Over the past few days, in my articles about Artificial Intelligence, I’ve presented what I hope is a fairly comprehensive survey of the likely consequences of AI. I have given what I believe to be an accurate, though inevitably somewhat depressing, idea of what potentially lies ahead. Although it is not an optimistic picture, many of the worst consequences are capable of being mitigated – but only if governments around the world (including the UK) take tough action in good time. That will require governments to immediately start seriously planning and investing in those mitigations now rather than when it is too late. Many governments realise that some mitigation will be necessary – but relatively few, if any, have fully taken on board the scale and magnitude of domestic and international intervention that mitigation will require.
In the UK, for instance not a single major political party has yet developed a fully thought-out and detailed publicly accessible strategy for avoiding disaster. And the situation in other countries is, on the whole, no better.
Not a single major political party anywhere adequately outlines how AI-induced mass unemployment might be averted or how the consequent likely AI-induced reductions in purchasing power (and the resultant economic disasters) can be avoided.
Likewise, no UK political parties have publicly addressed the massive security and mental and physical health problems that AI will pose, let alone how to solve them. And the picture in the rest of the world is generally no better.
There is, for instance in the UK, often an air of naivety pervading government and Opposition AI policy. For instance one key publicly available UK government document (the so-called Artificial Intelligence Playbook for the UK Government published last year) states that “government engagement with academia, industry and especially the broader civil society is crucial to dispel fears and foster understanding of the potential benefits of AI technologies”.
That sadly shows that, without doubt, the UK government currently sees itself as an enthusiastic advocate for AI, as do many other governments around the world.
AI is here to stay – but naïve enthusiasm/evangelism is probably not the right governmental/political mindset needed to effectively mitigate the horrific potential effects of the technology.
Too many politicians tend to be reactive rather than proactive – and often tend to be relatively poor at long-term planning. Inter-party political competition also tends to force politicians to address simpler less massive problems rather than huge challenges like AI, climate change or nuclear proliferation. Politicians don’t tend to like complexity – partly because the public and the media don’t like it either.
Too many politicians in Britain and indeed in other countries currently simply (and sadly probably simplistically) think that the AI revolution is like the computer and internet revolutions of the late 20th century and early 21st century or the Industrial Revolution of the late 18th century and early-to-mid 19th century. Those past technological revolutions created a lot of problems but in the end created more jobs than they destroyed and created more prosperity than they destroyed. However, the AI revolution is ofa totally different nature. It is not in any way comparable to previous technological changes. It has the capacity not simply to help humans but to fundamentally replace them.
That unpalatable fact has not yet being fully and actively taken on board by the vast majority of politicians – and that probably helps explain why AI is not being treated as the massive threat that it is (and why fully thought through solutions have not yet been developed, let alone adopted, by the political classes).
One key problem has been the way in which most of the big landmark AI economic impact reports (usually produced by major think-tanks and financial institutions etc) have been researched and produced. On the whole, they have been generated by economists, technology analysts, corporate strategists and business consultants. Sociologists, psychologists, medical researchers, demographers, climatologists, educationalists, historians, criminologists and ecologists etc (ie., the very people who know about the world beyond business/technology) have not, in the main, been involved in most of the really big economically-oriented landmark AI impact reports that politicians are more likely to read and be influenced by.
But there are potential solutions to the approaching AI-generated threat which politicians, public commentators and civil society leaders should now begin to evaluate.
So here is a list of the sort of things I believe will need to be done to ensure that AI does not lead to untold misery – and to make sure it does not destroy our world:
(1) Strengthen international organisations etc: Competition between countries is one of the principle driving forces behind the rapid development and expansion of AI – so stronger international regulation is critical, if the world is to successfully prevent the worst job displacement consequences of uncontrolled AI deployment. But the current global drift towards increased and often aggressive nationalism and increased economic rivalry (often caused by the drift to the political right in so many countries) is undermining the rules-based international order (and international organisations like the United Nations) rather than strengthening them.
(2) Interventionist policies: Mitigating the adverse employment, social and other consequences of AI will also require intervention by governments in their respective countries’ employment and AI deployment practices.
(3) Job sharing: For instance job sharing systems may need to be compulsorily introduced.
(4) Shorter working week etc: A compulsory shorter working week (potentially a four or even a three day one) with no reduction in pay – and longer annual holiday entitlements – will also need to be considered.
(5) Minimum remuneration: Minimum salary/wage levels may need to be set and enforced.
(6) Retraining: Some payments for people without jobs may, for some age groups, need, where appropriate, to become remuneration for attending retraining courses and reaching appropriate skills-acquisition targets.
