27 June 2014

Why Artificial Intelligences Will Never Be Like Us and Aliens Will Be Just Like Us.

"Yet across the gulf of space, minds that are to our minds as ours are to those of the beasts that perish, intellects vast and cool and unsym-pathetic, regarded this earth with envious eyes, and slowly and surely drew their plans against us."

cosmicorigins.com
Artificial Intelligence (AI) is one of the great memes of science fiction and as our lives come to resemble scifi stories ever more, we can't help by speculate what an AI will be like. Hollywood aside seem to imagine that AIs will be more or less like us because we aim to make them like us. And as part of that we will make them with affection for, or at least obedience to us. Asimov's Laws of Robotics are the most well known expression of this. And even if they end up turning against us, it will be for understandable reasons.

Extra-terrestrial aliens on the other hand will be incomprehensible. "It's like Jim, but not at we know it." We're not even sure that we'll recognise alien life when we see it. Not even sure that we have a definition of life that will cover aliens. It goes without saying that aliens will behave in unpredictable ways and will almost certainly be hostile to humanity. We won't understand them minds or bodies and we will survive only by accident (War of the Worlds, Alien) or through Promethean cunning (Footfall, Independence Day). Aliens will surprise us, baffle us, and confuse us (though hidden in this narrative is a projection of fears both rational and irrational).

In this essay I will argue that we have this backwards: in fact AI will be incomprehensible to us, while aliens will be hauntingly familiar. This essay started off as a thought experiment I was conducting about aliens and a comment on a newspaper story on AI. Since then it's become a bit more topical as a computer program known as a chatbot was trumpeted as having "passed the Turing Test for the first time". This turned out to be a rather inflated version of events. In reality a chatbot largely failed to convince the majority of people that it was a person despite a minor cheat that lowered the bar. The chatbot was presented as a foreigner with poor English and was still mostly unconvincing.

But here's the thing. Why do we expect AI to be able to imitate a human being? What points of reference would a computer program ever have to enable it to do so?


Robots Will Never Be Like Us.

There are some fundamental errors in the way that AI people think about intelligence that will begin to put limits on their progress if they haven't already. The main one being that they don't see that human consciousness is embodied. Current AI models tacitly subscribe to a strong form of Cartesian mind/body dualism: they believe that they can create a mind without a body. There's now a good deal of research to show that our minds are not separable from our bodies. I've probably cited four names more than any other when considering consciousness: George Lakoff, Mark Johnson, Antonio Damasio, and Thomas Metzinger. What these thinkers collectively show is that our minds are very much tied to our bodies. Our abstract thoughts are voiced using on metaphors drawn from how we physically interact with the world. Their way of understanding consciousness posits the modelling of our physical states as the basis for simple consciousness. How does a disembodied mind do that? We can only suppose that it cannot.

One may argue that a robot body is like a human body. And that an embodied robot might be able to build a mind that is like ours through it's robot body. But the robot is not using it's brain primarily to sustain homoeostasis mainly because it does not rely on homoeostasis for continued existence. But even other mammals don't have minds like ours. Because of shared evolutionary history we might share some basic physiological responses to gross stimuli that are good adaptations for survival, but their thoughts are very different because their bodies and particularly their sensory apparatus are different. An arboreal creature is just not going to structure their world the way a plains dweller or an aquatic animal does. Is there any reason to suppose that a dolphin constructs the same kind of world as we do? And if not then what about a mind with no body at all? Maybe we could communicate with dolphin with difficulty and a great deal of imagination on out part. But with a machine? It will be "Shaka, when the walls fell." For the uninitiated this is a reference to a classic of first-contact scifi story. The aliens in question communicate in metaphors drawn exclusively from their own mythology, making them incomprehensible to outsiders, except Picard and his crew of course (there is a long, very nerdy article about this on The Atlantic Website). Compare Dan Everett's story of learning to communicate with the Pirahã people of Amazonia in his book Don't Sleep There Are Snakes.

