Current mood:

contemplative

Basics of my Non-Turing Computer hypothesis.
I've been hinting about this for about a year, so I thought I lay my cards on the table and attempt to explain as to what I've been working on.

For years, computer scientists and cognitivist orientated psychologists [and neurologists] have been trying to create an artificially intelligent agent. From the functionalist [walks like a duck, quacks like a duck...] position to the emergent position in the 'AI' community have all been attempted with little or success. No matter how refined such theories have become, none have produced an AI analogous to the intelligence level of even some small animals [dogs and such]. Yet, the pursuit for the 'Holy Grail' of AI still continues for the simple fact that the reward still outweighs any potential risk or pitfall. Ultimately, the idea of having a conscious entity not dependent on limiting biological substrates is more than tantalizing to any businessman looking for a tireless mind able to work on problems, manage physical systems, and create new kinds of technologies to ignore. So, money pours down on such projects, even for just a glimmer of intelligent operation in machines.

My work focuses on what I see is the self-limiting factor in all the theories of AI. That factor happens to be the computers themselves. No matter how fast the processors get, no matter how many logic gates we cram into the recesses of silicon chips computers cannot act as independent or self-made agents. The reason is quite obvious if you have a cursory experience in programming in any modern machine language (C, C++, and etc), that is no matter how absurd your series of commands might be to you and other rational beings the computer will always execute them without question of consequence or implication. One common logical error in programming is divide by zero. It often leads to software crashing, databases becoming irreparably corrupted, and lots of headaches for developers debugging said software. Yet it seems clear that such software should not even run on computers if they are inherently 'smart.' The reality is that processors do not have the inherent capacity to develop logic implications as humans [and possibly other species] can. What is the secret behind logical assessment? Is it an intuition? Is it that logic has irreducible components that are analogical to physical systems? Such questions arise because these are the same questions that many philosophers ask when considering the problem of AI.

My solution tends to seem backward to some because it implies that computers are, in themselves, backward. What I propose is to suspend what preconceptions you come to this exposition with and consider the following. Imagine, if you will, how mental operations might occur. There are certain steps that arise through the jungle of the human brain and mind. Such steps are invariant no matter how many times a person goes through them, or how many times a person ceases to follow them for the simple fact that these steps are necessary to lead to the solution that allows for logical assessment.

The first step starts with our senses. Why the senses, you may ask? Simply put, all animals depend on them to operate, so our patterns of thought are geared toward them and not toward something purely internal [or purely external]. The fact that we learn from what we observe is integral to reproducing how we human animals think. From the senses we accrue the next step: perception. No matter what you may think about the idea of perception it's not what you think, it's what you act upon. It's all the jumbled sensory 'messages' that are ordered by your nervous system so your brain can actually operate on them to make conclusions. Essentially, your perception is your VR/Sense-o-Rama. It's the 'map' of the world you use to make conclusions about the world around yourself. Without it, you can't make any conclusions about the world. Nor could you say there was/is a world around you. The fact that perception, like your senses, is independent of your thought makes this very easy to represent in other physical systems. In fact, everything from operating systems to BIOS control chips are analogically similar to perception [the nervous system] in that both OSes and BIOS control chips order the state of computers for them to operate properly. So too does your perception order your senses for you to operate properly.

The third step is the one where things become complicated. That step is mental conception. This step can be broken into a number of other steps; measurement, isolation, and integration. Measurement is obvious enough to assume because everything we do with regard to our mental states is a form of measurement. We measure the quantity of objects, and we measure their quality. Our measurements allow us to order what we think into sets that are hierarchal within themselves and in relation to other such sets. Isolation may seem redundant, but in fact it's not. Isolation is not measurement for the simple fact that a person can still measure an entity without isolating from others. In fact, one has to measure an entity against another entity [mental or material] for one to know what one is measuring. Isolation, on the other hand, requires you to exclude all other entities from it. Isolation allows one to accrue all the data required to represent the entity being isolated in one's mental states as a concept. In short, isolation allows you make a map or copy of the entity you perceive.

Integration is important as it is the step in mental conception that allows you to store what you perceive, measure, and isolate in your memory. Integration is an 'emergent' step in relation to measurement and isolation in that both measurement and isolation allows one to easily integrate what one perceives into a concept for memory. Yet, integration is not automatic, it requires something extra: free will. Free will is important here because if we automatically integrated every thing we perceived, measured, and isolated, then what would become of our capacity to remember things? Clearly humans don't literally integrate everything perceived just by cursory examination of one's daily routine. At work, we don't remember every conversation held. At home we don't remember every little quip or joke on a sitcom we watched. And in our sleep, we don't remember every dream as if it were an automatic procedure. Ultimately, our free will is that little trick that allows us to ignore other entities in perception and to focus on the ones we wish to integrate. In short, free will is our halting command. In this context, computers do not have an internal halting command of their own self-control. Such halting commands or statements have to be hardwired or setup as logic loops that will be tested in which one case of said loops will yield a FALSE [or 0 bit] state, thus allowing the computer to continue operation with the rest of its programming.

With this basic blueprint of how the human animal thinks, it may be possible to recreate these steps within a non-Turing computer. Why did I choose the non-Turing paradigm? Simply put, it is the best possible candidate for success. Non-Turing, or hypercomputation, is a branch of computer science that examines physical systems to understand why they operate. Entire sections of Chaos Theory, neurology, and psychology all are devoted to this venture. One particular mathematician, Kurt Godel, hit upon a possible means to discover the methods required to make artificial intelligence possible. Godel observed that certain kinds of logical propositions and systems lead to absurdities [infinities] when they are forced to self-reference. One example of this would be recursive functions in ANSI C. When a recursive function in ANSI C is given no test condition [logical loop] it will operate forever, referencing itself without end. Essentially, such a function is left to roam in a hall of mirrors of its own nature. So, the test condition is a reference made outside of the recursive program. Godel would call a recursive function with a test condition an open system. Certain logical systems mirror this across many kinds of situations [Catalog Paradox is a good example if you wish to search for it at Wolfram's website], with no end in their operation due to self-reference. The solution, as it was for recursive functions, is to reference another system outside of the logical system itself. Basically, certain kinds of systems must operate as if it is not 'alone' or self-contained. Computers, specifically our current paradigm of the Turing Complete kind, are unable to operate like this on their own, they must be given test conditions to ensure they work properly. And in many cases such test conditions require human intervention to continue [Installation software is a great example of this as it queries the user for the next step in its procedures]. Free will in this context may be that test condition in mental conception that allows humans to operate with sufficient capacity as to be rational agents all on their own. Yet how we do quantify the nature of free will? I would consider this question to be the most important problem of this proposition only because I have yet to find anyone to sufficiently solve it [Not even myself]. The steps of mental conception as they stand [sensation, perception, and mental conception (measurement, isolation, and integration)] will be easy to implement in part, excluding integration [and free will] from the current set of steps. Yet, when only implemented in part, we do not get the desired result, but we do get other kinds of results such as semi-autonomous robots like the kind used by NASA on Mars, or 'intelligent' agents as they are used in certain hospitals to double check a doctor's diagnosis. Such 'intelligent' agents are only as good as the information we give them, so their intelligence is the result of ours. Essentially, they are condensations of human cognition into a more utility driven form.


And with all that being said, I think I'll end this post by stating feel free to reply with questions, criticisms, and what not.