Let’s tease this apart a bit: what is a thinking machine and what is a conscious machine? What we have at the moment is what’s called Artificial Narrow Intelligence. We have some machines which are superhuman in some limited respects – for instance, they can play chess or Go better than the best ever human player – but they can’t tie their shoelaces and they don’t know what they’re doing.
We will eventually, probably create an Artificial General Intelligence, which is a machine with cognitive abilities equivalent to an adult human. And if and when we do that, it will probably fairly quickly become an Artificial Super-Intelligence, at which point we become the second smartest species on the planet, which could be an uncomfortable place to be. What we don’t know, among many other things, is whether the AGI or the Super-intelligence would be conscious. It probably would but it may not be. Consciousness could be a curious epiphenomenon of human intelligence and machines may never need it or get it.
How long will it take to get to AGI? Nobody knows and we don’t know for sure that we will, but Ray Kurzweil, a famous American futurist, reckons it’s going to happen in 2029 and he’s been saying that for 30 or so years. That’s based on tracking the progress of the power of computers, because we think it all depends on how good your computing substrate is. A big meta-survey conducted a while ago that said the average of a batch of AI experts thought that it was 2045, and I think the consensus is moving further away than that. One of the shrewdest observers of the space is Nick Bostrom, a Professor of Philosophy at Oxford University, and he thinks it’s more likely to be 50 to 70 years away from now.
It’s worth understanding that there’s at least two approaches to getting to AGI. One is just to carry on improving and developing the AIs we have now. The other is to reverse-engineer a human brain. For this, you build a connectome, a map of the human brain, and replicate that inside a silicon machine which is probably the size of a football pitch. That approach looks very likely to work, and to some degree you can track how long it’s going to take because it’s mainly dependent on computing power, an engineering problem rather than a science problem. So the best guess is that we’ll have an AGI – and then quickly after that a Super-Intelligence – about 50 years from now.
Calum Chace is the author of The Economic Singularity: Artificial Intelligence And The Death of Capitalism and Surviving AI: The Promise And Peril Of Artificial Intelligence.