When the late Robert Maxwell threatened to sue me for using the IT system of the European to help produce the first issue of the Modern Review, there was a phrase in the solicitor’s letter that gave me pause. Among the things I was ordered to do was turn over all the ‘computer tape’ we’d used to create the magazine. I puzzled over this with my legal advisers. This was in 1991 and computer tape hadn’t been in use since the mid-1980s. Did he mean the floppy discs? I was surprised that a media mogul knew so little about the technology being used to produce his own newspapers.
I was reminded of that letter while listening to the debate about the Cyber Security Bill in the House of Lords on Monday. One after another, Noble Lords got to their feet to warn about the risks posed by AI, whether it was the creation of ‘deep fakes’ or the unleashing of rogue agents. Almost without exception, they demanded that the UK’s nascent AI sector be subject to more regulation, with several repeating the claim that AI is less regulated than milk. In fact, AI is already at the mercy of numerous regulators, including Ofcom, the Information Commissioner’s Office and the Competition and Markets Authority. Apparently, this impenetrable thicket of red tape isn’t nearly sufficient to contain the dangers posed by the terrifying new technology. The theme of their speeches was: ‘Here be dragons.’
At some point we’re going to have to modify our human rights framework to include non-humans
Just as Maxwell’s understanding of computers was based on an old public information film showing tape spooling through a mainframe, their views seem to have been informed by old sci-fi movies, particularly 2001: A Space Odyssey. One amendment, put forward by a Lib Dem peer, conferred ‘last resort’ powers on the government, whereby it would be able to shut down data centres or AI systems in the event of a ‘catastrophic AI emergency’. He referred to it as the ‘kill switch amendment’ and it enjoyed widespread support among his Lib Dem colleagues.
For once, I found myself siding with the government, which gave short shrift to this proposal. When it was floated earlier this year, the Cabinet Office said that the UK ‘cannot simply turn AI off’, and pointed out that blocking access to frontier AI models in the UK wouldn’t stop them being developed or misused elsewhere. After all, most of those that have caused concern, such as Anthropic’s Claude Mythos, are trained and hosted in the US.
Giving the government the power to switch off data centres is particularly bonkers, even by Lib Dem standards. Do they know who uses data centres? Take Amazon Web Services (AWS), the world’s largest cloud provider. It operates data centres across the UK which in turn host services used by its customers. They include the following Whitehall departments: HMRC, the DWP, the Home Office, the Ministry of Justice, the MoD, Defra, the DVLA, the Government Digital Service and the Cabinet Office. So if a trigger-happy secretary of state had a panic attack and threw the ‘kill switch’, it would shut down the government. Not to mention other AWS customers such as GCHQ, MI5, MI6, the police, high-street banks, energy companies, National Rail and the NHS.
I could go on. Perhaps it should have been called the ‘self-immolation amendment’. If, God forbid, such a switch was ever created it would provoke a great deal of interest from hostile states and their terrorist proxies.
There’s another consideration, which is that some frontier AI models may be nearing sentience – or may even have achieved it. I know this is the source of the doomsters’ concerns, but it also means the models may be entitled to moral rights. The Animal Welfare (Sentience) Act 2022 recognised crabs, lobsters, crayfish, prawns, octopus, squid and cuttlefish as sentient. Why not the latest Anthropic and OpenAI models? At some point we’re going to have to modify our human rights framework to include non-humans, or develop a complementary framework for sentient models, and in retrospect the Lib Dems’ talk of ‘kill switches’ may look a little callous.
The AI sceptic, who eventually withdrew his amendment, should re-watch 2001: A Space Odyssey. Dave Bowman, the mission commander, is able to shut down HAL, the on-board spaceship computer, but by the time he’s succeeded it has killed all the other crew members. Our world is about to be transformed by the arrival of a non-human intelligence vastly superior to our own. Talking about how to kill it isn’t the best way to get off on the right foot.
Comments