Lately while doing research on automated language translation, I’ve come to realize that there isn’t a clear, concise, well accepted definition of human language itself. A quick check on google reveals the wealth of interpretations. So then, kindly, let me proffer one more.

Language is the serialization of thought.

The term “serialization” should be ready accessible to programmers et al. For others, a quick explanation is in order. Serialization is the process of taking a complex (e.g wide) entity and transforming it, re-constructively,  such that is can be passed through a much simpler (e.g narrower) channel. A typical requirement for correct serialization is the ability to de-serialize the serialized data to result in exactly the original entity.

A very simple analogy would be “serializing” a bunch of untidy children through a narrow gate, one child at a time. The only caveat being that, if the serialization process was perfect, then the children would regroup on the other side in the exact same configuration as before the serialization started.
Once the concept of serialization is clear, the intent of my definiton of language should also be clear, although you may or may not agree with it.

If we go along this line of thought (pardon the pun), a corollary immediately follows:

Language is the ultimate compression engine.

Why?

If we believe that human thought is one of the most complex phenomena known to us, and if language allows serialization of a complex thought into a small compact representation that can be communicated and stored in a myriad of ways, and ultimately easily de-serialized by target humans to reveal the original thought, then the corollary must be true.

When one thinks hard to solve a problem, it is likely he or she uses bits of language in self-communication to focus attention on particular aspects, annotate the intermediate results, and proceed one by one onto higher level steps. Although the most revealing flashes of insight most likely occur during a moment of unbounded “massively parallel” thought, knowledge of language undoubtedly plays a role in allowing the thinker to carry out elaborate thought experiments.

In primitive cultures (or in someone never exposed to the concept of language), undoubtedly the enterprising inventors of that time devised their own methods of mentally labelling specific ideas with individual symbols, and then using those symbols to ease the task of deriving higher level constructs.

Also, if the above corollary holds weight, I think it has some additional fantastic implications.

This means that if we were to one day achieve “brain dumps”, or “downloading a brain” and their ilk, the best format to allow accurate storage and re-construction would be plain text! If the brain could somehow be tricked into emitting a high speed lecture on its current and past states, then the language best suited for that would be the mother-tongue of the brain’s owner. Nothing else we conceive will probably ever come close in accuracy or compactness.

Of course, this does not bode well for automated machine translation attempts. To be fully successful at that task, implies the ability to de-serialize a piece of text into the original speaker’s thoughts. If we accept that human thought is one of the most complex or mysterious activities known to mankind, then we are accepting that automated machine translation is a pipe dream for many years to come.

On the other hand, if the code of language does get cracked soon, will it mean that human thought is not so complex after all?
 

What do you think is the most difficult AI problem of all? I’m not sure there is even a debate around this, but regardless, I’d like to clarify the issue, at least for my own sake.
Here are some candidates, let me know if I missed something important.

1. Natural language understanding.
2. Vision, Image/Scene understanding.
3. Creating “consciousness”.

I’m very sure some of you will immediately rank the difficulty in the order presented above, with natural language understanding being easiest and creating consciousness most difficult, and scene understanding being somewhere in the middle.

I, however, would like to argue exactly the opposite.

Looking at things from an evolutionary point of view, what evolved first, and what evolved last?

1. “Conscious” beings emerged first as life evolved, reacting to various stimuli in their environment.

2. Soon various life-forms were able to “see” their surroundings and react accordingly.

3. Much later, only humans, the pinnacle of evolution’s achievement yet, are able to create and learn an innumerable number of communication mechanisms using symbolic languages, in a multitude of mediums including sound, light, surface markings and so on.

Consciousness, at one level, could be treated just as the presence of a feedback loop. It could be argued that a vast number of computer systems in existence today are already “conscious”, although the degree of consciousness varies significantly.
Vision, or image understanding, is present in species vastly lower on the scale than the human, i.e insects, birds and so on. And since the known communication mechanisms in those species are enormously rudimentary compared to human speech and language, and coupled with the fact that even species only a notch below humans cannot “talk” in the fine detail that we do, language must be the most difficult piece of engineering to achieve (I do know this one personally, I’ve been trying very hard to build an NLU algorithm for a while).

If you find flaws in my reasoning, please do point them out.

Because if my assertion is true, the corollary would be: “The moment a complete NLU system is built, it means the code of intelligence is cracked, and that fully intelligent self-aware beings are then only a matter of combining the various existing building blocks”.

