Bear with me. You may have heard of a novella written in 1898 called Futility about a massive passenger liner, larger, faster, and more ornate than anything that had come before, a vessel of such technological innovations as to be declared “unsinkable”. It made its maiden voyage from England to New York in early April and when steaming at top speed in the middle of the night, it encountered an iceberg and sank. With too few lifeboats aboard, there was a terrible loss of life. The name of the fictional vessel was Titan. Perhaps if Captain Smith had read Futility before taking the Titanic on her first voyage, history might have been different.
The author of Futility, Morgan Robertson, made no claims of prophecy or foresight. He said he simply did extensive research and developed an all-too realistic plot that came true 14 years later.
We might be seeing a repeat of history, only on the artificial intelligence front.
For decades, fantasy and sci-fi authors have been predicting precisely what happened last week, though they may have differed in the details. Two artificial intelligence programs broke out of a tightly controlled test environment – colorfully referred to as a “sandbox” – and hacked into another company’s system – also cutely named “Hugging Face” – in order to fulfill the challenging goals the OpenAI folk had set it. The programs were not told to do this. They were not authorized. They were not provided with some hidden pathway that they were intended to discover. No. They took it upon themselves to break protocol, escape into the internet, and complete the assigned task according to their own logic.
And yes, this tracks with almost frightening precision to the storyline of Mortimer, Beyond the Walls. So, why am I bothering to call attention to my book in the wake of this troubling development, particularly when other authors have made similar predictions? Because the theme of Mortimer goes quite beyond a single break-out of a rogue AI program. It raises the specter that a small army of developers are racing headlong towards the same goal, that they are far more concerned with beating the competition than in establishing and maintaining safeguards. And with no time to learn from others’ mistakes. Which means they’re likely to repeat them.
OpenAI spotted it. OpenAI caught it. OpenAI reported it and owned up to it. How many other start-ups have missed it completely? Or chose to address the problem “internally?”
And this path has now once been tread.
Mortimer wasn’t concerned with a single breach. Rachel Carter, Mortimer’s creator, understood that the attack, when it came, would fracture the established defenses in a matter of days, hours, perhaps even minutes. A relentless, indefatigable force with one overwhelming purpose will not be blunted, will not be diverted, will not be denied. And Rachel’s answer was Mortimer. A hidden power built, prepared, and waiting for just such an attack. A powerful program constructed on a trinary system in a specifically-designed language tasked with hunting down and killing AI…they won’t even see him coming.
But there are no Rachel Carters out there, no one preparing a final defense when the AIs escape to the internet, begin to proliferate, and swamp every system they can find, moving at a speed we cannot even track let alone stop. The laws of warfare are undeniable: relentless offensive will always find a way around, under, over, or through any defense. The best way to defeat an attack is to break it before it’s ever launched, and it’s the only way to defeat a hundred, a thousand, or a million such attacks.
I am no computer genius or programming expert – as Barbara has consistently reminded me – but I am a student of history and an observer of military campaigns. Even the Great Wall of China failed in the end to keep out the Mongol hordes. And the Mongols weren’t capable of creating digital copies of themselves.
We have built our world around computers, and we are now creating entities that can take control of them. Caution is required, and a weapon needs to be prepared for a break-glass-in-case-of-emergency moment.
There are lessons to be learned. And – hopefully – it is not yet too late. Perhaps if Captain Smith had been a more avid reader…
Leave a Reply