Written evidence submitted by Professor Huw Price (ROB0031)

The Future of Artificial Intelligence – the Long View

Written evidence submitted in a personal capacity by Professor Huw Price (Bertrand Russell Professor of Philosophy, Cambridge)

Executive Summary

• There are good reasons to take seriously the possibility that artificial intelligence (AI) will eventually outstrip human intelligence, perhaps greatly so.
• This may happen in the lifetimes of our children and grandchildren. Its impacts, for better or worse, are likely to be immense.
• Our best prospect of ensuring that this development is beneficial is to tackle it cooperatively, and as early as reasonable foresight allows.
• The UK Government can play an important role in fostering the academic, technological and policy-level coordination this process is likely to require, both nationally and internationally.
• In particular, by establishing an appropriate standing body, the Government can help to ensure that we do not fall into the trap of ignoring this important long-term issue, when short-term issues always seem more pressing and more tractable.

Introduction

1.              I am a Co-Founder, with Baron Rees of Ludlow and Mr Jaan Tallinn (Skype), of the Centre for the Study of Existential Risk, Cambridge. I am also Director of the new Leverhulme Centre for the Future of Intelligence, which is to be based in Cambridge, with partners at Oxford, Imperial College, and UC Berkeley. In these roles, I have been involved in recent discussions about the long term future of AI. I am writing with respect to the fourth issue on which the Committee seeks submissions: “The social, legal and ethical issues raised by developments in robotics and artificial intelligence technologies, and how they should be addressed.” I focus exclusively on long-term issues, and offer my personal recommendation concerning the role that the UK Government can most usefully play in the short and medium terms, with its eye on the long term. My focus is as much on benefits as on risks, but this response is coordinated with that of Simon Beard (Centre for the Study of Existential Risk, Cambridge), which deals with the uncertainty in evaluating possible long term risks of AI.

The prospect of superintelligence

2.              I J Good was a Cambridge-trained mathematician, who worked with Alan Turing at Bletchley Park, and at Manchester after the War. In their free time, Good and Turing often talked about the future of machine intelligence. Both were convinced that machines would one day be smarter than us. In the 1960s, when Good emerged from a decade at GCHQ, he began to write about the topic.

3.              In his first paper[1] Good tries to estimate the economic value of an ultra-intelligent machine. Looking for a benchmark for productive brainpower, he settles impishly on John Maynard Keynes. He notes that Keynes' value to the economy had been estimated at 100 thousand million pounds, and suggests that the machine might be good for a million times that – a mega-Keynes, as he puts it.

4.              But there’s a catch. "The sign is uncertain" – in other words, it is not clear whether this huge impact would be negative or positive: "The machines will create social problems, but they might also be able to solve them, in addition to those that have been created by microbes and men." Most of all, Good insists that these questions need serious thought: "These remarks might appear fanciful to some readers, but to me they seem real and urgent, and worthy of emphasis outside science fiction."

5.               In one sense, the prospect that concerned Good remains the same, fifty years later. It boils down to four key points:

• We have no strong reason to think that high-level intelligence is less possible in non-biological hardware than it is in our skulls.
• We have no reason to suppose that "human level" marks an interesting limit, in non-biological systems, freed of constraints (e.g., of size, energy consumption, access to memory, and slow biochemical processing speeds) that apply in our own case.
• So we should take seriously the possibility that AI will reach places inaccessible to us – kinds and levels of intelligence that are difficult for us to understand or map, and likely to involve capabilities far beyond our own, in many of the tasks to which we apply our own intelligence.
• Finally, there’s a prospect that AI systems will themselves contribute to improvements in AI technology, at some point. As Good saw clearly, there’s then a potential for an exponential rate of development.

6.              The big change since 1965 is that the incentives that will lead us in this direction are now much more obvious. We don’t know how long the path to high-level machine intelligence is, but we can be certain that its individual steps will be of huge commercial value, and immensely important in other ways – for security purposes, for example. So we can be sure that these pressures will take us in that direction, by default. AI is already worth trillions of dollars – perhaps not yet a mega-Keynes, but well on the way.

