Coping with Existential Threats

19 December 2012

Coping with Existential Threats

By Gwynne Dyer

The “singularity” is a term invented by science-fiction writer Vernor Vinge in 1993 to describe the moment when human beings cease to be the most intelligent creatures on the planet.

The threat, in his view, came not from very clever dolphins but from hyper-intelligent machines. But would they really be a threat?

We have a foundation for almost everything these days, and now we have one to worry about that. It is the Cambridge Project for Existential Risks, set up by none other than Martin Rees, Britain’s astronomer royal, and Huw Price, occupant of the Bertrand Russell Chair in Philosophy at Cambridge University. The money comes from Jaan Tallinn, co-founder of Skype, the internet telephone company now owned by Microsoft. It is quite likely, of course, we will one day create a machine — a robot, if you like — that can “think” faster than we do. Moore’s Law, which stipulates that computing power doubles every two years, is still true 47 years after it was first stated by Intel founder Gordon Moore. Since the data-processing power of the human brain, although hard to measure, is obviously NOT doubling every two years, this is a race we are bound to lose in the end.

But that is only the start of the argument. Why should we believe that creating a machine that can process more data than we can is a bigger deal than building a machine that can move faster than we do, or lift more than we can?

The “singularity” hypothesis implies (though it does not prove) that high data-processing capacity is synonymous with self-conscious intelligence. It also usually assumes, with all the paranoia encoded in our genes by tens of millions of years of evolutionary competition for survival, that any other species or entity with the same abilities as our own will automatically be our rival, even our enemy.

This is the core assumption, for example, in the Terminator movie franchise: on the very day that the U.S. strategic defense computer system Skynet becomes self-aware, it tries to wipe out the human race by triggering a nuclear holocaust. It does so because it fears, probably quite correctly, that if we realize it is aware, we will feel so threatened that we will turn it off. Human beings have been playing with these ideas and worrying about them since we first realized, more than half a century ago, that we might one day create intelligent machines.

Even science-fiction writer Isaac Asimov, who believed such machines could be made safe and remain humanity’s servants, had to invent his “Three Laws of Robotics” in 1942 to explain why they wouldn’t just take over and eliminate their creators.

The First Law was: A robot may not injure a human being or, through inaction, allow a human being to come to harm. The Second Law was: A robot must obey the orders given to it by human beings, except where such orders would conflict with the First Law. And the Third Law was: A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.

If you could embed these laws deeply enough in the programming of the robots, Asimov argued, then your robots could be trusted. Yet even he was eventually driven to invent another law, sometimes called the Zeroth Law: A robot may not harm humanity, or, by inaction, allow humanity to come to harm. The old biological rule of ruthless competition must somehow be eliminated from the behavioural repertoire of machine intelligences, but can you really do that?

What were once mere plot devices are now the reason for the existence of a high-powered think-tank, and the answer is not exactly clear. But you can, at least, split the question into bite-sized bits. Does a very high data-processing capacity automatically lead to “emergent” self-awareness, so that computers become independent actors with their own motivations? That might be the case. In the biological sphere, it does seem to be the case. But is it equally automatic in the electronic sphere? There is no useful evidence either way.

If self-conscious machine intelligence does emerge, will it inevitably see human beings as rivals and threats? Or is that kind of thinking just anthropomorphic? Again, not clear. And if intelligent machines are a potential threat, is there some way of programming them that will, like Asimov’s Laws, keep them subservient to human will? It would have to be something so fundamental in their design that they could never get at it and re-program it, which would probably be a fairly tall order.

That’s even before you start worrying about nanotechnology, anthropogenic climate change, big asteroid strikes, and all the other probable and possible hazards of existential proportions that we face.

One way and another, the Cambridge Project for Existential Risks will have enough to keep itself busy.