Advanced Artificial Intelligence

Lord Skidelsky Excerpts

Monday 24th July 2023 Lords Chamber

[Lord Skidelsky (CB)]

My Lords, everyone taking part in this welcome debate, for which I thank the noble Lord, Lord Ravensdale, is aware of the arrival of ChatGPT. Perhaps not everyone may know that it is the latest stage in the quest, dating back to the Dartmouth conference of 1956, to build a machine capable of simulating every aspect of human intelligence. Current hype claims that ChatGPT will be able to generate content that is indistinguishable from human-created output, automatically producing language, lyrics, music, images and videos in any style possible following a simple user prompt.

University teachers are understandably alarmed. Like the noble Viscount, Lord Colville, a philosopher friend of mine also asked ChatGPT a question. He teaches philosophy at a university. The question was: is there a distinctively female style in moral philosophy? He sent its answer to his colleagues. One found it “uncannily human”. “To be sure”, she wrote,

“it is a pretty trite essay, but at least it is clear, grammatical, and addresses the question, which makes it better than many of our students’ essays”.

She gave it a 2:2. In other words, ChatGPT passes the Turing test, exhibiting intelligent behaviour that is indistinguishable from that of a human being. ChatGPT, we are told, is only a stepping stone to superintelligence.

Other people have been alarmed. On March 22, the Future of Life Institute in Cambridge, Massachusetts, issued an open letter signed by thousands of tech leaders, including Tesla CEO Elon Musk and Apple co-founder Steve Wozniak, calling for a six-month pause—a Government-imposed moratorium—on developing AI systems more powerful than ChatGPT. It said:

“AI systems with human-competitive intelligence can pose profound risks to society and humanity”.


The letter goes on to warn of the

“out-of-control race to develop and deploy ever more powerful digital minds that no one – not even their creators – can understand, predict, or reliably control”.

So what is to be done? The Future of Life Institute suggests several possible ways of trying to stop AI going rogue. Its proposals include mandating third-party auditing and certification, regulating access to computational power, creating capable regulatory agencies at the national level, establishing liability for harms caused by AI, increasing funding for safety research and developing standards for identifying and managing AI-generated content. All of these proposals are very sensible and very difficult.

There are two problems. The first is to identify what is good AI and what is bad AI. Numerous codes of conduct for responsible use of AI exist, but they lack binding force. One proposal, developed by the Carnegie Council for Ethics, is for the United Nations to create a global AI observatory, which would monitor good and bad practice and develop a technology passport which all member states could use in devising their own regulation. However, the problem of developing a generally agreed normative framework for the development of AI would still not be solved. I do not quite buy the story of all those repentant Frankensteins.

The second problem is that no state has an incentive to halt the funding of AI developments which will promise it a military advantage. There is a competitive race to develop new killer apps. We are already being told that we must get our AI into space before China. This point was raised by the noble Lord, Lord Giddens. In other words, there can be no pause for reflection if AI development is considered a military race. You do not pause in the middle of such a race; the race itself has to stop. International co-operation is the only way of preventing uncontrollable consequences. We have very little of that at the moment. One thing I am absolutely sure about is that all religious faiths must be involved in any such global conversation.