Along with the millions who saw 2001 A Space Odyssey, Peter Cochrane FRSA was enthralled by HAL9000. He asks whether, three decades on, advances in artificial intelligence means we are near to producing a ‘real HAL’ .
Shortly after the release of 2001 A Space Odyssey in 1970, futurists and scientists declared that HAL9000 would be a reality within the next 30 years. Well, 2000 came and went and they were still saying it would be another 30 years. So, what went wrong and where is Hal 9000?
For sure artificial intelligence (AI) has made great strides over the past 40 years and it now powers everything form passenger aircraft to elevators, automobile engines and production lines. But the ‘intelligence component’ is always of a very limited kind and apparently not generic like ours. A big, and very fundamental, problem that we face here is the fact that we don’t understand intelligence, and we can’t describe, define or quantify it in any meaningful way. Descriptions such as “the faculty of rational behaviour” or “the ability to create order out of chaos”. turn out to be meaningless and do not add an iota to our understanding. Moreover, the IQ measure of Alfred Binet (working in the late1800s) also turn out to be useless although still widely used in education circles despite his warnings about its validity at the time.
It gets worse. Many people ascribe ‘intelligence hero status’ to those capable of feats of memory and acts of computational prowess regularly appearing on TV game shows such as Brain of Britain and University Challenge. Chess champions, poker and bridge champions are afforded a similar status. But in all cases humans now take second place to AI. Machines are now better than humans at playing chess, poker and bridge, and when it comes to feats of memory and processing data including mathematical competition they are streets ahead.
IBM Big Blue beat Gary Kasparov at chess in 1997, whilst last year IBM Watson became world champion against all comers in the US TV quiz Jeopardy. Again, there is little point in playing any more matches as human abilities have been left in the dust. So the next big challenge is creativity; can AI systems be creative? Surely this is the preserve of the human mind? No so!
The machine I am typing these words on employs chips designed by machines and constructed by AI guided robotic. NASA are now using AI systems to ‘evolve’ new designs for planetary walkers and other robotic systems, and many of the electronic and mechanical inventions of the last century are being overtaken by machine invention that sees results that we do not understand.
How can this be? As a species we are very limited and fundamentally capable of coping with three to seven variables at a time, and we can solve mathematical problems with up to five feedback (or feed forward) loops, but our machines suffer no such limitations. They cope with hundreds. So the decoding of the protein stack and the interaction with the genome will not be solved by humans using pencil and paper, or a laptop, but by AI systems that can cope with the inherent complexity. On a more prosaic level; humans are unlikely to untangle the complexity inherent in the financial system of the planet, and we are reliant on the machines to show us the way to stability; if indeed such a panacea is possible.
So, if AI systems are now so smart, where is HAL9000? He is coming fast, but like a human child he is still learning, and so are we. True intelligence relies upon cognition, context and awareness. Only recently have we built sufficient databases; comprehended the necessary rules of logic; and the frameworks of common-sense examples, that allow a machine to ‘understand’ a human utterance or situation. For the past 20 years machine speech recognition has outclassed human abilities by 2 or 3%, but there has been no understanding. For example the phrases Let us pray, Let us spray and Lettuce spray all sound the same and can only be understood in context. Are we in church, the garden or in the kitchen? We know, but a machine does not.
And so to the final components of intelligence – sensors and actuators – uniquely giving the ability to understand and influence the world in which you live. Until recently our machines had a limited awareness as they were effectively cut off from our world. But now we are giving them cameras, microphones, location-based information, chemical, magnetic, electric, vibration and force sensors. We are also giving them far more than screens and loudspeakers. Robotic arms, hands, manipulators and complete production lines are being added by the day.
We might thus anticipate that machine awareness will erupt to be far broader than our own, especially as they have the ability to network, communicate and share on a scale that is far superior to humans. But the question remains; just how intelligent will they be? We do not know for sure but we have interesting insights and evidence startling to accumulate.
It turns out that intelligence is possible without any memory or central processor. Jellyfish and slime mould are obvious examples. However, we also know that intelligence is impossible without receptors or actuators. In recent experiments with long-term comatose patients in an apparent vegetative it has been proven that they can hear but have no means of output, and so appear brain dead. As shocking as this might be, it has opened up new lines of research and an effort to understand more and perhaps affect some form of repair.
In the mean time we now understand the basis of intelligence and the fact that it grows logarithmically with processing power, memory, sensors and actuators and not exponentially as previously assumed. This is the reason HAL9000 has been a long time coming. The human model and assumptions were wrong; exponential change is fast whilst logarithmic is slow. So when, and will he be malevolent? I reckon within 15 years, and only if we teach him to be evil.
Dr Peter Cochrane (OBE, BSc, MSc, PhD, DSc, CGIA, FREng, FRSA, FIEE, FIEEE) is an entrepreneur, business and engineering advisor to international industries and governments. He currently runs his own company across 4 continents, and was formerly CTO at BT, The Collier Chair for the Public Understanding of Technology @ Bristol.