Thinking For Itself

How the film was created. This edition published by New English Library/Sidgwick & Jackson Ltd 1980. Cover illustration by Bruce Pennington.

As space probes and rovers on the surfaces of other worlds are deployed further and further from Earth, they will need to become increasingly autonomous.  The time delay in communication imposed by that irritating restriction imposed by the speed of light means that operation from the home planet in real time will become ever more difficult.  Controlling a rover or an aerial drone on Titan, the largest moon of Saturn, with a delay of well over an hour each way, could be a problem.

The need for autonomous systems, leading to artificial intelligence which can effectively think for itself and react to situations as they develop, has been appreciated for some time.  The next logical step is for such advanced and (hopefully) reliable systems to be in charge even when humans are along for the ride.  Probably the most famous such intelligence in the realms of science fiction is the all-controlling computer HAL in Arthur C Clarke’s 1968 film and novel 2001: A Space Odyssey.

But once AI develops to this level, it can have a full personality of its own, along with all the foibles that come with it.  In his 1972 book The Lost Worlds of 2001, Clarke takes us through the stages of the film’s creation and how the plot was constructed, including some insights into HAL’s genesis.  He admits that the movie “has often been criticised as lacking human interest, and having no real characters – except HAL.”  And there’s a key point – the AI that was HAL had to be a strong character in order to establish a major element of the plot, but any sufficiently advanced AI, given the role of controlling the humans with which it is associated, will end up stronger than them in many ways.

We are already well on the way to building autonomous systems, primarily at present in terms of the driverless vehicles that are appearing on our roads.  An assessment within the automotive industry, When Will We Have Fully Autonomous Cars?, suggests that we will not see what is referred to as Level 5 automation, requiring no human help whatsoever, for a while – at least twelve years – but there is no doubt that it will become a reality.  And then the automated systems will be in control.

In a previous Odyssey post, Driven to Affection, I pointed out some of the potential disadvantages of developing AI for driverless cars since it may end up with a character and personality of its own.  If it thinks for itself, in whatever form that might take, it must have a relationship of sorts with human beings.  Ultimately, it could end up with the sort of connection that HAL had with his colleagues on board the Discovery, which definitely had its downside.  But even if one is comfortably travelling the roads on Earth, away from the harshness of outer space, in a vehicle controlled entirely by an AI, one would want to feel assured that one had a friendly, uncontentious link with it at all times.

There’s a lot to be gained from possible interaction with an AI.  In Two minds are better than one (New Scientist, 24 August 2019), Douglas Heaven describes how machine opponents are routinely beating humans in computer games, not least because they develop entirely new strategies.  AI is evolving and we can learn from the way it adapts, and change our own perspectives – it often does know best.  There’s no problem in leaving it in complete control of the car if that’s really the case.

But there are still some problems to be ironed out.  In Fooling AI (New Scientist, 27 April 2019), Chris Baraniuk explained how worryingly easy it is (at present) to hack into an AI or to confuse it visually so it mis-classifies images.  Tampering with just a few pixels can result in serious misinterpretation for something such as facial recognition systems.  The obstacle could be even more serious, or even life-threatening, for an autonomous car, or possibly even worse again in a spacecraft where simply walking away from a vehicle controlled by a defective AI is impossible.  Developing AI for driverless cars is happening right now, but thinking through the safeguards to go with it is equally essential.

Richard Hayes, Assistant Editor (Odyssey)

With a lifetime’s interest in science, history and human behaviour, Richard Hayes focuses his writing on how the imagination has created the world in which we live, and where it may lead us in the future.  Odyssey, the e-magazine of the British Interplanetary Society, draws on the rich treasury of science fiction to explore many fields of speculation, enabling us to glimpse what might yet be, both here on Earth and out amongst the stars.

For related Odyssey posts, please click here: The Machine Starts | Defensive Adaptation

Be sociable; support the BIS!