Genitum, Non Factum
Thoughts on a Biological Turing Test
Page 1 of 3
Table of Contents
In what follows, I ask how we can distinguish between an organism and an artifact, such as a robot. The discussion is leisurely, as the question is rich.
By Fred Cummins, 2016
The Goddess Kali strikes a fierce and awe-inspiring pose. The Hindu Goddess Kali invites us to recognise ourselves in the world. This will be explored at leisure. In one of her left hands, she holds a decapitated head that bleeds into a bowl in another left hand. Around her neck, she sports a chain of shrunken heads, each identical to the fresh victim. Her skirt is made of human hands, lower arms still attached. Her right foot rests on Shiva's chest, her right hands hold a trident and a sword. Her skin is blue (or black), but the feature that always leaps out at me is her tongue, bright red and maximally protruding.
This is not a God in the sense familiar within the Abrahamic traditions (Judaism, Christianity, Islam, in all their multifarious varieties). It is not some kind of super-person, with special powers. There are more-or-less conventional interpretations of what the elements in such images stand for, though such interpretations vary greatly from one context to another. The severed head, for example, is not a specific person, but the ego, or fictitious autonomous self, caught up in the world of maya or illusion, that must be transcended. There is little agreement on the remarkable tongue though. To some it signifies bloodlust, to others, shame. [Table of Contents]
To me, though, the tongue seems to speak of something else; something akin to the famous reaction of a very young infant sticking its tongue out at an adult, in the classic study by Meltzoff and Moore (1977). Meltzoff, A. N., & Moore, M. K. (1977). Imitation of facial and manual gestures by human neonates. Science, 198(4312), 75-78. In the original study, it looked much as if the infants were veritable imitation machines. The authors documented three facial gestures and one hand gesture (pointing) that their infants seemed to imitate, even though the infants were only 2 to 3 weeks old, and thus had no knowledge of how they, themselves, looked. The results claimed seemed to speak of a very direct kind of bond between adult and infant, one based on mutual recognition or on reciprocal knowledge, and one that did not seemed to be based on learning.
A thought-provoking study like that one will attract replication and variation, and the strong claims of the original study are no longer widely vouched for. But the tongue gesture, alone among the four, seems to have survived the test of time. Very young infants will indeed stick their tongue out, Kali-like, in response to an adult doing the same. This has even been replicated in young chimpanzees Myowa, M. (1996). Imitation of facial gestures by an infant chimpanzee. Primates, 37(2), 207-213. , although they revert, as they get older, to imitating mouth opening, rather than tongue protrusion. So when we interact with them, the apes are, appropriately, a bit like us and a bit like themselves.
I want to suggest that what the infants are doing, and perhaps fancifully, what Kali is doing, is an act of self-recognition. To interpret this as self-recognition will require us to stick our own necks out a bit. That might just be dangerous. [Table of Contents]
A Biological Turing Test
The Turing Test (about intelligence) provides a model for our question (about living beings). At the beginning of the information age, as computers began to tickle the fancies of bright minds in mathematics, military science, anthropology, physics, and psychology, Alan Turing proposed a now infamous test that might be used to establish whether an artificial system could appear to a human as intelligent. It was not meant as a test to establish whether the artificial system was intelligent; just to see if it could exhibit sufficiently convincing behaviour that a human observer would not be able to clearly distinguish it from a human. In line with interface technology of the time, the test required the human to communicate with either a computer programme, or with a real human, by typing English text on a keyboard, and reading the textual responses. Based only on these text exchanges, the human observer would have to judge which partner was human, and which artificial. This may seem like an extraordinarily narrow lens through which to view allegedly intelligent behaviour, but it was very much of its time. The intellectual climate of the latter half of the 20th Century saw text as the cleanest way to represent language, and language as the hallmark of the human, and by inference, the intelligent. The interface may seem particularly impoverished by todays standards, but the test has survived as a landmark, if not as a practical goal.
