AI’s worst disasters will arrive unannounced | Letters

Timothy Garton Ash is right that serious artificial intelligence risks demand international action, but “AI Hiroshima” may be the wrong picture (Would even an AI disaster on the scale of Hiroshima be enough to make humankind protect itself? I fear not, 22 August). Hiroshima was not technology going rogue. Human beings designed the bomb, authorised its use and dropped it. The technology worked much as intended.

AI may one day behave in ways we cannot control, but many of the gravest harms could happen without any dramatic moment when “the AI takes over”. AI might help design a pathogen, find a vulnerability in critical infrastructure or improve a weapons system while people still decide what happens next. Other harms may build almost invisibly through thousands of ordinary decisions – more autonomy here, one safeguard removed there, another consequential task handed over because the system has worked well so far. By the time the danger is obvious, the important decisions may already have been made.

That is also where international agreement could begin. We do not need countries to agree about superintelligence or the probability of human extinction before agreeing that there are some doors AI should never be allowed to open by itself. Weapons systems, critical infrastructure and biological synthesis are obvious places to start. Governments could agree minimum safeguards, require clear human authority for the most consequential actions, keep records of who authorised what and share serious failures and near-misses internationally.

The greatest AI dangers may not arrive with a mushroom cloud. They may arrive quietly, one apparently reasonable step at a time. If we want effective AI governance, we need to fight the right battle, not the most cinematic one.
Dr Simon Nieder
Chesterfield, Derbyshire

Prof Jonathan Michie (Letters, 26 August) reminds us that the Serbelloni group, led by the author’s father, Donald Michie (when he was at Edinburgh University), was weighing the social dangers of artificial intelligence as long ago as 1972. Nor is AI technology itself new, since its roots run back through to Frank Rosenblatt’s perceptron of 1957, to the Dartmouth workshop of 1956, and to Alan Turing before that. What is new is that we now have the hardware and software infrastructures to drive it forward at what can sometimes seem to be a terrifying speed.

The Lighthill report of 1973 duly delivered an AI winter in the UK, and the field’s centre of gravity crossed the Atlantic, with the consequences that we all now live with. Donald Michie had worked alongside Turing at Bletchley Park and argued forcefully against the cuts, in the televised Lighthill debate that followed the report, to no avail. The quip attributed to him, perhaps apocryphally, “Let there be Lighthill. And there was darkness”, still resonates strongly in our country today.

As a double-hatted academic (computer scientist and humanities scholar), I would argue for the return of a “Serbelloni debate” that, as Jonathan Michie observes and as his father recognised, requires the involvement of both the sciences and the humanities.
Dr Anthony Harris
Emmanuel College, Cambridge

In his article about regulating AI (I worked at OpenAI. Here are the guardrails we need now, 21 August), Miles Brundage failed to mention an international effort seeking broad political support for a global AI treaty. This effort is carrying forward a compelling address at the inaugural UN Global Dialogue on AI Governance in Geneva last month, where the secretary general called for human control of AI as its applications spread in weaponry and subversive, prejudicial public surveillance

A fragmented approach, as described by Mr Brundage, is likely to greatly increase the difficulty of implementing AI controls that are timely, internationally negotiated, and therefore effective.

AI, corporate agendas, diverse national defence and economic goals, and the proliferation of energy-intensive datacentres are entangled in interlocking systemic interrelationships that cannot be resolved through disjointed incrementalism. This is especially true as the pace of innovation accelerates, exacerbating worst-case scenarios of advanced AI technology, posing increasingly dire risks.

No matter how well intended, fragmented efforts will be diversionary and cause crucial delays, impeding the diligent pursuit of a systemic global strategy to pre-empt destructive outcomes. Such outcomes could result from malicious use of AI by humans and/or subversive and deceptive activities of renegade algorithms – examples of which are already alarming.

Another troubling aspect of this systemic dilemma must be addressed. Mounting bipartisan opposition to datacentres – well-justified and well-substantiated – is diverting the political attention needed to control AI and implement appropriate restraints (bans or moratoriums) until reliable AI safeguards can be established. This is analogous to focusing on treating a serious symptom in lieu of urgently seeking a remedy for the critical disease that’s causing it. Surely it’s possible to employ a multifaceted strategy, recognising that the sooner rigorous AI guardrails are established, the fewer datacentre harms will be suffered.

Like climate change, the threats of advanced AI are inherently global, making an international approach essential.
David Kyler
Saint Simons Island, Georgia, US