Monday, 30 October 2017

Robots, The Turing Test, Defining Life, & All That Jazz

With the rapidly increasing technological capacity of artificial intelligence, there are all sorts of thoughts and opinions doing the rounds about robot intelligence, and whether robots will ever have intelligence on a level with humans, and whether they can ever have feelings that are qualitatively comparable to the feelings we experience.

There is a sense in which things like technology are combinatorial; that is, where new technologies arise from existing ones - the constructive aspect of technology and human endeavour is an example of creating systems out of other aspects of the self-same systems. The term for this is called 'autopoietic' (it's from the Greek, it means 'self-creating'). Of course, these systems don’t create themselves on their own - they require the agency of human minds, a little like how coral reef creates itself from itself with the agency of small organisms.

Now when one thinks of nature and the huge potential with which she endows us, we see a similar situation occurring - only in the grand scheme of things the mind is a stupendously complex thing doing the creating: it builds on its neurological composition, on its innate cognition, on experiences, on receiving information, on analyticity, on processing, and it even has facilities (specifically, the dopamine neurons in our angular cingulated cortex) through which we can become aware of patterns and formations which, when conflated with memory and experience, alert us to deviations which help with perceptions like 'trust' and 'reliability'.

Some biologists define life in a similar way - as an autopoietic system - that is, one whose constituent products sustain the structural integrity in order to propagate its genetic material. But that doesn't really tell us very much, because it fails to zoom in on what is a whole ladder of vast complexity regarding sentience and consciousness.

We have often asked whether other animals have consciousness like us, and whether a machine could ever think in the same way humans think (in this blog post here I elaborate on why I think it's impossible to replicate the human brain). But here’s the fundamental problem; the trouble with consciousness is that it is so hard to define that it is all but useless to claim such a thing to be existent in something that isn't human.
In Language, Truth and Logic, Ayer gave his own version of the Turing Test by considering a man and machine. To determine whether a machine is conscious it would have to pass tests that look for the presence of the same kind of consciousness that humans exhibit.

In the past I've volunteered to interact with one or two Turing-type programs in which participants were able to engage with a 'mind' that responds to your typing, and you it, in order to see if volunteers can distinguish between a computer pretending to be a person, and an actual person on the other end of a keyboard. My conclusion was that it was quite evident that it is easy to tell a computer with its rule that transforms my comments into a reciprocated sentence. 

Just like Searle's Chinese Room experiment, a program that simulates the ability to understand Chinese is not proof that the machine understands the meaning of what it returns - it can simply be programmed to respond by use of correct symbols. The qualitative difference between this kind of rule-based response and a genuine human is immense.

Incidentally, from feedback I received, I was one of only three people to successfully *out* the computer as being non-human. I am not sure how the other two volunteers managed it, but my method was to try to use a linguistic trick that I conjectured a human would comprehend that a robot unfamiliar with subtle human nuances may not. So I asked the computer to give a view on a little syllogism I created:

Nothing is better than eternal paradise
A weekend in Great Yarmouth is better than nothing
Therefore a weekend in Great Yarmouth is better than eternal paradise

You can see what I did there with the wordplay on 'nothing'. I figured that a human would get it but a robot would not, and that turned out to be the case as I exposed the program doing the typing as being non-human.

My gut feeling is that one of the major hurdles in our ever being able to declare a robot as being alive in any meaningful sense is that the term 'life' is either subjectively definitional as a human construct, or it is far too complex to be properly defined (sometimes both).

Consider the question of whether a virus is classed as a living organism. Definitions of this kind are somewhat imprecise simply because we are relying on our own definitions of what is living and what is not. When dealing with such issues, we mustn’t ever get caught in the trap and forget that the things we are defining are done so within the definition of the languages employed, and that definitions become less fuzzy as we become more precise about where to draw the line. The chances are the question of whether a virus is a living organism or not will not apply to the science of thirty or forty years henceforward.

The notion of things like species, genetic similarity, and even in many cases ‘life’ itself, are humanly constructed notions that help us classify and categorise what we have discovered. The reason people disagree so much on abortion issues is because they can't agree on what constitutes life - it is not because they disagree on whether murder is bad. Some people argue that life begins at the point of conception because they consider life to be 'information', and they maintain that the information to create a human is already present upon fertilisation. In that case, if we copied the information onto a CD (it would fit) I assume they must then believe that the CD is alive. Further, if a cell is life, does that mean a virus is alive? Such issues show how people get into epistemic difficulties.

A virus has the genetic information necessary for its own growth and propagation, but it requires the machinery from a host cell to do so. Thus if we define ‘life’ as autonomous growth and reproduction, then by that definition a virus is not truly alive: a virus is acting in nature’s physical laws but it is not answerable to human descriptions – it is humans that have defined living organisms as being able to adapt to their surroundings, and being able to achieve homeostasis, and being able to identify with proteins, and having a characteristic genetic code, and having the ability to reproduce. 

Viruses do fit some of the criteria; they do have genetic material and they have both living and nonliving characteristics, but as we’ve already said, they do not survive without the metabolic machinery of host cells for survival and propagation.

That was the definitional ambiguities, but now to complicate matters further, consider a thought experiment I wrote a while back, which tries to cover the second part of the equation - the vast complexity of life:

Think of the notion of removing atoms one by one in the physical world, and imagine we have a method of physically doing so with ultimate computational precision and high speed capacity. If I reduce bit by bit a plane or a car or a microwave to a random aggregation of atoms and then reassemble them exactly as they were, then I would have a fully working plane or a car or a microwave, because neither of these systems is biologically alive. But if I did the same to an insect, a bird or a human (at several trillion atoms at a time), there would come a point when its being 'alive' would cease. 

If I reassembled those atoms exactly as they were I would never reconstitute life, because once a thing dies it cannot be brought back to life. At least that is our current understanding of biological systems. But do we believe this only because of our limitations in reassembling the atomic or sub-atomic structure back to full constitution?

In other words, if, when a young bird died by hitting a tree, I had the apparatus to reassemble its structure into the exact atomic form before it flew into the tree, would it be alive as it was before? I think the idea of life as being explicable in terms of matter, information and computation is interesting, because it leads to the question of whether it can be reconstituted with the ability to reassemble matter, or whether there is some law in nature that would preclude this.

To know if the bird's life can be reconstituted after death with the right retaining of atomic structure we have to know what life is. There are practical problems with this. In the first place, we are alive, but we cannot step out of this state of being alive to measure its true complexity, and we can't therefore look back in on this perspective and know whether our judgements of the relations between life's constituent parts are accurate. 

Yes, on the face of it, we know the difference between a live mouse and a dead one, but we don't know if the complex internal arrangements of substances that make up the mouse's state of being alive can be brought back once that state has ceased, or whether there are barriers unknown to us.

Our definition of ‘life’ seems to me to be far too simple to capture all the goings on. What my thought experiment indicates is two things; 1) definitions of life probably are arbitrary and humanly constructed to remain consistent with the utility of definition. And 2) the physical universe probably conceals enough complexity to render those definitions nigh on impossible to ascertain once we begin the componential process.

Consequently, then, if we do ever get to the stage where some people think they can define life sufficiently to ascribe those properties to the cognition of artificial intelligence, you can bet your bottom dollar that there'll be all sorts of quarrels and protest groups much like there are now in the abortion debates, the genetically modified food debates and the cloning debates.

Further reading -----    Why Robots Won't Make Most Of Us Unemployed

                                    Apocalyptic Chickens Coming Home To Roost