Intelligence is a very poorly defined concept, and the Turing Test served to provide a kind of operational benchmark (still probably not completely passed) with which one could work to design more complex programmes. It was never intended to resolve the intractable question of what "intelligence" really is, and there was an underlying assumption that this operational test might just be as much as one could do.
Our goal here is to distinguish between biological organisms in the most general sense, and artifacts or mechanisms. I wish here to consider a related question. Rather than distinguishing between a programme and a human, though, I want to understand how we might distinguish between an artifact and a biological creature. Artificial systems these days can be richly embodied. Our robots are getting better and better at the humdrum tasks such as locomotion that biological organisms seem to find so easy, but that turned out to be really really hard to imitate. We are not yet in the future of Bladerunner, in which replicants (robots) are essentially indistinguishable from humans, and nor will we be in my lifetime. But the question I want to consider is more general, and is not restricted to humans, or any specific animal. I can perhaps best illustrate it using the following simple example, one that has taken a great deal more of my thinking-time than I supposed when I first posed it. [Table of Contents]
Here are two relatively simple systems. Top: A tumblebot. The furry weasel is optional. The action lies in the motion of the ball. Bottom: Mr Sparkles, my goldfish, 12 years old and counting. The lower is my goldfish, Mr Sparkles, who has lived 12 years in the same tank. His life seems boring, and he exhibits little in the way of intelligence or emotion. He swims around, eats, excretes, and not much else. The upper system is a toy known, in some instances at least, as a Tumblebot. The furry weasel-like animal attached to the ball by a string is optional. It is the ball itself we are concerned with. I first encountered a tumblebot (no weasel attached) in 1993 in Colorado, where I was attending a Summer School for graduate students interested in artificial neural networks. The school was run by all the greats of the field (Rummelhart, McClelland, Hinton, Elman, and many many more), and was attended by a bright bunch of students, many of whom are now themselves establishment figures in training subsequent legions of researchers. [Table of Contents]
It was on a break from class that myself and my friend, Devin, first saw a tumblebot in a downtown store. What we saw was a self-driven ball that rolled around the floor in what looked like an uncannily agentive fashion, as if it was aware of its surroundings, and as if it was seeking something. It would roll under chairs and couches, roll into corners and behind bulky pieces of furniture, and just when you thought it must get stuck, it would seem to reconsider its plight, and it would find a way back out to open space in the middle of the floor again. When it came at your shoes, you could not help but think that it was sniffing your shoes, trying to find something, before moving on.
Inside the ball is a very simple mechanism. Mystery revealed: the inside of a tumblebot. There is a battery case, attached to an overbalanced motor, a central spindle, and not much more. The overbalanced contraption is responsible for the complexity of the behaviour. As the internal structure rotates about a fixed axis, it generates wobbly, inconstant motion of the ball as a whole, and those wobbles are sufficiently complex to extricate it from almost every predicament it gets itself into. It also helps that the surface of the ball has a friction-characteristic well matched to the domestic environments the toy finds use in: carpets and parquet floors.
Devin and I bought a tumblebot and returned to the class. We demonstrated the crazy motion and the convincingly goal-directed behaviour of the toy, and we claimed that this was a fancy robot we had built, whose motion was controlled by a fancy neural network algorithm. The intense searching behaviour it exhibited, we claimed, was a goal we had engineered using state-of-the-art tools from artificial intelligence. Most of the class seemed to believe our fiction, at least until the toy was opened up and its remarkably simple interior put on display. The claim of a high-tech cognitive mechanism in a hidden interiority seemed to be sufficient to fool very many very bright people, for a while at least. I am reminded of similar and common reaction to Terry Winograd's Eliza programme, a product of artificial intelligence research of the 1970's that appeared to converse with a user, through text of course, even though it was mindlessly recombining what it had just heard with some stock responses, in a relatively simple manner, one that now fools nobody.