Although Alan Turing was a mathematical genius he was not a genius of psychology. And he made a fundamental error in his Turing Test in my opinion. Our Theory of Mind is tuned to assume that other minds are like ours. If we can conceive any kind of mind independent of us, then we assume that it is like us. This has survival value, but it also means we invent anthropomorphic gods, for example. A machine mind is not going to be at all like us, but that doesn't stop us unconsciously projecting human qualities onto it. Hypersensitive Agency Detection (as described by Justin L Barrett) is likely to mean that even if a machine does pass the Turing Test then we will have over estimated the extent to which it is an agent.

The Turing Test is thus a flawed model for evaluating another mind because of limitations in our equipment for assessing other minds. The Turing Test assumes that all humans are good judges of intelligence, but we aren't. We are the beings who see faces everywhere, and can get caught up in the lives of soap opera characters and treat rain clouds as intentional agents. We are the people who already suspect that GIGO computers have minds of their own because they breakdown in incomprehensible ways at inconvenient times and that looks like agency to us! (Is there a good time for a computer to break?). The fact that any inanimate object can seem like an intentional agent to us, disqualifies us as judges of the Turing Test.

AI's, even those with robot bodies, will sense themselves and the world in ways that will always fundamentally different to us. We learn about cause and effect from the experience of bringing our limbs under conscious control, by grabbing and pushing objects. We learn about the physical parameters of our universe the same way. Will a robot really understand in the same way? Even if we set them up to learn heuristically through electronic senses and a computer simulation of a brain, they will learn about the world in a way that is entirely different to the way we learned about it. They will never experience the world as we do. AIs will always be alien to us.


All life on the planet is the product of 3.5 billion years of evolution. Good luck simulating that in a way that is not detectable as a simulation. At present we can't even convincingly simulate a single celled organism. Life is incredibly complex as this 1:1 million scale model of a synapse (right) demonstrates.


Aliens Will Be Just Like Us.

Scifi stories like to make aliens as alien as possible, usually by making them irrational and unpredictable (though this is usually underlain by a more comprehensible premise - see below).

In fact we live in a universe with limitations: 96 naturally occurring elements, with predictable chemistry; four fundamental forces; and so on. Yes, there might we weird quantum stuff going on, but in bodies made of septillions (1023) of atoms we'd never know about it without incredibly sophisticated technology. On the human scale we live in a more or less Newtonian universe.

Life as we know it involves exploiting energy gradients and using chemical reactions to move stuff where it wouldn't go on its own. While the gaps in our knowledge still technically allow for vitalistic readings of nature, it does remove the limitations imposed on life by chemistry: elements have strictly limited behaviour the basics of which can be studied and understood in a few years. It takes a few more years to understand all the ways that chemistry can be exploited, and we'll never exhausted all of the possibilities of combining atoms in novel ways. But the possibilities are comprehensible and new combinations have predictable behaviour. Many new drugs are now modelled on computers as a first step.

So the materials and tools available to solve problems, and in fact most of the problems themselves, are the same everywhere in the universe. A spaceship is likely to be made of metals. Ceramics is another option, but they require even higher temperatures to produced and tend to be brittle. Ceramics sophisticated enough to do the job suggest a sophisticated metal-working culture in the background. Metal technology is so much easier to develop. Iron is one of the most versatile and abundant metals: other mid-periodic table metallic elements (aluminium, titanium, vanadium, chromium, cobalt, nickel, copper, zinc, etc) make a huge variety of chemical combinations, but for pure metal and useful alloys, iron is king. Iron alloys give the combination of chemical stability, strength to weight ratio, ductility, and melting point to make a space ship. So our aliens are most likely going to come from a planet with abundant metals, probably iron, and their space ship is going to make extensive use of metals. The metals aliens use will be completely pervious to our analytical techniques.

