It was a warm sunny noon and to keep selves from breaking sweat, me and a friend of mine decided to remain in-doors. We were debating on cognition and stretched it far and wide till our wits could sustain. A furlong down, we changed gears into Artificial Intelligence and since neither had any background nor had done any homework upon it, both were essentially playing devil’s advocate.
We hit a juncture where both held diametrically opposite stand. She proposed that ‘no computer could ever beat human intelligence because its a human who has programmed its intelligence’. I was aghast! It seemed so flatly wrong. One, the logic basing which she had forwarded her claim seemed quite bleak. Moreover, the claim was clearly smeared in the implicit desire to remain on top of the evolutionary tree and not go redundant. I have always considered an emotionally backed claim as the stumbling block in any good argument. Emotion essentially always is just so plain irrational!
I gathered my thoughts and set to plunder her stand. I held that the proposition could be approached through three steads, each adding to why its plain wrong that ‘a human cant create a computer that beats human intelligence’. Note, by intelligence i do not mean just computational prowess, but all cognitive attributes that we take to be quintessentially human, thus also referring to creativity, imagination, inventiveness, artistic-genius, et al.
So here’s the conquest. Lets begin with the west gate! We humans are intelligent, allegedly, but lets take it to be true for arguments sake at least! Now we didnt require an intelligent designer to make us intelligent. We evolved to this state through purely non-deterministic stochastic processes guided by natural selection. Note, evolution doesnt have foresight, its blind. Now, evolution gives a mechanism for how complex can emerge from the simple. Having said this, the fact that we are intelligent and our intelligence didnt need a designer comes out clear. Savour this, we too are part of this ecosystem. So any creation by us, though in an anthropic sense is man-made, is in fact, still natural because we are also part of nature! When nature, which is blind and non-cognitive, can by evolution create human intelligence, why not we, also part of nature, while with the ability to cognition and foresight, create another intelligence, purportedly better?! Now if you hold the contention that the intelligence as evolved by nature is purely by chance and if by design one were set to create an intelligence, one cant beat ones own ceiling well then, what makes you think that the mental processes happening in our brain are not ‘non-deterministic stochastic processes guided by natural selection’?!
The second approach. Lets barge in through the North gate this time! Consider intelligence as a confluence of faculties. Each faculty being encoded and processed by a neural circuit. Now the algorithm for the faculty could be computational or non-computational, but be it either, it is going to have a rule (a rule to smooth-en out the
randomness in the latter case). If we can figure that rule and better it, we could have an algorithm that performs that particular faculty still better. Extend the argument to every faculty and we have a smarter brain. Note, as my tricky quirky friend pointed out, this second point stands on an implicit assumption that the algorithm for the faculty has no ceiling, ie, that there is no best was of performing a particular instruction which cannot be bettered. While logically true, we hardly see a ‘perfect’ in nature. And all through our evolutionary history, we have been pushing further the optimum and redefined perfect. So i can hope the argument is to stay.
And finally the third one. Lets ambush from all around! A very subtle yet powerful point that we learn from evolution is that randomness, when operated upon a very long long long time, under suitable selection pressure, can churn out stable systems (stable as in accordance to the selection pressure applied). Why not try the same?! Why not use the principle of evolution as a mathematical tool to create algorithms?! Make a mundane program for say ‘identifying faces’. Now create a mutator program. It keeps messing with the coding of the ‘identifying faces’ program every once in a while as the program copies itself into two. Apply a selection through an operator who says when the program identified a face correct and when incorrect. Also include a clause wherein as the correctness of the response rate increases, the frequency of the operation of mutator in the program decreases and the copy-number of the program increases. Viola! We recreated the evolutionary process in an electronic system. The coding is the trait, mutator is akin to mutation introducing variations and the selection-operator playing the role of natural selection providing direction (again in a non-deterministic way). As evolution always occurs in direction of the selection pressure, we are sure to end up with a program that doesnt stop getting more intelligent, though, when given sufficiently long long long time. Thus, we end up with a program that’s more intelligent than us in face identification (note, i have preferred the term face-identification over the more prevalent term of neuroscience face-recognition because there is more to recognition than identification, and i wanted the example to remain simple). Now extend the same to all faculties of brain and bingo! we got a better brain.
And as i finished with my three arguments, i sincerely hoped she would fall prostrate at my feet in surrender. But she instead chose to brush her hair to behind her ear, make a pout and barged into the kitchen to help herself with coffee. Such lovable creature she is! Though am sure i would need some illogical aka dil-logical pleading and prodding to be done before she would make an extra cup for me.