Our prank has backfired. But our prank seems to have backfired. Now I sit and I regard the Tumblebot, and Mr Sparkles, and I pause to consider on what basis one might identify one as an artifact, and the other as a creature. The very word "creature" seems to imply a creator, but even within devout Christian traditions, and to Christian philosophers such as Descartes and Kant, the bodies of animals and people have long been regarded as a special kind of machine. If animal bodies are, in fact, machines, then there might be no means, in principle, by which we could tell artifact from organism. The question I want to dwell on is this: are we stuck with a Turing-like test of behavioural adequacy, to distinguish the living from the merely mechanical, or is there in principle something that can tell them apart? [Table of Contents]
To be perfectly clear, the question is whether there is, in principle, any way to tell an organism from a machine. We can make this more concrete by imagining we encountered something that was so weird that we are not sure if it was a space ship or not, but it jettisoned something, which we can now examine. Is it an artifact from an unimaginable technology, designed by others, or is it a living being, far from home. Ideally, it should be alive, or active, when we view it, but there is a problem here: we have stripped it of all context. Unless we know it's "natural" environment (whatever that means in this strained context), we are in a poor position to assess its activity. Dolphins don't look real smart in the forest.
The question is not a trivial one. And just to make sure this quest is worth the entrance price, we might note in passing that we are in the same territory as a classic argument for or against the existence of God: the Argument from Design, and there is a scientific consensus that living beings are not externally designed, but evolved through a process of natural selection, given means of (1) reproduction, (2) variation, and (3) competition. We will return to this point.
Of Goals, Functions, and the Sciences
Science is not just physics. A rather old-fashioned way of looking at the sciences takes physics to be the gold standard. Scientists talk with tongue firmly in cheek of physics-envy within all disciplines that are not physics. The physicist Ernest Rutherford audaciously claimed that all science is either physics or stamp collecting. The term "physical" is used to describe something which is indubitably real, in the fond hope that there is such a class of things. In everyday language, "physical" is frequently a synonym for "real".
On this outdated and indefensible view, chemistry is just a branch of physics that concerns itself with specific phenomena—the atoms and molecules that make up matter. There is nothing within the domain of chemistry that could not be deduced, in principle, from basic physical laws, for basic physical laws govern what is real (or what really "exists", if you prefer). Biology, then, sits atop of chemistry: biology concerns itself with a very specialised and constrained subset of chemical and material processes, those found in living beings. In theory at least, biological facts then reduce to chemistry which in turn reduces to physics. And the softer, more problematic "sciences" of psychology, sociology, economics, geography, all, in turn reduce to physics, at least to the extent that they are ascertaining matters of fact.
Emergence and levels of explanation. Now happily almost nobody argues for this extreme reductionist view of the sciences these days. There are many reasons why. One of them is the widespread recognition that different phenomena require unpacking within different courts, admitting of different kinds of evidence. Let us first stay within the domain of facts construed as "physical". It is commonly accepted that there are macroscopic phenomena that cannot be found by examination of the components that give rise to them. The sphericity of a soap bubble, for example, is an emergent property of soap and water that demands being taken at its own level. No study of individual molecules can plausibly reveal the phenomenon of a bubble. Arguing along these lines, we might indeed continue to encounter new phenomena that demand that we take them seriously at their own level: bubbles at the scale of bubbles, not molecules; traffic jams at the scale of traffic, not individual cars; bodies at the scale of bodies, not cells or atoms. This, emergentist, pluralistic approach is perhaps more representative of most views of the relationship between biology and physics; non-reductive, each with domain-specific characteristics, and each making claims only within its own domain. A kind of bounded rationality, if you will.