(7) Control of AI deployment: Use of AI may need to be limited to a particular share of a company’s productivity – and that will require giving government agencies the ability to inspect and ensure that those limits are being kept to.
(8) Non-AI-impacted jobs: Government investment is needed in developing employment sectors that are less vulnerable to AI displacement, including any that might be generated by a 3 or 4 day weekend.
(9) Agreed International common standards: What’s more, to succeed, a lot of these interventionist practices will need to be adopted by most economically developed nations across the world – otherwise non-adopting countries will be able to unfairly compete with adopting ones. Again, that will require an increase in the international rule of law, the supremacy of international treaties (not, as at present, a decrease).
(10) Controlling extreme artificial intelligence: The most extreme AI (known as ‘Artificial General Intelligence’ [AGI] has not yet been developed – but it very likely will be within a few years. AGI will be more intelligent than the combined intelligence of the entire human population of our planet. It is therefore essential that its development is brought under very strict control or indeed prevented. Sadly, it is already probably too late to ban its development – but it should only be developed under government and international licence. AGI is not just an economic threat. It is also a huge security threat to the world – so ways of restricting and potentially combating it will urgently need to be devised.
(11) Data centres and other AI infrastructure should become tightly regulated public utilities, either in private or public ownership.
(12) Price controls on all AI-produced or semi-AI-produced goods and services will probably need to be introduced.
(13) Hardware level verification – compulsory tracking of key microchips so as to ensure that all data centres are properly regulated and cannot fully or partially operate clandestinely
(14) Stricter limits on the size of tech companies will need to be introduced to prevent monopolies or excessive market domination.
(15) Strict limits will need to be set and enforced to ensure that AI cannot supply information which would assist terrorists, assist in the manufacture of weapons of mass destruction and other weaponry or threaten societal stability and democracy – and that AI can never function with total autonomy.
(16) Strengthen human control and alignment by (a) funding and enforcing research [to ensure AI systems inherently share human values]; (b) creating third-party bodies to test models before public release; (c) mandating hard-coded, unalterable emergency shutdown procedures for all autonomous AI networks; and (d) requiring AI decisions to be traceable and fully understandable by human overseers.
(17) Strengthen AI’s digital Integrity by (a) cryptographically signing all AI-generated media in order to combat deepfakes; (b) creating unalterable registries for authentic human-made journalistic and creative content; and (c) holding AI developers legally responsible for systemic harms caused by their AI system outputs.
(18) Strengthen economic and societal safeguards by (a) levying specialised taxes on automated productivity to help mitigate AI social impacts; (b) creating public wealth funds from tech giants; (c) very strictly regulating all AI companion systems; and (d) restricting and strictly regulating the utilization and commercialization of human biological and voice data.
(19) Strengthen open-source and proliferation control by (a) compelling developers to robustly secure the underlying codes of new advanced AI frontier systems (so-called ‘frontier models’); (b) banning the public release of models capable of cyberwarfare or bioweapon or other weapons design; and (c) enacting tough ‘whistle-blower’ protection legal shields for tech employees who report safety violations.
(20) Strengthen environmental and resource-use controls by (a) setting strict electricity-grid-use limits on data centres to protect climate goals; (b) strictly regulating the massive water resources required by data centre etc infrastructure for equipment cooling purposes.
(21) Ensuring that all or most of the costs of these mitigation strategies are met by taxing AI tech companies
(22) Finally and crucially there is an urgent need for more multi-disciplinary (and thus more comprehensive and objective) academic research to be carried out into the likely and possible future impacts of AI. Too many think-tank and financial institution research papers so far have concentrated on purely economic impacts (and even many of those have been inexplicably over-optimistic). To be credible, future research will need to be truly multi-disciplinary, covering the psychological/cognitive, mental health, physical health, security, crime, environmental, climate change, educational, societal, political, geo-political, trade, demographic and other impacts as well as, of course, the purely financial and economic ones (which should never be assessed in isolation).
AI is likely to very seriously impact the world at exactly the same time that the effects of global warming and (potentially of nuclear proliferation) will do so – and to reduce and mitigate all those problems (and their interaction with AI-caused ones) will also require stronger international rules and systems.
Sadly, although all these global challenges will affect us and our world even more than most purely domestic issues, they are not near the top of most government’s political agendas (including the UK’s).
That’s because of the nature of the media and of most political systems. That is, of course, at least theoretically capable of being changed – but only with immense grassroots pressure and political involvement, and that sadly is not currently happening.
I live in hope.
If you found today’s article of interest (but have not yet read the previous four articles in the series – on AI’s likely economic impacts, on its military/security impacts, on its psychological and social impacts, and on its environmental ones), you can access them, below:
