It is certain that one day it would be happen.

Thank you for your kind answer, but I do not use to believe anything, and this is why in my question I have asked for a prove instead of a prophecy or a good desire.

Dear Joachim Pimiskern,

I appreciate your answer and the links. Nevertheless, I am also able to perform similar programs for a fixed collection of pictures. However, face analysis can work over an infinite set of examples and neither a human nor a machine can handle infinite sets. This is why in my question I am requiring a proof and not a particular case.

To illustrate the topic, consider an infinite collection of all possible pictures each of which together with a description of its expression. A machine can find up the proposed picture in its memory and, then it can retrieve the description. But this requires an infinite picture set E. To do the same task by handling only a finite information subset S, this subset must generate the whole set E, therefore it is required some generating procedure. It is the existence of such generating procedure what I am asking for.

To understand the question properly, let us see another illustration. A machine can calculate square roots by means of a two-column table the first column of which contains every integer and in the same row, but in the second column the corresponding square root can be retrieved. With this table together with a searching algorithm, lying in any machine memory, every square root can be calculated but, it is required an infinite table, since the integer number set is also infinite. However, using any known algorithm machines can calculate any square root handling a finite information set. This is possible because the calculating algorithm generates all square roots from a small table of them.

Notice that with an infinite two-column table every problem can be solved. It is sufficient to contain the problem in the first column and the corresponding solution in the second one. But such a table must be infinite. Indeed, handling infinite tables every problem can be solved by a machine, perhaps in an infinite time!!!. Indeed,

for some particular subset the method works fine, but only for the contained cases, unless some generating procedure is available.

It is clear that human mind possesses a lot of generating procedures by means of which it is possible to retrieve a lot of information from a picture. Usually observing a face any human can notice whether or not the face corresponds to a very intelligent man, whether or not he seems to be aggressive, how old could be, whether he looks like Mozart or some film actor etc. but since only finite sets can be handled, some generating algorithm are needed. Of course, when a human can perform these processes must be because the problem is possible. I am aware of this circumstance, but human mind from pixels builds continuos images, while machines do the converse process, that is, split continuos images into pixels or information units. From this difference arises my doubts.

Best regards.

Dear Ines Domingues

I appreciate your "good intention", but in my previous post I have declared that an algorithm to detect expressions for some collection of pictures I am also able to build. Even equipped with a learning algorithm increasing its powerfulness.

My questions is wider. The literature I had in mind when I have written the sentence that you have quoted is included in the collection entitled

"50 years of artificial Intelligence, Lecture Notes in Artificial Intelligence vol. 4850, Springer Verlag (2007)."

For instance at page 11 you can find the following claim:

The holy grail of AI, the creation of general (human-like) machine intelligence (think of HAL from Stanley Kubrick’s 2001 or Bladerunner ’s Roy Batty), has remained elusive. Natural intelligence is far from being understood and artificial forms of intelligence are still so much more primitive than natural ones. Seemingly simple tasks like object manipulation and recognition –" tasks easily solved by a 3-year-old – have not yet been realized artificially". A look at the current research landscape reveals how little we know about how biological brains achieve their remarkable functionalities, how these functionalities develop in the child, or how they have arisen in the course of evolution.

Dear researchers,

After analyzing the answers I have obtained to my question , I have observed that nobody has payed attention to the subconsciousness part of human mind. Psychoanalysts, like S. Freud have devoted a lot of literature to this topic. I know, from my own experience, that every mental phenomenon is the result of the information obtained from the external world together with the influence of subconsciousness. Perhaps do you know machines equipped with a subconsciousness machinery? If so, this is a good topic to be analyzed. In any case, a soon as possible, I will write a partial solution from a very algebraic view point.

Dear Thomas,

If you read my posts, you can see that I have declared that I am able to build such a kind of software, and I have no interest in any software. Perhaps my English is poor and perhaps I cannot make myself understood, but I am only interested in some proof either for or against the existence of some procedure. I am working in this topic and as soon as possible I will expose a partial proof, based upon algebraic invariant group theory. Perhaps, when this partial solution will be exposed, I will be able to get software-free answers.

Thank you for your interest.

Best regards.

Dear contributors,

I hope, that the following formalism can give rise to some proof for the proposed question.

The method I propose consists of detecting properties. To this end, consider the set E of all pictures satisfying some property P. It is not difficult to see, that any transform sending a member X of E into another one Y preserves the property P. In other words, P remains unaltered under every transform between two members of E. ¿How can we formalize this method?

First, let us formalize pictures. Let R be a screen, that is to say, a countable subset of R^2. If C denotes a color set, then a picture is nothing but a map f: R ➝ C; accordingly the set of all possible pictures is C^R.

A transform between two pictures is induced by a bijection ß: R ⟶ R; for instance the composition f·ß is a transformation of the picture f. Let G denote the group of all transforms. If #(R) is N then the cardinality of G is greater or equal to N!.

It is a known fact that every property P determines a subgroup P(G) of G consisting of all transforms preserving P. If two systems of picture treatment can detect the property P the corresponding transformation groups must possess the subgroup P(G). Now, it is easy to state a negative criterion, namely, if the transformation group of the visual memory of some human contains P(G) and the transformation group in the memory of a machine does not, then the human can detect P and the machine cannot, and vice versa. However, a positive criterion is very difficult to be stated since the cardinality of G is N!.

For instance, a picture of 1000 000 pixels is of cardinality 1000000! which is a google of goolgles of goooooooogles!!! Even if we only consider the Sylow's subgroups of G the number is excessively large. If humans can simplify this task, perhaps evolution has provided them with some powerful algorithm.

In any case, as far as I know, in actual artificial intelligence researche the subconsciousness part of mind is not considered. The mental phenomena in any human is always the combination of the information obtained from the external world and the subconsciousness matter. This is why I think that while machines will not be equipped with a subconsciousness machinery, they will not be able to do some tasks that a child can perform easily.

Notice, that this post only deals with algebraic structures, and it is prejudice-free.

From now on, I hope that no new posts about software will be added to this thread.

Best regards.

If you want an existence proof, then just consider a massive machine simulating every neuron in the human eye and brain. If it behaves the same as a brain, then it recognizes everything you want.

Of course, it is possible to program machines to interpret sounds and images. However, my dog is able to know my mood without being programmed. Perhaps, because it feels similar sensations in similar situations. I am not negating nor affirming any possibility. What I mean is that, presently, machines only work as prothesis of the human mind, not as autonomous systems.

I think this is possible in some extent by using image processing techniques. By which on machine can study face and eye expressions and detect the emotion.

Dear Juan-Esteban,

I'm going to go out on a limb here since AI is not my field. The possibility of such a proof or disproof probably lies not so much in the algorithm as in the concept of information. As you said, the theoretical structure of such a machine works by splitting "information" into elementary units like bits or pixels or what have you. Facial recognition algorithm, for example, analyses the relative positions of facial feature points. All such "information" comes in as external data. As long as we treat "information" as a kind of external data, we most probably will not be able to have an AI system that detects, for example, internal sadness, although an AI system that detects sad expression on a face is not difficult. Shannon's work (A Mathematical Theory of Communication) in 1948 merely took "information" for granted and the entire exercise was to mathematise communication. We may need to rethink what information really is. When we have a different understanding of "information", we then may have some idea of whether we can come up with an algorithm to detect internal sadness, not just a sad facial expression.