Different kinds of explanation are required in the realm of the living. But that picture will not do for the present discussion. While it admirably leaves phenomena their own domains—biological phenomena find description in biological terms, atomic phenomena in terms of atoms—it leaves the kind of explaining done within each domain untouched. It suggests that the way in which we establish matters of fact when considering chemical reactions or cometary orbits is the same as the way in which we establish matters of fact when considering organs of the body, human behaviour, the role of monetary policy on unemployment, or the character of lived experience. A single kind of "truth" obtains, irrespective of the domain, and science sees itself as the means by which facts, or truths, are established. A central plank of this belief system is the dogma that values cannot be included in the discussion: no "ought" from and "is", as Hume claimed.
I am forced to argue that this singular approach to truth can not be sustained, because when we deal with the goings on of the living, value and goals leak in, and that transforms the manner in which we need to frame our observations and descriptions. I am by no means alone in this claim, and it is in no way an anti-scientific claim, but it may be seen as such by some. [Table of Contents]
One way to frame the discussion is to point out that talk of "function" is never needed when discussing the goings on of the inanimate, but becomes necessary when developing a rational account of living beings. Function implies purpose or goals—teleology—and explanation in terms of goals is something that the scientific enterprise spent a long time getting away from. Before the birth of modern science, causal accounts developed in the Aristotelian tradition appealed in an undisciplined manner to goals and purposes. Final cause, which casts end states or goals as a causative agent, was the reason a seed grew into a plant, a rock fell to earth, and a knife cut leather.
The scientific picture that grew from the work of Galileo and later Newton banished such quaint notion of inherent purposes in everything. It described a mechanical universe in which efficient cause, or cause-and-effect as captured by laws written in the form of differential equations, came to be the prime mover. No purposes anywhere, just the lawful force-driven exchanges among moving bodies. Along with the banishment of inherent purposes, something else vanished from this cosmological view: Agency. Nobody, in such a mechanical world, does anything.
But you can't keep purposes out that easily. In the realm of biology, talk of purpose and function never went away. Perhaps the clearest example is that of the heart, which we all conventionally understand as having the function of pumping blood around the body. One way to understand this is to say that any talk of "function" of this kind is actually shorthand for a more benign (less purposeful) extended process of evolution by natural selection, the blind mindless chain of events by which life forms gradually became more complex and adaptiveness came to decide which structures and processes got passed to the next generation and which did not.
There is something terribly unsatisfactory with this picture. If talk of function is actually shorthand for the meaningless processes of evolution by natural selection, we lose the ability to speak with confidence of any functions. Cardiologists would have a hard time explaining what they are about if we did not share their background assumption that hearts serve a specific purpose, and that they can fail as well as succeed. Failure is not an option for a copper molecule or for a cloud, but it is for an organ, or an ant within a colony, or a germinating seed. In each case, there is a perspective that we must acknowledge, that of the system that may or may not succeed in what it is about. This perspectivalism is very important. It lies at the heart of the living.
In consideration of the living, we simply cannot keep teleology out! In banishing agency and purpose, the mechanical cosmos also banished the singular perspective of the living. This perspectivalism underlies that which is conventionally called mind, consciousness, lived experience, anima, dasein, or the soul. There has been a great deal of work expended in the meantime to try to further our understanding of a mechanical universe to the point where the reality of the present, the here and now that attends singular minded existence, can be accommodated. The orthodox approach of late has been to suggest that "consciousness" is something that arises as a result of the activity of brains, or more specifically, the electrical activity of nerve cells. But many find this unsatisfactory. To discuss "consciousness" as if it were something to be found in the world, is to effect a bit of sleight of hand. It does not actually address the singularity of lived experience in first person mode. It ignores it. On the received view the "world" which we seem to agree exists is entirely transcendental, known only by inference and induction. The world discussed by a science that prides itself on objectivity is known only in an indirect manner, quite unlike the direct unmistakable character of lived experience, which must assert no lesser claim to reality or being. All the while, the singular perspective of the living subject is annihilated by rebranding it as "consciousness", and treating it as if it were a substance exuded by brains.