Economic disruption and runaway AI: what can governments do?

by Richard Stirling and Hannah Miller

“The existing international organisations and governance struggle to keep up with technological change.” With this statement, Francis Fukuyama neatly captured the prevailing sentiments of the AI discussions at the 2018 World Government Summit in Dubai.

The World Government Summit brings together 4000 influential leaders from the public and private sectors, prominent thinkers, and policy-makers from around the world for three days of discussions. The aim is to discuss the future of government and improve the lives of citizens, including by capitalising on innovative technologies such as artificial intelligence (AI).

The central theme that emerged from the talks and debates was the importance of balancing short-term government action with long-term international regulation in AI development.

Short-term action

Governments around the world are grappling with the same questions on AI: What is it? What can it do? What does it mean for our country? There are three pressing questions that governments need to answer in the short-term:

• How to manage the transition in the economy? AI is forecast to cause significant economic disruption, especially through changes to the labour market. Governments should have a comprehensive strategy in place to manage these economic changes, including potential job losses.
• How to ensure the use of AI reflects your country’s values? AI development should be unique to each country’s circumstances and values. The danger of AI development is that it can involve relying on Silicon Valley, and in particular a small cluster of companies in Silicon Valley, who control the majority of the world’s data and most of its AI talent. How can Governments ensure that decision-making in their country continues to reflect the values of their citizens, and not just the preferences, approaches and commercial strategies of Silicon Valley?
• How to use the power of AI to deliver better services to citizens? Governments are now waking up to the transformative potential of AI in the field of public services. How can AI be used most effectively – and most ethically – to help deliver better public services, to more people, more cheaply?

Governments’ responses to these three questions will help determine whether they become world leaders in the AI revolution. Comprehensive, wide-reaching AI strategies enable countries to set out their policy aims for addressing these key issues. As we discussed in a previous blog post on the current state of global AI policy, some countries have made much more progress developing AI strategies than others.

Regulating artificial general intelligence

Alongside the excitement surrounding AI’s potentially enormous economic and social benefits, there was plenty of hand wringing and existential angst. These concerns focused on the rise of artificial general intelligence (AGI). AGI is considered by many researchers to be the ‘holy grail’ of AI; a general-purpose system that can successfully perform any task a human can.

In theory, the value of artificial general intelligence is huge. There is, however, a risk that it could turn into a less benign super intelligence, and present an existential risk to the human race. The type of risk is unclear: would we end up working for it? Would it commandeer all our resources? In his talk, Nick Bostrom gave a useful example to demonstrate a potential unintended consequence of developing an AGI. An artificial general intelligence sees you are hungry – but you don’t have any food in the house. So, it kills your cat – to feed you, because you are hungry.

So, can AI, and in particular artificial general intelligence, be trusted? And how can these technologies be regulated – especially if they are as powerful as we think they might be?

Some have compared artificial general intelligence with nuclear weapons, a threat to our future that brought nations together to negotiate the Nuclear Non-Proliferation Treaty in 1968. They have suggested that a similar treaty could govern the development and use of artificial general intelligence.

We think the analogy is flawed. Nuclear weapons are very expensive to build, requiring the wealth and resources of a nation state. Once they’ve been built, they need to be tested – something which is now impossible to carry out in total secrecy.

These constraints do not apply to the development of AGI. It requires smart people, and computing power. From the outside, it can be difficult to tell the difference between a super-computing cluster used to run an AGI, or one being used to run blockchain.

With this in mind, what could a government regulate?

• primary research – in other words, prevent people from researching interesting questions, because they might lead to bad outcomes.
• applied research – regulating access to the technology that can help to realise the impacts of primary research, such as supercomputers.
• a framework for artificial general intelligence – government could provide a set of limitations that must be designed into any use of artificial intelligence, for example to ensure that an AI always followed certain defined human interests.

Achieving short-term action and longer-term regulation will be essential to the development of AI that is both effective, and beneficial, to citizens around the world. The World Government Summit was a useful way to help leading policymakers come up with a collective definition of what the problems might be. The next step is to begin finding solutions.