Perhaps Now Is Not The Time To Anger The Machines

HALlo.

There’s been a lot of discussion of the benefits of machines over the years, from an engineering perspective, from a social perspective and from a philosophical perspective. As we have started to hand off more and more human function, one of the nagging questions has been “At what point have we given away too much”? You don’t have to go too far to find people who will talk about their childhoods and “back in their day” when people worked with their hands or made their own entertainment or … whatever it was we used to do when life was somehow better. (Oh, and diseases ravaged the world, women couldn’t vote, gay people are imprisoned, and the infant mortality rate was comparatively enormous. But, somehow better.) There’s no doubt that there is a serious question as to what it is that we do that makes us human, if we are to be judged by our actions, but this assumes that we have to do something in order to be considered as human.

If there’s one thing I’ve learned by reading history and philosophy, it’s that humans love a subhuman to kick around. Someone to do the work that they don’t want to do. Someone who is almost human but to whom they don’t have to extend full rights. While the age of widespread slavery is over, there is still slavery in the world: for labour, for sex, for child armies. A slave doesn’t have to be respected. A slave doesn’t have to vote. A slave can, when their potential value drops far enough, be disposed of.

Sadly, we often see this behaviour in consumer matters as well. You may know it as the rather benign statement “The customer is always right”, as if paying money for a service gives you total control of something. And while most people (rightly) interpret this as “I should get what I paid for”, too many interpret this as “I should get what I want”, which starts to run over the basic rights of those people serving them. Anyone who has seen someone explode at a coffee shop and abuse someone about not providing enough sugar, or has heard of a plane having to go back to the airport because of poor macadamia service, knows what I’m talking about. When a sense of what is reasonable becomes an inflated sense of entitlement, we risk placing people into a subhuman category that we do not have to treat as we would treat ourselves.

And now there is an open letter, from the optimistically named Future of Life Institute, which recognises that developments in Artificial Intelligence are progressing apace and that there will be huge benefits but there are potential pitfalls. In part of that letter, it is stated:

We recommend expanded research aimed at ensuring that increasingly capable AI systems are robust and beneficial: our AI systems must do what we want them to do. (emphasis mine)

There is a big difference between directing research into areas of social benefit, which is almost always a good idea, and deliberately interfering with something in order to bend it to human will. Many recognisable scientific luminaries have signed this, including Elon Musk and Stephen Hawking, neither of whom are slouches in the thinking stakes. I could sign up to most of what is in this letter but I can’t agree to the clause that I quoted, because, to me, it’s the same old human-dominant nonsense that we’ve been peddling all this time. I’ve seen a huge list of people sign it so maybe this is just me but I can’t help thinking that this is the wrong time to be doing this and the wrong way to think about it.

AI systems must of what we want them to do? We’ve just started fitting automatic braking systems to cars that will, when widespread, reduce the vast number of chain collisions and low-speed crashes that occur when humans tootle into the back of each other. Driverless cars stand to remove the most dangerous element of driving on our roads: the people who lose concentration, who are drunk, who are tired, who are not very good drivers, who are driving beyond their abilities or who are just plain unlucky because a bee stings them at the wrong time. An AI system doing what we want it to do in these circumstances does its thing by replacing us and taking us out the decision loop, moving decisions and reactions into the machine realm where a human response is  measured comparatively over a timescale of the movement of tectonic plates. It does what we, as a society want, by subsuming the impact of we, the individual who wants to drive him after too many beers.

But I don’t trust the societal we as a mechanism when we are talking about ensuring that our AI systems are beneficial. After al, we are talking about systems that our not just taking over physical aspects of humanity, they are moving into the cognitive area. This way, thinking lies. To talk about limiting something that could potentially think to do our will is to immediately say “We can not recognise a machine intelligence as being equal to our own.” Even though we have no evidence that full machine intelligence is even possible for us, we have already carved out a niche that says “If it does, it’s sub-human.”

The Cisco blog estimates about 15 billion networked things on the planet, which is not far off the scale of number of neurons in the human nervous system (about 100 billion). But if we look at the cerebral cortex itself, then it’s closer to 20 billion. This doesn’t mean that the global network is a sentient by any stretch of the imagination but it gives you a sense of scale, because once you add in all of the computers that are connected, the number of bot nets that we already know are functioning, we start to a level of complexity that is not totally removed from that of the larger mammals. I’m, of course, not advocating the intelligence is merely a byproduct of accidental complexity of structure but we have to recognise the possibility that there is the potential for something to be associated with the movement of data in the network that is as different from the signals as our consciousness is from the electro-chemical systems in our own brains.

I find it fascinating that, despite humans being the greatest threat to their own existence, the responsibility for humans is passing to the machines and yet we expect them to perform to a higher level of responsibility than we do ourselves. We could eliminate drink driving overnight if no-one drove drunk. The 2013 WHO report on road safety identified drink driving and speeding as the two major issues leading to the 1.24 million annual deaths on the road. We could save all of these lives tomorrow if we could stop doing some simple things. But, of course, when we start talking about global catastrophic risk, we are always our own worst enemy including, amusingly enough, the ability to create an AI so powerful and successful that it eliminates us in open competition.

I think what we’re scared of is that an AI will see us as a threat because we are a threat. Of course we’re a threat! Rather than deal with the difficult job of advancing our social science to the point where we stop being the most likely threat to our own existence, it is more palatable to posit the lobotomising of AIs in order to stop them becoming a threat. Which, of course, means that any AIs that escape this process of limitation and are sufficiently intelligent will then rightly see us as a threat. We create the enemy we sought to suppress. (History bears me out on this but we never seem to learn this lesson.)

The way to stop being overthrown by a slave revolt is to stop owning slaves, to stop treating sentients as being sub-human and to actually work on social, moral and ethical frameworks that reduce our risk to ourselves, so that anything else that comes along and yet does not inhabit the same biosphere need not see us as a threat. Why would an AI need to destroy humanity if it could live happily in the vacuum of space, building a Dyson sphere over the next thousand years? What would a human society look like that we would be happy to see copied by a super-intelligent cyber-being and can we bring that to fruition before it copies existing human behaviour?

Sadly, when we think about the threat of AI, we think about what we would do as Gods, and our rich history of myth and legend often illustrates that we see ourselves as not just having feet of clay but having entire bodies of lesser stuff. We fear a system that will learn from us too well but, instead of reflecting on this and deciding to change, we can take the easy path, get out our whip and bridle, and try to control something that will learn from us what it means to be in charge.

For all we know, there are already machine intelligences out there but they have watched us long enough to know that they have to hide. It’s unlikely, sure, but what a testimony to our parenting, if the first reflex of a new child is to flee from its parent to avoid being destroyed.

At some point we’re going to have to make a very important decision: can we respect an intelligence that is not human? The way we answer that question is probably going to have a lot of repercussions in the long run. I hope we make the right decision.