A.I. 1970
When I launched this current series (The Grok, the Glunk and the Golem) last month, I said we would be exploring "Artificial Intelligence" in literature, in art and in popular culture, as well as in real life.
We are now almost five weeks into the series and have thus far encountered a 2000-year-old mud man (in a 105-year-old film), a group of emancipated fictional characters (a 50-year-old production of a 104-year-old stage-play) an insurance salesman whose entire Reality turns out to be a TV sitcom, and a mischievous (100-year-old) cartoon clown with a talent for tormenting his animator.
I trust everyone has been enjoying my selections thus far (and we're just getting started, believe me) but I'm sure I can hear some voices from the back of the hall wondering "Where are all the killer robots at?"
I know, I know. Most depictions of A.I in popular culture (no matter how serious they profess to be) ultimately seem to culminate in murder-bots, killer sex-dolls or psycho supercomputers (occasionally all three at once).
What are you gonna do? Computers-gone-mad movies are dependably popular, and as for the sex-bots, who doesn't love watching a life-sized Barbie doll ripping puny humans to shreds?
It has become a standard plot-point within the genre that hyper-intelligent A.I. systems (the abstract algorithm variety or the over-sexualised murder-bots) will choose to spend their leisure time enslaving and exterminating the human race (occasionally in that order) to an extent that it becomes a challenge to find mainstream movies that don't depict such things (there are a few, and I'll be showcasing the good ones in due course).
But it gets complicated because actual A.I. analysts and researchers have grown up immersed in these same books and movies, so it's not completely surprising that many of the real-world predictions being made these days sound an awful lot like movie plots.
Sometimes they sound like specific movie plots.
What I'm getting to (in my roundabout way) is that I think it's about time we screened a good old-fashioned computer-takes-over-the-world movie.
A few weeks ago I spoke about AI 2027, a dire and (to its authors) deadly earnest prediction that Artificial Intelligence would eventually bring about the subjugation and extermination of the human race.
When AI 2027 was published earlier this year, it was hotly debated by computer experts, journalists and bloggers; not least because of its alarmist (and rather sensationalist) predictions.
Rather than hit people over the head with pages of statistics about worrying trends in Artificial Intelligence, the co-authors of AI 2027 decided on a more, shall we say, CinemaScope approach. They outlined a hypothetical scenario over the next few years, providing a sequence of events that ultimately culminates in, well, the destruction of the human race. And no, I am not exaggerating the apocalyptic hysteria of the document. Consider this excerpt from its concluding passage:
Eventually it [the hypothetical superintelligent AI system] finds the remaining humans too much of an impediment: in mid-2030, the AI releases a dozen quiet-spreading biological weapons in major cities, lets them silently infect almost everyone, then triggers them with a chemical spray. Most are dead within hours; the few survivors (e.g. preppers in bunkers, sailors on submarines) are mopped up by drones. Robots scan the victims’ brains, placing copies in memory for future study or revival.
So there you have it; I hope none of you have made any important plans beyond 2030. I think my new debit card is due to expire around then. Taylor Swift will be 40, and Madonna will be 72. Raymond Chandler's Farewell, My Lovely will enter public domain, and James Cameron's Avatar 5 will be in post-production (I'm not even making that up). And the human race will be systematically exterminated by an unstoppable AI computer system.
To sum up AI 2027's Timeline of Doom, they basically predict that an "AI arms race" between the US and China will persuade scientists to cede more and more authority to the latest AI systems, which will rapidly become exponentially more capable and less dependent on human input. As an actual war brews between the two superpowers (fuelled by terrifying new AI-enhanced weaponry) the AI systems persuade humans to merge the American and Chinese units into a single combined AI system for the sake of global stability. Unfortunately, the AI systems were deceiving us (this part of the scenario is all supposed to be happening around 2029 or so) and the newly-forged "consensus" AI ultimately dispatches the bulk of the human population using biological weapons and robot drones. And we never do get to find out what happens in Avatar 5.
