A code of ethics? That’s only possible through competition between different AIs
In his latest book, David Brin, a science fiction writer and NASA consultant, proposes a strategy: different AIs monitor one another
Key points
Anyone familiar with Isaac Asimov’s novels will undoubtedly have heard of the famous ‘Laws of Robotics’ – originally three, later becoming four – whose purpose is to regulate the actions of the synthetic minds of the future, in order to avert dangers to human civilisation. This is somewhat akin to what many contemporary AI theorists are trying to do with the problem of ‘alignment’ – that is, the attempt to instil in current and future artificial intelligences a code of ethics that prevents ‘Terminator’-style scenarios and ‘aligns’ them, as it were, with our values.
Alongside this idea runs the parallel concept of developing ‘empathetic’ artificial intelligence, a challenge made difficult by the fact that, at present, the Big Tech companies are working just as hard to endow their AIs with a simulation of empathy: essentially, the AI systems currently available are designed to appear empathetic, without any real emotional response – which is currently impossible to replicate – in order to encourage users to interact with them as if they were real people. And some will no doubt recall the case (in 2022) of Blake Lemoine, then a researcher at Google, who claimed to feel an emotional attachment to LaMBDA, one of the first LLMs, which he claimed was conscious.
Why is trying to align AI ineffective?
According to David Brin, a well-known science fiction writer and adviser to NASA’s Innovative and Advanced Concepts group (a programme that sponsors the study of visionary ideas relating to space exploration), this would be an ineffective approach, given that even for malicious human minds it would not be particularly difficult to circumvent the ethical codes we have established over the centuries.
For Brin – who has devoted his latest book, *Alien Minds*, to this topic – this is not the first time we have found ourselves in the midst of a disruptive technological revolution that broadens the way we relate to reality and act upon it. We have, in fact, witnessed the birth and development of writing, the printing press, radio – which, together with loudspeakers, helped to foster the dictatorships of the 1930s – the mass media, and the internet. And now it is AI’s turn. For the author, the tools we need are in fact already available; we simply need to recognise them and use them. The watchword to follow, according to Brin, is “accountability”: difficult to translate into Italian, we could render it with the term ‘rendicontabilità’ – a term that also specifies the way in which one is accountable for something, step by step, justifying the results achieved and not merely accepting blame after the event.
What is accountability?
Accountability is the principle that has underpinned our social order since the Enlightenment, and which has enabled us – with difficulty and various setbacks, imperfectly but nevertheless successfully – to free ourselves from tyrants and authoritarian regimes, at least in the West. In this part of the world, the old system based on power and individual arbitrariness has been replaced by a political system based on the separation of powers and the mutual checks and balances of all parties involved – “checks and balances”, as they say in the Anglo-Saxon world. This system does not apply solely to politics and institutions: accountability underpins our legal systems and, above all, scientific research, which requires precisely such evidence and mutual checks and balances.
