits are curved, or 1,1,1,1,1,1,1,2,1,1,1,3,1,2,... as the number of syllables in the English pronunciation, or 10,5,9,6,6,6,7,8,8,5,... as the number of letters in the last names of US presidents.

Problems like that are interesting, but unhelpful to the program for learning, and will probably only serve to confuse it.

DaveE

 

[NateTG]

Why not just start by populating it with sloane's db then?

 

[davee123]

Why not just start by populating it with sloane's db then?

They might've! Hard to say, really. If they did, it's likely that it didn't help, or else at the very least it would've been able to figure out a sequence of primes.

It did give me one I couldn't solve right away. Anyone?
1, 2, 6, 14, 29, 56, 102

DaveE

 

[Curious3141]

Why not just start by populating it with sloane's db then?

Sloane's is just too complex to be ported en masse to this project. At least, they haven't done it yet, otherwise it would've gotten the two sequences I posed it.

 

[EPitter]

Why not just start by populating it with sloane's db then?

After reading the project page ( www.kitbit.com ), I don't think they're going to do it. It doesn't look like a DB based AI model. I might be wrong, but I think the high expectations they have put on their system are not compatible with making a huge DB.

 

[perry123]

Serie solved

It did give me one I couldn't solve right away. Anyone?
1, 2, 6, 14, 29, 56, 102

the solution is:

1, 2, 6, 14, 29, 56, 102, 176, 289, 454

 

[davee123]

the solution is:

1, 2, 6, 14, 29, 56, 102, 176, 289, 454

So, what's the logic behind it?

DaveE

 

[NateTG]

The best I can figure out is that the sequence (including the extension) can be written as:
$$x_i=2*x_{i-1}+p(i)$$
Where $p(i)$ is a quartic polynomial.