Is human reasoning rational?

Shamelessly plugging course work into the blog. Why not, right? Since I have to write once a week, might as well get my blog back up to healthy numbers.

            To tackle the problem of rationality, we must first define explicitly what being rational means. It seems that, through the readings, rationality is taken to be equivalent as deductive logic, which is then studied in laboratory settings and results in surprising performance incongruences. I will address the irrationality of that in the latter half of this essay, but first, I want to share some anecdotal experiences of my encounter with rationality.

            I went to college for engineering, and lived with 3 of my best friends who were all mathematically-inclined and extremely logical people- in other words, the most rational people I knew. We’d have debates on a regular basis, during which we try to subdue each other’s argument by finding logical inconsistencies. But these arguments were not entirely scholarly in nature, as they are often a true reflection of our opinions, which meant that the defeat of one’s argument was equivalent to denouncing one’s rationality, and that was a cardinal sin in the undergraduate engineering population. To give an example, one argument went like this: since we’re all studying to become engineers, and that we want to contribute to society, then we should do everything we can to tilt the balance towards positive progress. If this is the case, then it is only rational that when the time comes to have children, we should instead adopt children from less fortunate circumstances and raise them as our own, which will simultaneously improve the quality of a person’s life by removing them from all kinds of risk factors associated with orphanages, as well as remove a burden from the social welfare system. After flailing about in trying to find a rational counter to this argument backed by any sort of empirical data- stretching from evolution reasons of passing down one’s genes, to the chemical pheromones of one’s own offspring facilitating bonding- in the end, I gave up my own rationality, because I just want my own kid and I don’t quite know why. This was not the only argument to end this way, as it wasn’t a question of IF we’re proven to be irrational, but a question of when and after how many “why”s.

From these exchanges, I concluded that human behaviour is actually very rational, or at least we have the capacity to act rationally. The problem was that at the very bottom level, there are always axioms that cannot be further produced through logical deductions. In fact, this is true of any formal system: a logical deduction must have a point of origin, which serves to be the axioms of the system and are taken unquestioned. And as Kurt Godel has pointed out, no meaningful formal system can be complete and consistent, which boils down to that some things cannot be proven and will have to be taken for granted. Of course, some axioms are useful, and ideally they’re consistent, but they don’t have to be, especially in humans. So really, being rational just means that we are able to go from axiom to theorem, and one theorem to the next, while consistently following a set of syntactical rules. I don’t claim to understand these human axioms or their origins, nor am I a proponent of Freudian theories that some things are just the way they are, like wishing to sleep with one’s mother. But I do believe that there is a clear parallel between the human mind and formal systems. A big caveat, though, is that formal systems exclude the possibility of learning anything new, and if humans strictly operated using deductive (or rational) reasoning, then we will not be able to survive, even as individuals. It could be that humans are rational within our scope of knowledge, but outside of it, we need to be able to act irrationally – inconsistently – in order to learn new behavioural rules to survive.

Having set up my position, I want to address some of the points raised in the readings. On the surface, it feels like the 3 pieces seem to progressively increase in their reasonableness, as a function of time. Newell and Simon’s general problem solver is an archaic and ridiculous model of human cognition, as it “replicated” a very narrow set of behaviors and called it mission accomplished, not to mention its assumption of logical operations in the human mind to start with. Oaksford and Chater, on the other hand, summarized the more agreeable, Baysian models of thought and decision making. But if we take a step back, it should be apparent at the time of each of the three articles, the proposed model for human thought was nowhere near to be the complete blueprint for cognition, but it was simply claimed to be a worthwhile model because that was the state of the art in our advancement of inference algorithms. In other words, these models weren’t cognition-inspired, but we decided to fit our conception of the mind to these models because that’s the best we could do at the time - in the 60s, it was deductive logic, and in the 90s, it was Bayesian inference. Knowing that, perhaps we should be reminded of our humility the next time we claim to have a model of the human mind.

My biggest issues with modeling human rationality, though, are in the following two problems. The first being the scientific methods employed in testing rationality so far, or rather, what we deem to be rational in a laboratory setting. Given that this is a scientific enterprise, I understand that we must tightly control parameters in order to extract meaningful interpretations from experiments. Although, having indeed controlled for the parameters, we must then present conclusions with qualifiers for the experimental settings, and not paint with broad strokes on human rationality. This was touched on in Oaksford & Chater, though I’d like to draw emphasis to it. Plopping a subject down in a room and presenting a deduction type puzzle certainly provides insight on one’s ability to operate within a well-defined system. However, it should not come as a surprise to both the logic and the models people that the subject comes from an outside system, having practiced rationality using a separate set of rules for some 20 years (for your average undergraduate participant anyway). It may be true that the performance in the Wason task significantly improves when a real life scenario is used instead of its original wording, but that should be expected as well, because if you present a calculus problem, people do much better when they can utilize their intuition and experience on velocity, rather than the “derivative of distance”. So in the end, are people’s failures on the Wason task an indication of irrationality, or an inability to translate between parallel situations?

My second problem comes from the disregard for the level of abstraction from the mind to the model, and thus what meaningful conclusions we can and cannot draw from it. In this particular aspect, I agree with John Searle, which is that a model of something is not the thing itself. The deductive GPS mimics the behavioral process a human uses when solving a logic puzzle, then humans must be deductive! Or even now, because Bayesian models can predict human behavior, we must (implicitly) think in Bayesian terms! Then, if we build a robot that can intake food and lie down for 7 hours at night, would we call it human as well? I am aware that no one claimed that the GPS IS a human, but I do believe that we should make these kinds of general comments with extreme caution. This, then, begs the question of can we ever create something that thinks, as no matter how close the abstraction is to the real deal, there is always a gap, even if the model consists of physiologically accurate neuronal processes. To that, I offer the point that no two human minds are even alike, so if the closest thing to a human mind isn’t even really like that particular human mind, how close do we hope to approximate using a machine? But, depending on how much detail we want to smear away, we CAN make (often useful) abstractions and lump certain human faculty together, such as deductive reasoning and probabilistic inference, we just have to be conscious of that fact.

In closing, I found it amusing that while Newel/Simon and Laird attempt to argue that the mind is a symbol manipulating machine, Searle argues that the machine cannot be the mind – and I disagreed with both accounts. I think this really boils down to personal philosophy, and mine is that no matter how close a machine gets, it can never BE human. But it can approximate human thoughts very accurately, depending on the level of abstraction, though it will have to be years removed from the ones in the readings to be considered even close to human. It is, in a way, asymptotic, but since each individual human mind can only be asymptotically similar to another, once a machine gets within range, we may not be able to tell the difference.