Otherwise entitled: Descartes writes a 17th century turing test. I was rereading Descartes “Discourse On Method” when I came across something completely unexpected. Descartes the 17th century polyglot was concerned about humanoid machines. He writes a guide on how to tell the difference between authentic human intelligence and machine intelligence. What in the hell is this doing in a manuscript published in 1637?
Across his canon, Descartes makes reference to human beings as meticulously constructed machines. In his “Meditations”, he frets that when he looks out of his window down onto the street at passers by, “what do I see from the window but hats and coats which may cover automatic machines?” For him, it is believable that a machine which performs the physical functions of a man would be constructed, even that it would excel the capabilities of an ordinary man. He doesn’t forecast it, he doesn’t predict a timeline, but he believes that it is possible and he is worried enough about robots passing for humans that he feels the need to argue that we will always be able to tell the difference.
In the “Treatise of Man”, one of Descartes’ posthumously published works, he describes such automatons that he himself encountered. In a certain castle that he visited, one could stepped on a pressure plate tile that was “so arranged that if, for example, they approach a Diana who is bathing they will cause her to hide in the reeds, and if they move forward to pursue her they will cause a Neptune to advance and threaten them with his trident; or if they go in another direction they will cause a sea-monster to emerge and spew water onto their faces; or other such things according to the whim of the engineers who made the fountains.” In fact there has been a great deal of interest in humanoid automatons throughout history.
The most remarkable example that I have come across comes from 12th century mathematician and general genius Al-Jazari. In his book entitled “The Book of Knowledge of Ingenious Mechanical Devices,” he writes, “I say that I was commissioned by one whom it is impossible to disobey to construct a boat upon which were figures of boon-companions and of a company of female musicians and servants of the court.” He says, “Its functioning: the boat is placed on the surface of a large pool, and is seldom stationary but moves in the surface of the water. All the time it moves the sailors move, because they are on axles, and the oars move it [i.e. the boat] through the water-until about half an hour has elapsed. Then, for a little while, the flute player blows the flute and the [other] slave-girls play their instruments with sounds that are heard by the assembly…They do not desist until they have performed about fifteen times.” The point of these machines was not to be labour saving devices. As far as I can tell, they were first and foremost spectacles.
So why does he write this turing test-esque advice? He was giving an explanation of how he studied the human body, attempting to account for things like how “heads, shortly after being severed, still move about and bite the earth, even though they are no longer alive.” He says that this shouldn’t be too surprising to anyone who is “cognizant of how many different automata or moving machines the ingenuity of man can make, without in doing so, using more than a very small number of parts…For they will regard this [human] body as a machine which, having been made by the hands of God, is incomparably better ordered and has within itself movements far more wondrous than any of those that can be invented by men.” For the atheist reader, replace “hands of God” with “natural selection”, and his basic idea holds up quite well. But isn’t it odd, how we and Descartes compare the human body to a machine? The human body preexists man-made machines, so wouldn’t it be more sensible to compare “automata” and “moving machines” with a living body?
The human body was not made to be like a machine, instead machines are made somewhat in the image of the human body. Whenever I look at a digger operator, I imagine the operator and the digger forming this cybernetic individual, the operator imposing his will with his absurdly strong robot hand. The digger is a stronger, bigger, better human arm. Descartes finds it completely likely that human-like machines “might perform many tasks very well or perhaps better than any of us,” but only specific tasks. He is thinking that the human body might be exceptionally well designed for general use, but there might be bodies and materials that would excel humans at specific tasks. Indeed, the course of robot design hasn’t been leaning towards machines that have the same physiognomy as human beings. If you look at a factory assembly line, you won’t see humanoid robots, you will see arms, pincer-gripping fingers, and the likes. It is an effective deconstruction of the human body down to the strictly necessary components, strengthened and hardened.
So according to Descartes, how can we tell that a human-like machine is not a “true man?” When it comes to monkey machines (apenoids?) “we would have no way of recognizing that they were not entirely of the same nature as these animals,” whereas when it comes to the humans, “we would always have two very certain ways of recognizing that they were not at all, for that reason, true men. The first is that they could never use words or other signs, or put hem together as we do in order to declare our thoughts to others.” Of course, a chatbot uses words and signs, although they don’t seem to “put them together as we do in order to declare our thoughts to others”. They are still, in some measure, “call and response” machines. Although for us maybe it is less inconceivable that a machine will never “declare” its thoughts. What he is getting interestingly wrong is he thinks that the automation of the body is easier than the automation of mental work. We are learning from the success of LLMs that knowledge based workers have as much reason to fear for their job security as operational workers, perhaps even more.
