There are about several thousand active posters on Veeky Forums, with a (supposed) average IQ of 130...

There are about several thousand active posters on Veeky Forums, with a (supposed) average IQ of 130. This is basically a huge scientific team, so you should be able to solve this:

>Come up with a way to make an AI discover patterns without providing any kind of a pattern beforehand (unsupervised learning)

Other urls found in this thread:

thedailyliberator.com/liberty-discord/
en.wikipedia.org/wiki/Fluid_and_crystallized_intelligence
twitter.com/NSFWRedditVideo

get a gurl prego

What are restricted Boltzmann machines

Two questions:
1. Is the AI allowed to count?
2. Is defining the addition operator considered "providing it with a pattern"?

Define "pattern".

Use an EA that has its fitness bound to how well the end set is sorted. Pretty easy desu. Unless the guidelines for the fitness count as a pattern. Clarification would help.

1. A pre-defined rule of having it somehow evaluate higher values, or "count", yes, but not with a directly programmed number comparison as that's obvious interference from your side
2. Yes

The set {1, 2, 3} - a pattern with an addition of +1

I would venture that it is impossible to devise an unsupervised learning AI

If you're not allowed to "teach" it addition, there is no way for it to organize numbers into a continuum of ascending or descending values. This also would prevent it from checking if two variables were equal.

I could elaborate further.

And yet, humans having unsupervised intelligence is a proof that such things do exist.

>If you're not allowed to "teach" it addition, there is no way for it to organize numbers into a continuum of ascending or descending values.
Maybe that's the problem to begin with - it's supposed to somehow realize that they have to be organized and organize them itself without our interference, maybe something towards this direction?

>This also would prevent it from checking if two variables were equal.
Our universe has differing quantities, the sums of which are measured by gravity and (considering we're not a simulation) without any predefined rule, so there should exist a way

>>Come up with a way to make an AI discover patterns without providing any kind of a pattern beforehand (unsupervised learning)
can you prove that humans are not provided with patterns beforehand from birth?