# Intelligence over simple data

Let's say we already have an intelligent system (as a computer program). We give it some simple data from outside world, and system tries to predict it. The question would be: what kind of patterns this system should be able to recognize?

For example, system gets binary data stream(zeroes and ones).
1) Should it recognize patterns like: 1 0 11 0 111 0 ....?
2) Should this system be able to recognize prime numbers in ascending order written with "1" with "0" between them?

Is there any research about it?

• Welcome to AI! Are you meaning "intelligence" in the context of "information of value", or the application of algorithmic intelligence to random datasets? If the latter, "Intelligent analysis of contextless data" might be more helpful. Interesting question, nonetheless.
– DukeZhou
Oct 4 '17 at 20:37
• Douglas Hofstadter talks a lot about this kind of stuff in one of his books. Not G.E.B., but one of the other ones. Maybe "Fluid Concepts and Creative Analogies". Oct 6 '17 at 0:28
• @mindcrime What do you think would be a more suitable title for this question? (It's interesting and I feel like there's a useful answer out there, but no one with the relevant knowledge is going to find it.)
– DukeZhou
Oct 8 '17 at 1:42

It depends on the algorithm.

A classical algorithm would not be able to recognize the pattern of primes unless it was programmed to do so. This pre-programming would not have to be exclusive to primes, but the algorithm would have to "understand" the nature of primes including 1 (the number can only be divided by 1 and itself.)

Because computers are good at number crunching, it should be quite possible to program an algorithm that looks for numerical relationships, such as primes, parity and factors, without being taught those specific concepts.

Extending this to alternate representations of numbers, 1=1, 11=2, 111=3 requires a type of abstract thinking (i.e. the literal representation is not the true meaning. In an 8-bit byte 11101101 = 237. In Base-2 1,11,111 are the primes 1,3,7;)

However, if the algorithm was superintelligent, figuring out the example would be trivial, and, ostensibly, correctly deriving patterns of any kind, no matter how opaque, would be possible, given enough time and memory.

In my opinion: If we could predict accurately what an intelligent system capable of, it ain't really intelligent.

About research, maybe facebook AI chatbot being able to develop its own language qualify as one.

As @DukeZhou stated, "context", If we build an AI system that ONLY writes creative poet, I can't imagine it would be able to launch nuclear attack or take off power grid! but what about building a general purpose AI? it should be able to surprise us in the context of performing unpredictable behavior.

• It is certainly possible to predict an intelligent system's behavior when you run it second time over the same data. If such system is functioning in a very simple environment, it should be possible to predict its behavior analytically. Oct 5 '17 at 10:43
• hmm, why would the system behavior unchanged while iterating over same data?! shouldn't intelligent system learn? Oct 5 '17 at 16:11
• @SteepSlope if it's a deterministic algorithm, the intelligent system would produce the same result for the same data, but purely deterministic algorithms are rare for stronger AI these days. My comment would be that intelligence alone does not connote human-level intelligence, superintelligence, or even strong AI, just basic decision-making or analytic capability (but I'm sort of a fundamentalist in this regard b/c my own endeavors partly involve the utility of AIs that are not-so-smart.)
– DukeZhou
Oct 5 '17 at 20:16

If the system would be able to predict with certainty doesn't It mean the data set input would be a product of a world of determinism and the system had to be created using the deterministic laws of the world whence that data came from? It's a tough one because randomness to me reeks of no laws because they create order not random behaviour and a world with no known laws would be systemless as systems employ laws.

• Main characteristic of an intelligent system is the ability to acquire a knowledge by itself. If laws of the world are already known, then what is needed is a model with adjustable parameters, not an intelligence. Oct 5 '17 at 9:47
• You might be interested in this excerpt from the Strong Free Will Theorem: "Some believe that the alternative to determinism is randomness ... It may well be true that classically stochastic processes such as tossing a (true) coin do not help in explaining free will, but, as we show, adding randomness also does not explain the quantum mechanical effects ... It is precisely the “semi-free” nature of twinned particles, and more generally of entanglement, that shows that something very different from classical stochasticism is at play here."
– DukeZhou
Oct 5 '17 at 20:25
• That's all a bit beyond my paygrade, but I thought you'd be interested that you're not the only one dissatisfied with classical stochasticism ;)
– DukeZhou
Oct 8 '17 at 0:08
• @Steep Slope reality as we perceive it is limited with its associated parameters imagination enables us more parameters but they are not the subject of the physical world our brains are contained within. The physical world is attempted to be explained with reference to these parameters and it's varience within it. The reduction of it by this method as far as I know is limited and does not give the complete picture but even reduction and holism and it's attempts could be conceived as reductionism it's an enigma rapped in a puzzle rapped in a mystery mate and that's the eternal fascination.
– Bobs
Oct 8 '17 at 0:29
• @Steep Slope systems don't aquire knowledge they process a state applied to that system based upon the laws it employs the outcome of that state is detetmined unequivocally as a result of that initial state and the process that system has upon that information that results in an outcome of the initial state and is replicable. Consiousness is not a system whose determination and outcome is a result of the conditions you feed it thats absolute, not independent ,finite and replicable.
– Bobs
Oct 8 '17 at 5:19