Now in the early stages of working iron one needs a fairly robust body: one has to work a bellows, wield tongs and hammer, and generally be pretty strong. That puts a lower limit on the kind of body that an alien will have, though strength of gravity on the alien planet will vary this parameter. Very gracile or very small aliens probably wouldn't make it into space because they could not have got through the blacksmithing phase to more sophisticated metal working techniques. A metal working culture also means an ability to work together over long periods of time for quite abstract goals like the creation of alloys composed of metals extracted from ores buried in the ground. Thus our aliens will be social animals by necessity. Simple herd animals lack the kind of initiative that it takes to develop tools, so they won't be as social as cows or horses. Too little social organisation and the complex tasks of mining and smelting enough metal would be impossible. So no solitary predators in space either.

The big problem with any budding space program is getting off the ground. Gravity and the possibilities of converting energy put more practical limitations on the possibilities. Since chemical reactions are going to be the main source of energy and these are fixed, gravity will be the limiting factor. The mass of the payload has to be not too large to be to costly or just too heavy, and it must be large enough to fit a being in (a being at least the size of a blacksmith). If the gravity of a n alien planet was much higher than ours it would make getting into space impractical - advanced technology might theoretically overcome this, but with technology one usually works through stages. No early stage means no later stages. If the gravity of a planet was much lower than ours then the density would make large concentrations of metals unlikely. It would be easier to get into space, but without the materials available to make it possible and sustainable. Also the planet would struggle to hold enough atmosphere to make it long-term liveable (like Mars). So alien visitors are going to come from a planet similar to ours and will have solved similar engineering problems with similar materials.

Scifi writers and enthusiasts have imagined all kinds of other possibilities. Silicon creatures were a favourite for a while. Silicon (Si) sits immediately below carbon in the periodic table and has similar chemistry: it forms molecules with a similar fourfold symmetry. I've made the silicon analogue (SiH4) of methane (CH4) in a lab: it's highly unstable and burns quickly in the presence of oxygen or any other moderately strong oxidising agent (and such agents are pretty common). The potential for life using chemical reactions in a silicon substrate is many orders of magnitude less flexible than that based on carbon and would of necessity require the absolute elimination of oxygen and other oxidising agents from the chemical environment. Silicon tends to oxidise to silicon-dioxide SiO2 and then become extremely inert. Breaking down silicon-dioxide requires heating to melting point (2,300°C) in the presence of a powerful reducing agent, like pure carbon. In fact silicon-dioxide, or silica, is one of the most common substances on earth partly because silicon and oxygen themselves are so common. The ratio of these two is related to the fusion processes that precede a supernova and again are dictated by physics. Where there is silicon, there will be oxygen in large amounts and they will form sand, not bugs. CO2 is also quite inert, but does undergo chemical reactions, which is lucky for us as plants rely on this to create sugars and oxygen.

One of the other main memes is beings of "pure energy", which are of course beings of pure fantasy. Again we have the Cartesian idea of disembodied consciousness at play. Just because we can imagine it, does not make it possible. But even if we accept that the term "pure energy" is meaningful, the problem is entropy. It is the large scale chemical structures of living organisms that prevent the energy held in the system from dissipating out into the universe. The structures of living things, particularly cells, hold matter and energy together against the demands of the laws of thermodynamics. That's partly what makes life interesting. "Pure energy" is free to dissipate and thus could not form the structures that make life interesting.

When NASA scientists were trying to design experiments to detect life on Mars for the Viking mission, they invited James Lovelock to advise them. He realised that one didn't even need to leave home. All one needed to so was measure the composition of gases in a planet's atmosphere, which one could do with a telescope and a spectrometer. If life is going to be recognisable, then it will do what it does here on earth: shift the composition of gases away from the thermodynamic and chemical equilibrium. In our case the levels of atmospheric oxygen require constant replenishment to stay so high. It's a dead give away! And the atmosphere of Mars is at thermal and chemical equilibrium. Nothing is perturbing it from below. Of course NASA went to Mars anyway, and went back, hoping to find vestigial life or fossilised signs of life that had died out. But the atmosphere tells us everything we need to know.