7.              At present, however, AI is very good at (some) narrowly-defined tasks, but lacks the generality of human intelligence. The term artificial general intelligence (AGI) is used to characterise a (hypothetical) machine that could perform any intellectual task that a human being can, including tasks not tied to specific set of goals. The term artifical superintelligence (ASI) refers to an AGI that greatly exceeds human capacities, in these respects.

When is AGI or ASI likely to arrive, and what would it mean for us?

8.              A time-line for the development of AGI is difficult to predict, in part because it may depend on an unknown number of future conceptual advances. A recent survey of AI researchers reported that most regarded AGI as more likely than not, well within this century.[2] It does not seem alarmist to say that while it is not on our doorsteps, it may be only “decades away” (as a leading AI researcher puts it recently, intending to dispell the popular impression that it is just around the corner).[3]

9.              Concerning the impact of AGI or ASI, we know little more than Good. It does not seem controversial that its impact is likely to be very big indeed. The world-leading AI researcher Professor Stuart Russell (UC Berkeley) is convinced that – for better or worse – it would be “the biggest event in human history.”[4] But the sign is still uncertain, as Good put it. The potential benefits are immense, not least in the light of AGI’s potential to solve many other problems. But there’s also a risk. As Turing himself put it, “It seems probable that once the machine thinking method has started, it would not take long to outstrip our feeble powers. ... At some stage therefore we should have to expect the machines to take control.”[5]

What can we do now?

10.              These issues are going to be with us for a long time, and are likely to become more pressing as AI develops. In the short term, the obvious strategy is to attempt to foster the level of interest, expertise, and cooperation that the task is likely to require in the future. In effect, we should be trying to steer some of the best of human intelligence to the job of making the best of artificial intelligence. Most of all, in my view, we should avoid the mistake of putting off the issue to another decade or generation, on the grounds that it seems too hard or too much like science fiction, or because other issues in the same area simply seem more pressing.

11.              There are encouraging recent signs of rapidly growing interest in these issues, for example in an open letter now signed by many AI professionals and others, following an international meeting in Puerto Rico in January 2015.[6] In particular, there is a growing sense of the desirability of cooperation between technology, policy, and academic partners. Some of this cooperation will necessarily be pre-competitive sharing, for commercial and other reasons – but all the more reason to engineer the kind of trust and cooperation that make such sharing possible.

12.              The UK is playing a leading role in this recent collaborative effort. Britain has great strengths in relevant academic and technical fields – as is well known, for example, the world-leading AI company Google DeepMind is based in London. On the academic side several UK centres are already prominent in international discussions in this area: the Future of Humanity Institute in the Oxford Martin School, Oxford; the Centre for the Study of Existential Risk, Cambridge; and the new Leverhulme Centre for the Future of Intelligence (Cambridge-based, with partners in Oxford, Imperial College, and UC Berkeley). The new Alan Turing Institute is also likely to play an important role, and there are many excellent individual researchers elsewhere in the country.

What useful role is there for government, given the long time horizon?

13.              The likely time-scale of these developments, and their dependence on ongoing research and progress in the field, makes a decisive intervention at one point in time impractical. More than in most cases, we are bound to be scanning a moving horizon. Nevertheless, there is a clear role for government that is likely to be beneficial, no matter how the field develops. It can foster, promote, and add its voice to a cooperative effort, both nationally and internationally, to monitor developments in the field, to flag opportunities and challenges as they arise, and generally to try to ensure the community of technologists, academics and policy-makers is as well prepared as possible to deal with both.

14.              What the UK government can most usefully add to this mix, in my view, is a standing body of some kind, to play a monitoring, consultative and coordinating role for the foreseeable future (and hopefully well beyond it). By ensuring from the beginning that the focus of this body is explicitly on long-term issues, there is an opportunity to lessen the risk that long-term issues will always be pushed aside in favour of short-term concerns.

Recommendation

15.              I recommend that the Committee propose the creation of a standing body under the purview of the Government Chief Scientific Adviser, charged with the task of ensuring continuing collaboration between technologists, academic groups including the Academies, and policy-makers, to monitor and advise on the long-term future of AI.

April 2016