Uptil yesterday, I used to think that thinking in 4 spatial dimensions is a purely analytical play (and “Time” as the 4th dimension never seemed intriguing enough anyway). But I just finished reading “The Möbius Strip” by Clifford Pickover, and now I believe I am actually able to mentally and somewhat visually integrate at least some key concepts. Now I not only accept, but also understand why two infinite planes which intersect each other in our 3-D world, could actually be non-intersecting planes in a 4-D world. The book has many other such amazing concepts explained with remarkable ease, and is highly recommended for the spatially inclined. If you are an origami aficionado, it is a definite must read.

There are theories out there that electricity, magnetism, gravity, and all the invisible forces are manifestations of the unseen 4th dimension. Sounds plausible enough. But String Theory is said to include 10, 11 or even 26 dimensions. Just like our 2 intersecting planes in 3-D can be made non-intersecting by adding a 4th dimension, it appears physicists keep adding an extra dimension to their theories to “get around” every new spatial incongruency they encounter in their thought experiments.

The inevitable questions..does “intelligence” have anything to do with all of this? Does it “emerge” by way of some unseen interaction in the 4th dimension. Is this why despite the hundreds of thousands of hu-man-hours spent pondering this problem, we have not been able to truly emulate a high form of intelligence? Does it mean that because of this extra-dimensional dependency, we will never be able to grasp the real mechanism?

I’m certainly not convinced that the brain’s functioning is somehow dependent on a 4th spatial dimension with a separate manifestation over and above the basic forces of nature (electricity, magnetism etc). I think “massive parallelism” is basis of the the complexity (quite a traditional view) that we are unable to overcome in our thought experiments. In fact, massive parallelism is the cause of the complexity in everything science is not able to do well today. The weather, fluids, the stock market, are all examples of systems where a huge number of entities comprise the system, and most of them are in turn dependent on even more other types of variables.

The problem in understanding our own brain’s function, I believe, also stems from some kind of innate capacity limit upon an intelligence trying to fathom its own composition.  For example, by my definition, to be called an “intelligence”, it must be “aware” of itself, or rather, of “something” that it perceives to be itself. What form that “itself” takes is the crux of the matter. If someone built a computer simulation of an intelligence which solves some problems given some reward/pain stimuli, imagine what it would look like if this intelligence suddenly understood its own construction! This would mean that it has become “aware” that it is a piece of software code which runs inside a computer. But get this, this would mean that it also understands what “software is”, and even what a “computer” is. This would imply that at the instant of its birth, it is pretty much as intelligent as the average hu-man!

For those that have pet dogs, I’m sure they appreciate the levels of emotionally complex and intelligent behavior these animals are capable of. They are all “intelligent” beings, and also self-aware. But does this mean that every dog out there primally understands that the seat of its intelligence lies in a glob of soft matter just behind its eyes? Certainly not. Even in hu-mans, most likely this knowledge is usually gained by virtue of the education system. To be a successful self-aware intelligence, it need not imply being aware of the low level physics of the awareness.

Imagine if a particular hu-man suddenly becomes fully self-aware. He can close his eyes, and look at his brain solve a mathematical problem, while solving it. Neurons are firing left and right for solving the problem, chemicals are rising and ebbing in tiny but significant portions, but at the same time this same brain is also looking at itself doing all of this activity. This act of “looking” would also cause various other neurons to fire, which would then need to be looked at as well. Sounds recursive, doesn’t it? I guess this is why I always get a headache whenever I contemplate this stuff too much.

A simple way to get out of this recursion would be to get a hu-man A to look at the brain activity of another hu-man B. As soon as A’s brain truly understands what is going on in B’s brain, it means A’s brain is able to simulate B’s brain to completeness, all the while still retaining the identity of A. This happening can only mean that A has a vastly more architecturally complex brain than B (not just higher IQ, but structurally more advanced). This means that the above scenario is not possible where A and B are both hu-mans.  To grasp why this must be generally true, think what would happen if while A is looking at the activity of B’s brain, B turns around and starts trying to follow A’s brain. Have you ever pointed a live video camera at the TV it is connected to? In summary, I theorize that:

a) To observe and understand an intelligence of complexity X, we would need another intelligence of at least complexity X^2. 

From which follows the corollary:

b) If an intelligence A is able to comprehend another intelligence B’s construction, it automatically means that B can never hope to comprehend A, no matter how hard it tries.

