I'm a bit of a fan of Ray Kurzweil, and had been planning to get his new book How to Create a Mind. I've since discovered a full synopsis on a website, which saves me some time and money. (For non-fiction authors the existence of such a site should be a worry - but that's another story.)
Kurzweil is that interesting type of thinker who drives scientific materialism to seemingly absurd limits. His idea is that having completely understood how humans are put together we can remake ourselves and overcome the imperfections that biological evolution saddled us with.
Kurzweil's transhumanist thinking sound kooky, even to many of his peers. But he's a credible figure, having made major contributions to artificial intelligence, notably in the field of voice recognition. I think Kurzweil's vision of the technological singularity, that machines will outstrip human intelligence by mid-century, is rather cool. I completely disbelieve it, for what I consider are sound empirical as well as practical reasons, so I'm not threatened by it, or feel the need to heckle (he gets plenty of flak as it is from other AI thinkers like Daniel Dennett). But I love it that science can let the imagination rip - it would be dull if we weren't allowed to dream.
I was curious to know, though, how does one create a mind? It's about understanding how it works. Kurzweil follows the well-trodden computational route, where lots of lower-level processes based on responses to the environment combine to produce higher level abstract thinking. He identifies the basic process as that of pattern recognition and thinks the underlying architecture is relatively simple, the kind of thing that could be readily replicated by a sufficiently powerful computer - by 2029, at the present exponential rate of progress.
He concedes that a supercomputer like IBM's Watson, currently the most advanced of its kind, could not yet pass the Turing Test, fooling a human interlocutor into thinking it is human. But that's only because it was not designed to engage in conversation but to succeed at specific tasks, like winning at chess and Jeopardy. The most advanced AI machines are already using these same principles and processes. Indeed, AI was using them even before it was discovered that the human neocortex is doing the same thing, and now the traffic has been reversed, with neuroscience feedings its discoveries back to AI.
Where reasoning is concerned, Kurzweil sees quality as an outcome of quantity. Nature endowed us with a mere 300 million pattern processors, but once we start making synthetic brains why not give them a billion, or even a trillion? This, he claims, will not only increase the kind of intelligence we already see in humans, but also generate higher orders of abstract thought and complexity. The synthetic brains could have an in-built critical thinking module that stops them holding a bunch of inconsistent ideas, as humans do. While humans are limited by evolution in terms of what we can achieve, these super intelligent, super rational machines could pursue goals like curing disease and alleviating poverty with a realistic prospect of success.
Meanwhile humans can beef up their own brainpower by adding new modules as brain implants. Lest we worry this would change our identity, Kurzweil thinks that identity is an effect of our entire system, and would not be compromised by changing individual parts.
The next step in the journey will be to spread this new intelligence throughout the universe.
If we can transcend the speed of light - admittedly a big if - for example, by using wormholes through space (which are consistent with our current understanding of physics), it could be achieved within a few centuries. Otherwise, it will take much longer. In either scenario, waking up the universe, and then intelligently deciding its fate by infusing it with our human intelligence in its non-biological form, is our destiny.
The book has stirred the pot, and it's interesting to see what sceptical perspectives have emerged. One is to point out just how incredibly distant we really are from the goal that Kurzweil thinks is just around the corner. Take the humble earthworm, C. elegans. Where the brain has 100 billion neurons connected by 100 trillion synapses, the worm has only 302 neurons connected by roughly 5,000-7,000 synapses, and a lot is known about it. Scientists can tie its reflexes and behaviours to individual neural pathways or brain circuits. But . . .
Can we use this to model or predict the actions of the worm? No. We're not even close. In fact, it takes a computer with a billion transistors to make a weak, incorrect guess at what a worm with 302 brain cells will do.
If we can't simulate 302 neurons and 5,000 synapses, how can we hope to conquer 100,000,000,000 and 100,000,000,000,000? Let's not even get started on the 100,000,000,000,000,000 electrical signals per second that form the traffic on that neural road network.
Then again, are mental phenomena really the outcome of mere pattern recognition processes? Colin McGinn, a long-time critic of mainstream thinking about mind, points out that this is in fact a very partial view of what the brain does. There's no perceptual recognition going on at all in thinking about an absent object - 'for instance in thinking about London when I am in Miami, or in dreaming, or in remembering that I have to feed the cat'. Kurzweil talks as if everything in the mind involved perception. But there are also other mental phenomena, such as
emotion, imagination, reasoning, willing, intending, calculating, silently talking to oneself, feeling pain and pleasure, itches, and moods - the full panoply of the mind. In what useful sense do all these count as "pattern recognition"? Certainly they are nothing like the perceptual cases on which Kurzweil focuses. He makes no attempt to explain how these very various mental phenomena fit his supposedly general theory of mind - and they clearly do not. So he has not shown us how to "create a mind," or come anywhere near to doing so.
McGinn also makes the point - widely noted by other critics of dominant trends in neuroscience such as Raymond Tallis - that these sorts of accounts are saturated in anthropomorphic language. Kurzweil's pattern-recognisers 'receive and send messages', they 'manipulate information'. Listening to this, it's easy to forget these are lumps of tissue: in fact they have no awareness of doing any such thing. A retort is that such representational talk is merely intended to be metaphorical. But that's not altogether true. Its effect is to create in the reader's mind a sense of a coherent process, a masking pseudo-explanation that leaves the fundamental mystery entirely untouched.
Kurzweil does include a chapter on such key matters as free will and identity, which he concedes his model can't really explain. Like many people he asserts that they somehow 'emerge' from lower level activity, just as conscious awareness does. This is problematic enough, but there's also an absence - as far as I can tell, having only read a synopsis - of any serious consideration of emotional or moral intelligence. It's not widely observed that the cleverest people are also the most emotionally mature - that is to say the wisest, the best at cooperating with groups, and most fully committed to finding the most widely acceptable outcome to any given social or political problem.
In principle I suppose one could program the machines with some utilitarian-type algorithm. However to design such a thing, humans would first have to agree on an ideal, which would be like competing groups getting together to write a national constitution - always an extremely fraught process. Kurzweil seems to think such matters can safely be left to the machines themselves, since they will be vastly more intelligent than us. This implies that there exists out there some single outcome that is so obviously rational they can all agree on it. But that is emphatically not the human experience, and what happens if the machines too start to compete with each other? This is the stuff of Hollywood dystopian nightmares, a world given over to indestructible beings with unimaginable strength and inhuman cunning.
One might say it's pure fantasy, not least because no machine could ever pass the Turing Test, as long as that test includes a demonstration of ESP - as Turing himself intended. It is course argued that ESP doesn't exist and that Turing naively let himself be seduced into accepting spurious claims. Or that even if ESP does exist, it plays such a marginal role that by meeting all the other requirements a computer could achieve human status. But if we take ESP to be a property of minds, then clearly minds are something more than the effect of brain processes.
It follows, therefore, that if some Frankenstein character were to try to use How to Create a Mind as a manual, he wouldn't get very far. Without understanding the true basis of mind we will only ever be able to mimic it. But that's not necessarily the end of the story.
Science can't create minds from dead matter, but perhaps it can create the conditions for minds spontaneously to come into being. This would mean taking seriously the idea of panpsychism, that consciousness is inherent in matter, and will emerge in systems of sufficient complexity. In this way, something of Kurzweil's vision could come about by quite different means. Or imagine that these thinking robots he anticipates, although not properly human, are sufficiently sophisticated that they can be possessed by the dead as a means to revisit the world in mechanical bodies.
That's why I like this transhumanism stuff. We don't have to take it literally, but it can take us to some interesting places.