So where are all the aliens visitors? (This question is known as the Fermi Paradox after the Enrico Fermi who first asked it). Recall that as far as we know the limit of the speed of light invariably applies to macro objects like spacecraft - yes, theoretically, tachyons are possible, but you can't build a spacecraft out of them! Recently some physicists have been exploring an idea that would allow us to warp space and travel faster than light, but it involves "exotic" matter than no one has ever seen and is unlikely to exist. Aliens are going to have to travel at sub-light speeds. And this would take subjective decades. And because of Relativity time passes slower on a fast moving object, centuries would pass on their home planet. Physics is a harsh mistress.

These are some of the limitations that have occurred to me. There are others. What this points to are a very limited set of circumstances in which an alien species could take to space and come to visit us. The more likely an alien is to get into space, the more like us they are likely to be. The universality of physics and the similarity of the problems that need solving would inevitably lead to parallelism in evolution, just as it has done on earth.


Who is More Like Us?

Unlike scifi, the technology that allows us to meet aliens will be strictly limited by physics. There will be no magic action at a distance on the macro scale (though, yes, individual subatomic particles can subvert this); there will be no time travel, no faster than light travel; no materials impervious to analysis; no cloaking devices, no matter transporters, and no handheld disintegrators. Getting into space involves a set of problems that are common to any being on any planet that will support life, and there are a limited set of solutions to those problems. Any being that evolves to be capable of solving those problems will be somewhat familiar to us. Aliens will mostly be comprehensible and recognisable, and do things on more or less the same scale that we do. As boring as that sounds, or perhaps as frightening depending on your view of humanity.

And AI will forever be a simulation that might seem like us superficially, but won't be anything like us fundamentally. When we imagine that machine intelligences will be like us, we are telling the Pinocchio story (and believing it). This tells us more about our own minds, than it does about the minds of our creations. If only we would realise that we're looking in a mirror and not through a window. All these budding creators of disembodied consciousness ought to read Frankenstein; or, The Modern Prometheus by Mary Shelly. Of course many other dystopic or even apocalyptic stories have been created around this theme, some of my favourite science fiction movies revolve around what goes wrong when machines become sentient. But Shelly set the standard before computers were even conceived of; even before Charles Babbage invented his Difference Engine. She grasped many of the essential problems involved in creating life and in dealing with otherness (she was arguably a lot more insightful than her ne'er-do-well husband).

Lurking in the background of the story of AI is always some version of Vitalism: the idea that matter is animated by some élan vital which exists apart from it; mind apart from body; spirit as opposed to matter. This is the dualism that haunts virtually everyone I know. And we seem to believe that if we manage to inject this vital spirit into a machine that the substrate will be inconsequential, that matter itself is of no consequence (which is why silicon might look viable despite it's extremely limited chemistry; or a computer might seem a viable place for consciousness to exist). It is the spirit that makes all the difference. AI researchers are effectively saying that they can simulate the presence of spirit in matter with no reference to the body's role in our living being. And this is bunk. It's not simply a matter of animating dead matter, because matter is not dead in the way that Vitalists think it is; and nor is life consistent with spirit in the way they think it is.

The fact that such Vitalist myths and Cartesian Duality still haunt modern attempts at knowledge gathering (and AI is nothing if not modern) let alone modern religions, suggests that the need for an ongoing critique. And it means there is still a role for philosophers in society despite what Stephen Hawking and some scientists say (see also Sean Carroll's essay "Physicists Should Stop Saying Silly Things about Philosophy"). If we can fall into such elementary fallacies at the high-end of science then scientists ought to be employing philosophers on their teams to dig out their unspoken assumptions and expose their fallacious thinking.

~~oOo~~
Related Posts with Thumbnails