AS if priming us for the Prime Minister’s Washington announcement that the UK will host the first global summit on Artificial Intelligence regulation in the autumn, Rishi Sunak’s ‘task force’ adviser on technology took to the airwaves on Monday to warn that we may have only two years to control AI before computers start threatening the human race.
Matt Clifford told Talk TV that he thinks AI systems may soon become so out of control they could wipe out swathes of humanity. Asked what keeps him awake at night, he answered: ‘The fact that the people who are building the most capable systems freely admit that they don’t understand exactly how they exhibit the behaviours that they do.’
Shadow digital secretary Lucy Powell struck a similar chord at the techUK conference on Tuesday, warning that AI creators need to show what is ‘under the bonnet’ in their software so ‘experts and regulators’ can ‘figure out’ the technology behind it.
Clifford anxiously envisages a future where a new ‘species’ of out-of-control, super-smart aliens could turn against us, saying: ‘It’s certainly true that if we try and create artificial intelligence that is more intelligent than humans and we don’t know how to control it, then that’s going to create a potential for all sorts of risks now and in the future.’
One can only hope that the new AI genus Clifford describes will program itself for critical thinking, since chief advisers and senior politicians seem terminally incapable of it.
Without wishing to be rude to Mr Clifford and Ms Powell, it is simply nonsense to believe we don’t know fully what is ‘under the AI bonnet’ – every last digit of its code is visible to its human creators, and much of it is in the public domain, accessible with a double click on any web page. As is the case so often, ignorance breeds fear, and when ill-informed hysteria is amplified by an equally susceptible news media the smog of illiteracy inevitably grows gloomier.
What is intelligence? What is a thought? What is consciousness? Is it really plausible that a computer can think? There’s little point in pontificating about this and that ‘existential threat’ unless you have examined these concepts. Without definition, words are meaningless and conversations which include them are disconnected from reality. Without a clear definition my washing machine, cooker, satnav, online calendar and pretty much the bulk of today’s technological aids might be described as ‘artificially intelligent’. But do we really want to use the concept so loosely?
I currently have the good fortune to be tutoring an 11-year-old boy in critical thinking. Recently we have been attempting to define ‘intelligence’ and are engaged in an enjoyable tussle over our different perspectives. Lucas’s view is that intelligence is essentially knowledge, and the more knowledge you have the more intelligent you are, an idea not too distant from many schools’ philosophy of education. Mine is that intelligence is a capability, to which Lucas consistently replies: ‘Ah, but you need the knowledge in order to be capable.’
This week I asked him if a library is therefore intelligent. We decided it isn’t, that a library is merely a vast warehouse of information. ‘Like Google,’ he added. ‘Yes,’ I agreed. ‘Google itself is not intelligent, it just goes and fetches stored stuff it is told to, like a librarian might fetch a book.’
‘But that means Google is intelligent,’ he pressed. ‘No,’ I replied, ‘it just does what you ask it without understanding, just as a librarian might bring you a book on a subject he or she has no understanding of.’
To inform our next discussion I have suggested a stunning, simple-to-read paper by Richard Feynman which brilliantly enunciates the fundamentals of computing. Feynman argues that computers – and by extension AI – are misnamed: ‘They are primarily filing systems. People in the computer business say, “They’re not really computers. They’re data handlers.” All right, that’s nice. Data handlers would have been a better name because it gives a better idea of the idea of a filing system. There’s a lot of data
on cards and you handle it. You take cards out, look at them, put them back, and so on.’
Like the librarian fetching books on subjects unknown to him or her, computers follow instructions to get stuff from various places and then, following other instructions, put it together in a potentially useful bundle. They do not need to understand what the filing cabinets are for, they just need to be able to find the files they are told to.
Today’s data handlers can retrieve and compile at almost the speed of light. You might say that is intelligent, and we do call people who can do maths quickly bright, but that ability is merely one small facet of rounded problem-solving capacity.
Critical thinking requires examination of fundamentals and exploration of key terms. Blindly accepting other people’s unexamined assumptions is anathema to critical thinking. Unfortunately this fact continually evades so many who hold forth on the ‘deadly dangers’ of AI.
A speedy filing system is not a new species, Mr Clifford. And computers have no desire to take over the planet simply because they have no desires, full stop. Please think about it. And perhaps point this out to the PM, too.