Does it possess intelligence?

Many artificial intelligence experts believe that the Turing test is no longer useful to determine if something that is claimed to possess intelligence is indeed intelligent. There were many other tests proposed, but none, to our best knowledge, is efficient enough for Aigarth. In this article you’ll find a description of an intelligence test that is so far the best we have. The test uses zero-knowledge proof as a key element improving efficiency, and (which is important!) allowing to detect intelligence in an Artificial Superintelligence (ASI).

Let’s assume that we have a set of common animals. You get the following list:

Owl, Pigeon, Eagle, Sparrow, Raven, Pigeon, Kiwi.

Is this list random? Clearly not, because there are only birds (we hope you are intelligent enough to know that a kiwi can be a bird sometimes) and even the fact of “Pigeon” appearing twice, which may happen if the animals have been chosen truly randomly, doesn’t fool you.

Now there is another list:

Dog, Owl, Fox, Ant, Eel, Cat, Cow.

Still not random, because they all are 3-letter animals. This time your knowledge of the language helps you to spot non-randomness.

We see that your knowledge and ability to reason have just helped us to detect that you possess intelligence. In a general case our intelligence test consists of asking if some data are random or not and counting correct answers. Of course, AI, which is being asked, may provide random answers trying to guess, but by using same logic as in the Alibaba cave experiment we can still get the ultimate answer with the required level of assurance.

Some of you may ask if randomness tests break our intelligence test. Well, look at the picture above, you should see Merylin Monroe face there (upside down to make it less obvious), drawn in ASCII art style. Maybe not this but another picture will look random to the randomness tests while being obviously the opposite.

And finally, we must know if the data we use are random or not, what if they look random to us (human beings with IQ around 100) while an ASI sees some pattern in them. We cannot feed the ASI with truly random data only, the test would become skewed, but we can add noise to non-random data. This should work because, after all, we, who possess natural intelligence, have been working with data full of noise all the time. If the ASI gives correct answers almost all the time then we shuld increase the level of noise, theoretically this should work, practically we’ll be able to verify this not as soon as we wish.