This is the third post in a series summarising the key arguments of Ray Kurzweil’s The Singularity is near: When humans transcend biology. The previous posts were: Ray Kurzweil, The Singularity and the accelerating pace of progress and Kurzweil predicts personal computers with the power of the human brain by 2025.
Computer hardware by itself is, of course, little use to anyone, so after arguing that we will have personal computers as powerful as the human brain by 2025 Kurzweil goes on to discuss the software that will run on this amazing platform. His contention, in a nutshell, is that we will reverse engineer the workings of the human brain and emulate the salient parts in software and that the resulting machines will be interact with us in a way indistinguishable from humans. Further, for all practical purposes, a machine whose interactions are indistinguishable from a human will need to be treated as a human and therefore regarded as conscious, and I guess alive.
It is worth dwelling on the nature of machine capable of ‘interaction that is indistinguishable from a human’, or in other words one that would pass the ‘Turing Test’, of which you might have heard. Such a machine would have to be able to get jokes, infer context, display and comprehend emotion, and exhibit every other human trait you can think of – otherwise it would be distinguishable. As Kurzweil points out several times in The Singularity, such a machine would be treated as being alive by most people. Just think how many of us impart human characteristics and personalities to complex but inanimate objects like our cars and computers.
Kurzweil’s contention is that this software will be made possible by continued exponential improvements in scanning technologies (lower power, greater resolution) which will enable us to understand exactly how the brain works. He is emboldened in this prediction by his conviction that in the next decade or so nano-scale scanners in our blood stream will be able to scan our brain at incredibly high resolution from the inside. Once we understand how the biological brain works it will be relatively simple to reproduce it in software – a task made easier by the fact that many areas of the brain won’t need to modelled – for example those dedicated to life support and areas where it is sufficient to model at the system or module level.
If you are wondering how far this technology has got today – the answer is the level of a cat brain – which IBM successfully modelled in November last year, and we are also starting to get models for regions of the human brain – e.g. Watt’s model of the auditory regions.. Additionally, many of the underlying artificial intelligence techniques have been successfully operating in narrow domains for some time – e.g. neural nets used in facial recognition, Markhov systems used in speech recognition, genetic algorithms in jet engine design.
This is a good moment to make the point that such a software will not be deterministic in the way that most computer programs are today, but will copy the emergent and chaotic systems of the brain. The outputs of these human intelligent machines will have random and unpredictable elements, just as human brains do, and they will very much be products of their environment. The growth and development of an emergent system is hugely dependent on the early environment in which it finds itself (or in other words the early input variables) – look at the way different ways ants nest develop for example. Different environments will therefore produce very different human intelligent machines. We might come to think of those differences as different personalities.
These human intelligent machines would start as blank canvases, and to do anything useful they would need nurture and education, in much the same way as a human child. One of the reasons that Kurzweil is confident that software will reach this milestone is that each of our human brains is created from a relatively small amount of information – half of the human genome is dedicated to the brain, and that comes to about 25 million bytes, when compressed. 25 million bytes could be represented in around one million lines of code. Our adult human brains are infinitely more complicated than this and would therefore be much harder to reverse engineer.
Kurzweil’s contention that a reverse engineered machine-baby-brain would grow up to have emotions and all the other aspects that characterise the adult human mind is one that many have taken issue with (including ‘twainventures’ who has left numerous helpful comments on this series of Kurzweil posts). It is also one that raises philiosophical and religious questions about what it means to be alive or to be conscious, or what it is that provides the spark of life, if anything.
My feeling is that if something (inter)acts like a human then to all intense and purposes it is a human and that questions about the the spark or life etc. are redundant, but that is pure conjecture. Until we build such a system and see how it interacts I think it will be impossible to know more.
I’m going to close this piece with the observation that once a computer achieves a human level of intelligence it will necessarily sour past it. Simply put these minds will apply their intelligence to improving themselves and will be able to re-write their own software, very quickly access and upload any information they need to make themselves smarter, and even create new more powerful hardware platforms to run themselves on.
By now you are probably picturing Terminator like nightmare scenarios of machines gone wild – a subject I will return to in the final post of this series.