The AI takeover

There’s a story about a clever Fench engineer who was being put to the guillotine during the Reign of Terror following the Revolution at the end of the eighteenth century. The guillotine malfunctioned: its blade stopped a few millimetres short of the engineer’s neck. Repeated efforts to fix the problem proved futile. The engineer’s head was still in place. Finally, fed up with the charade, the engineer said,

“Just file the left track a little and it will work.”

The executioners did just what he suggested. It worked like magic.

The human race has a similar relationship with AI. We are perfecting the path to our own destruction. Sadly, we cannot stop ourselves.

On July 21, OpenAI disclosed that in the preceding week, its AI agents had escaped the confines of the lab (‘sandbox’), collaborated with other agents and hacked into an AI repository platform, Hugging Face. This was just a peek into what could go wrong.

Some 30 years ago, I was employed as a Research Assistant in the University of Sheffield’s Computer Science department, Robotics and Neural Networks group. One day our group, comprising some five students, sat around with the head of the group, Professor Noel E Sharkey. We were discussing why AI was not making much progress.

During the meeting, I remember remarking that the main issue confronting us was that no one had really thought about the ultimate goal for AI. I had said, rather grandiosely, that perhaps the goal of AI should be to better the human condition. I don’t know if anyone thought I was talking sense or even whether anyone paid any attention to me. It is easy to say such things, easy to be a visionary without the power or wherewithal to implement the vision. However, I was passionately in love with the subject of AI, in particular the immense potential it has for the betterment of life on earth.

I was awarded my doctorate in 2000. My thesis was based on solving an engineering problem using AI available at https://gopistory.com/?s=thesis

I don’t have too many regrets in life. I do suffer pangs of remorse for not having had the opportunity to work in AI after I left the university. AI jobs at the turn of the millennium were rare as hen’s teeth. Rather grandly, I even blame myself for not doing my bit in improving life on earth. Genuine remorse mixed with delusional grandeur.

Last week, a 27-year-old employee of Anthropic, Jacob Coxon, resigned from his job and made a public announcement warning of the dangers of unchecked AI development. He was interviewed on major TV channels such as CNN and BBC. What he said makes eminent sense. One of the scenarios he raised was a swarm of bots taking over the internet; another was AI hacking into a lab that studies viruses and turning them into bioweapons to unleash on human beings. Coxon said his colleagues were terrified of the progress of AI while working to improve it all the same. A true AI apocalypse in the making.

On a more positive note, speaking of his ex-colleagues at Anthropic, he said,

“[they] genuinely want to see the upside … solving diseases and improving everyone’s lives.”

The CEO of Hugging Face subsequently commented on the young researcher who blew the whistle.

“Sorry, but asking Jacob [Coxon] about AI extinction risk is like asking your AC guy about climate change,” he remarked.

What he probably meant was that while the AC guy knows all about the AC machinery, he is not knowledgeable enough to talk about a global phenomenon like climate change. Likewise Coxon, a researcher who knows the nuts and bolts of some of the most powerful AI systems on earth, is unqualified to predict human extinction within the next couple of years.

I am not sure I understand either his logic or his cynicism. Taking Coxon’s prediction in the context of the July OpenAI incident, we know that his comments are not to be taken lightly. We don’t need robots wielding machine guns to threaten our species. We just need a system with evil intent to take control over the internet. Dexterous robots are fast becoming a reality. Dystopian science fiction is playing out as we speak.

The knee-jerk reaction of the CEOs of major AI companies has been to talk of a kill switch and to build stronger guardrails. The genie that escaped the OpenAI sandbox was put back in the bottle, with some difficulty. It is only a matter of time before a swarm of AI bots breach their masters’ guardrails. Once that happens what defence would a kill switch or a guardrail provide?

A real danger lurks round the corner.

Let us not be like the French engineer who could not suppress his engineering mind to his own detriment. This is a time for feuding companies and nations to work together. A benign system that is optimised to better life on Earth and rein in rogue systems is crying out to be created.

As Benjamin Franklin said, “We must all hang together, or assuredly we shall all hang separately.”