Descartes says, “one can well conceive of a machine being so made that it utters words, and even that it utters words appropriate to the bodily actions that will cause some change in its organs (such as, if one touches it in a certain place, it asks what one wants to say to it, or, if in another place, it cries out that one is hurting it, and the like.)” The design philosophy he is assuming is analogous to good old fashioned artificial intelligence. The earliest AI systems were largely designed as "call and response machines" that relied on hard-coded rules, mathematical logic, and symbolic reasoning to simulate intelligence. These systems were built on the premise that human reasoning could be reduced to formal logic, which could then be replicated computationally. Descartes notices that the world that humans navigate by means of their reason is so diverse that there isn’t, even in man, an intellectual organ that is designed for every single scenario that one might come across. We have found that the organ-based AI coding strategy just doesn’t work for complex problems. We cannot hard-code a response to every situation, even in game environments like chess, because there are too many possible scenarios to even render them all, let alone figure out a good response for each of them.
Descartes goes on to explain the difference between animals and humans. He says, “the fact that they [animals] show more skill than we do in some of their actions, we nevertheless see that they show none at all in many other actions. Consequently, the fact that they do something better than we do does not prove that they have any intelligence, for, were that the case, they would have more of it than any of us and would excel us in everything.” Descartes understands intelligence as a general thing. A human is intelligent, not because they can do math or build tools or anything else in particular. We are intelligent because we are multipurpose.
The origin of his opinion about human intelligence is his theology. Humans are endowed with intelligence by our creator, albeit a small share of it in comparison to God. Animals haven’t been blessed in the same way. According to Descartes, theres nothing that “puts weak minds at a great distance from the straight path of virtue than to imagine that the soul of beasts is of the same nature as ours, and that, as a consequence, we have nothing to fear or to hope for after this life any more than do flies and ants.” We might look at him and think, oh how Baroque! But I’m not sure how distant we are from his view. I frequently read news headlines saying things like, “It turns out, magpies are intelligent. They are capable of counting the days of the week!” or, “dolphins have been named 2nd most intelligent animal, because they use their fins like opposable thumbs.” These are fake headlines, of course. But what I see in media as regards animal intelligence still has this attitude that intelligence means doing human-like things. Its not so much that man is the measure of all things, but all things are measured against man.
What’s even more distasteful to my modern ears is the thought that since animals lack reason, they are like clocks, just well wound machines that carry about a mechanical function without any form of an inner life. I think this position is forced by Descartes broader philosophical framework. Famously, Descartes notices “I think, I am,” (I leave out the therefore, since his later work the Meditations leaves it out too. It isn’t a cognition so much as a pre rational recognition of an inner life.) There is no suggestion that animals have an inner life. You might be thinking, well can I we do any better as regards to other people? How can I say, you think, you are? Descartes never addresses this in his metaphysical work, which suggests that he has a difficulty strictly trusting that other people exist in the same way that he does. He cannot say it on strictly rational grounds, and I wonder if any of us can. But Descartes, and hopefully most of us, operate as though other people have inner lives in the same way that we do. Its not so much because there is a rationale behind it, but because the consequences of ignoring this would be socially annihilative.
We 21st century people are quite different from Descartes. We are far more used to our tools being able to do intelligent things: roombas, driverless cars, autocorrect and about a million other things. Descartes thought that intelligence was the proof of an internal life, of consciousness. I don’t think this was a very successful definition even for his time, because he couldn’t strictly reason to the existence of other people’s internal lifes, and had to admit that another human could just be a seemingly intelligent machine. The only proof that other people weren’t machines was that humans are capable of language. But now that we use LLMs in our daily life, we don’t have the option of using language capacity as the proof of consciousness. That is, unless we think that our LLMs are awake. I feel dirty even mentioning this question of consciousness. The question of AI sentience has been unhelpful for the development of safe AI systems. The locust of public concern has long been on an AI “waking up” and choosing to rebel against its human masters. While this is a distinctly troubling possibility in the moral sense, it overshadows more pressing concerns. It turns out, an AI agent doesn’t need to have an internal life and bad intent to become dangerous. It is enough that the reward function (a coded measure of the AI’s goal, that indicates what strategies are useful and which are harmful) is misaligned with our objectives, add a bit of power and agency, and we have a problem.