Where does all of this take us? I think it means that we will never be able to simulate human intelligence at the drop of a hat, or by turning on a computer and running a program, because only an intelligence vastly more complex than humans can do that.  A complete human like intelligence, if it has to run on different hardware than hu-mans themselves, will HAVE TO BE EVOLVED. And since we cannot replicate the hu-man hardware, the intelligence MUST EVOLVE DIFFERENTLY. As controllers of the hardware we will initially retain god-like status in controlling the direction of the evolution, but we will never know what lies next. Inevitably, one fine morning, a surprise will await us all.

Does the brain exist and evolve to prolong the survival of the body? Or does the brain use the body to prolong its own survival? Or are they one and the same thing and should not be discussed as two separate entities?

Hypothetically, is it possible to “evolve” a brain as complex and intelligent as a human’s, without attaching it to a human like body?

On the flip side, is it possible to create a human like body and expect it to survive out in the world, without providing it with a human like brain?

To me, asking questions like these instantly exposes the flaws in traditional AI techniques. Everything that has been built so far, learns only because it is force-fed information that we want it to be able to process.

If it doesn’t care about life and death, can it learn?

If it doesn’t yearn, does it learn?

For the last couple of years I’ve been wrapping my brain around the question of knowledge representation and the decision making process within the ..uh, brain. I’m not fully there yet, but I am close to the conclusion that the basic principles can be modelled using simple probability theory, applied in a hierarchical manner. The deeper the hierarchy, the more intelligent the being, in general.

 Conversely, looking at all the research around modelling neurons directly in software for emulating brain like behavior; maybe all that goo in our cranial cavity is actually nature’s way of building something that can do mathematical operations, albeit massively in parallel? Do we really need to model the brain dendrite by dendrite, axon by axon, synapse by synapse to be able to get the same basic functionality in software?

Recently I have been researching and writing code for sentence segmentation, a rather well-beaten topic in the research community, particularly for segmenting chinese sentences into words( chinese is ususally written without any spaces betweeen the words/characters). Since I only wanted to create a generic algorithm( and I don’t understand chinese), Ijusttakearegularenglishsentenceandremoveallthespacesbetweenthewordsasmytestinput. (I just take a regular english sentence and remove all the spaces between the words as my test input.)

The algorithm then figures out where exactly to put the spaces such that the output will make sense…or rather, the “most” sense. For example: consider a fragment of an input “isawhim”. This could be “i saw him”, or it could be “is a whim”. If you knew absolutely nothing else about the context in which this was said, what would you choose? Probably the first one. But if you knew that this fragment was preceeded by an “it”, as in “itisawhim”, then its obvious that the second choice is the better one, because “it i saw him” doesn’t “make sense”. Consider the case of “comealong”. If you knew it was succeeded by “now”, then it would be “come along”, but if it was succeeded by “way”, it would be “come a long”. Usually its not just the directly preceeding or succeeding words which provide the evidence, but many words past and beyond.

In layman terms, the algorithm works by creating a running measure of “goodness” across a streaming input of continuous characters without spaces. It doesn’t need to know a start or an end.  At every point it evaluates the thousands of possible valid combinations that could be formed over a large number of characters, and eliminates those which would cause goodness to drop in the future. The goal is not to move in the direction of short term maximum goodness, but rather long term lack of badness. In other words, it will do somewhat bad things to get a good future, short of doing something suicidal.

The “long term” is the key here. In time based terms, some activity happens at the milli-second levels( like interpreting the sounds in sentences we hear as words), some happens at the level of  “fractions of a second” (applying meaning to the words as we interpret the sentence), and some at seconds( the meaning of a sentence, the emotion within a sentence ). The important thing to note is that activity at higher levels affects past activity at lower levels. For example, someone starts saying the sentence in a nice tone “jack is a…”, the words “great, nice” etc start flashing in our minds, but then the tone changes, and it goes “..complete jerk”, with sarcasm. At this point, the nice image of jack dissolves, and the sentence is interpreted in a different way.

 Taking this up to higher levels, its not difficult to visualize how the task of going from point A to point B allows us to tolerate the nastiness of a bad journey, because the pleasure of reaching is much greater than problems encountered during the journey. Going up a level, point B might not actually be a particularly enjoyable city, but we accept it because it might result in a future career advancement.