‘Humans must stay in control of AI and the pace of its development.'
Rishi Sunak, former UK Prime Minister & now senior advisor to Anthropic
The BBC interviews with Professor Geoffrey Hinton (the ‘godfather’ of Artificial Intelligence) and Senator Bernie Sanders on Newsnight last Thursday must have sent shivers down the spines of most listeners. This was followed by Jacob Coxon, the AI researcher who quit Anthropic, reporting that staff were ‘genuinely frightened’ for humanity’s future. On Friday there was a lead comment article in The Times by Gerard Baker, entitled, ‘Be afraid; everything is escalating out of control’, with a strap line ending ‘… with little real leadership to stop it’.
These are serious comments about AI getting out of control and they follow a series of warnings from industry experts such as Sam Altman, the founder of Open AI, Elon Musk and Dario Amodei, the head of Anthropic. Gerard Baker is right: the quality of national leadership leaves much to be desired at present, and there is no prospect of global leadership in sight, democratic or otherwise. It is indeed a dangerous situation.
We have called repeatedly for global governance, most recently a couple of weeks ago. However, the threat from Artificial Intelligence is so substantial and so imminent that a much more direct process of human participation is needed. We were explicit in our arguments for this in November 2023, following the world leaders’ summit on AI at Bletchley Park, organised by then Prime Minister Rishi Sunak.
Whereas a significant move towards global governance presents a major challenge, the need for convergent regulation is more readily — although sometimes reluctantly — accepted. The European Union has demonstrated this over several decades, and regulators also work hard to avoid loopholes in order to make that convergence effective.
So, while the UK Government has rejected the idea of a ‘kill switch’ for dangerous Artificial Intelligence, it is reasonable to look for a form of internationally accepted regulation in order to ensure that there is proper human oversight. Indeed, the formation of the AI Security Institute (AISI) following the 2023 Bletchley Park summit illustrates that there has already been a move in this direction, although Rishi Sunak accepts that his own company, Anthropic, did not given their latest model to AISI to test — it is, at present, a voluntary process.
At the same time, it is broadly accepted that the extent of wealth concentration as a result of AI, together with its major impact on the prospects for human employment, are aspects which cannot be ignored — indeed, Bernie Sanders drew attention to these threats in his Newsnight interview.
It is therefore entirely logical that a process of mass human participation in AI ownership (in terms of both wealth creation and oversight) should be included as a potential solution for maintaining continuing human control of its development. This is precisely what Share Alliance has been researching over the past three years with its ‘Stock for Data & Creativity’ programme; in May 2023 I explained this project to Microsoft’s ChatGPT, who responded with wishing me ‘all the best with your research and your vision’. I wonder if that conversation would end similarly today.
‘Stock for Data & Creativity’ has made considerable progress over the past three years. In late 2023, we spoke of the need for careful study from an economic perspective and, over last winter, Capital Economics has undertaken a detailed project illustrating its potential impact. The way in which technology diffuses both data and creativity is a global process; its harvesting is therefore global in nature, as should be the rewards.
The Capital Economics study has not only shown how the Share Alliance proposal would enable wealth participation across the world, but it has also provided an inter-generational perspective in order to illustrate its potential impact for young people, who will be most affected by Artificial Intelligence.
Our commentary in November ’23 also referred to the need to study the effectiveness of distributed governance, an aspect in which Meta has been seriously interested — it also discussed the potential interplay between regulators and mass ownership structures. A shareholder vision which enables the participation of billions of human individuals will require new processes of oversight and education.
In the United Kingdom, a three-month consultation has just been launched by the Department for Business, Innovation, Science and Trade (DBIST). One of its proposals is to allow the annual general meetings of public companies to take place virtually where there is shareholder consent.
Not surprisingly, this has met with some opposition from retail shareholder groups such as ShareSoc, whose director Mark Northway said that in-person AGM's were needed ‘to look directors in the eyes and hold them to account’.
However, if ‘Stock for Data & Creativity’ does result in the development of globally-accepted regulations requiring mass participation in the share capital of tech giants, it will of course be necessary to develop alternative processes of virtual control. Indeed, because of the speed of AI development, this will need to be an ongoing, not periodic, process. In this way, we can ensure that the most meaningful and widespread nature of human control is maintained as new models are developed.
We therefore look forward to discussion with DBIST to explore how this might operate, and how it can be made global in its application.
Share Alliance’s research into ‘Stock for Data & Creativity’ has now reached the point where much wider awareness and investigation is needed. As we commented on 18th May, the journey from research towards implementation requires both education and advocacy; we are now entering those intermediate stages.
So, while the UK Government may hold back from installing a ‘kill switch’ for dangerous AI, we look forward to their thoughts on how our proposals for control of AI development through human participation for all can gather global momentum.
Gavin Oldham OBE
Share Radio