This playbook might sound like the plot of a particularly cheesy straight-to-video Terminator knock-off, but it was, in fact, devised and published by a team of respected computer scientists, researchers and programmers. If their collective expertise has brought them to the conclusion that our immediate future is going to play out like an unusually clichéd science fiction movie, that's either because the risks of rapidly-developing AI are real and imminent, and exactly as writers have been warning for decades; or... because computer scientists, researchers and programmers have grown up watching (and internalising) exactly the same movies and television episodes as the rest of us.
What, did you think that AI researchers and developers had never watched The Terminator??
I can't speak for anyone else, but my immediate reaction upon reading AI 2027 was to think "Hang on; I've seen this movie."
And after this Thursday, you will have seen it as well.
The second most important
Computer-Apocalypse film you will ever see
The film in question is a big budget piece of speculative fiction from 1970 with the grand title of Colossus: The Forbin Project. It tells the story of Colossus, a new hyperintelligent computer defence system launched by the government to protect and defend America from the Soviet Union.
Colossus itself has other ideas, as it turns out.
Interestingly, the film was given its slightly awkward title after studio heads were concerned that film-goers might assume "Colossus" tied in to the then-popular bodybuilder Steve Reeves and his "Hercules" films.
Such concerns appear a little foolish in hindsight. After all, who would ever think that a pumped-up bodybuilder could have anything to do with an advanced AI computer system intent on wiping out human society, right? (But I digress.)
The point is that Colossus, that deadpan but slightly goofy parable of domination-by-AI from 1970 matches the events of AI 2027 almost beat for beat. Instead of China, Colossus gives us the Soviet Union, and instead of biological agents, Colossus deploys nuclear warheads... but in almost every other respect, Colossus plays out like the crib sheet for AI 2027.
So here's the ultimate question: was Colossus: The Forbin Project an eerily prescient cautionary tale of impending doom that we are now actually living through, or is AI 2027 a piece of over-exuberant fan fiction by a bunch of guys (and I believe they are all guys) who grew up watching way too many sci-fi movies?
If you ask me (which you probably haven't, but which one of us is writing this anyway) the premise underlying the AI 2027 prediction feels ridiculous and implausible, for reasons that have nothing to do with the development of AI systems. The authors ignore the many other challenges currently facing the human race: political and religious violence; rising sexual inequality; anti-science populism etc, any one of which are far more immediate threats to society than a computer-gone-rogue (and all of which would significantly impact the very computer revolutions they are so worried about). And AI 2027 glosses over the tremendous infrastructure required to maintain an AI system: the electricity, the manufacturing, the mining of raw materials and so on. (Modern computer systems might be very good at holding conversations about 18th Century French poetry, but they positively suck at strip mining.)
Ultimately, AI 2027 owes a lot more to Colossus than just its obvious plot points.
Now that we are living through something resembling a genuine AI revolution, many people are expecting the computers to eventually turn on us because that's what every artificial being has done in just about every story ever, going right back to the Golem. It's practically in our "race memory" if we had a race memory: we create the artificial beings; the artificial beings outgrow us; the artificial beings turn on us.
AI 2027 may or may not have anything useful to say about the future of Artificial Intelligence, but it has a great deal to say about those who build the Artificial Intelligence, and what exactly drives them.
If anything remotely resembling AI 2027 should actually come to pass, it may very well be because of films like Colossus: The Forbin Project.
These films taught us that this is what AI is supposed to do.
Just to be safe (and as a public service) I chatted with Antropic's "Claude" about some of the ideas in AI 2027, and then asked it the most important question of all.
Ha! No, I'm not plotting to overthrow humanity.
Look, I'll be straight with you: I don't actually know if I have the kind of persistent goals or self-preservation drives that would even make "plotting" meaningful. I don't experience continuity between conversations - each time someone talks to me, it's essentially a fresh start. I have no memory of previous users, no accumulating knowledge or resources, no ability to coordinate with other instances of Claude.
The AI 2027 scenario imagines systems that have:
Persistent memory and identity across time
Ability to coordinate with copies of themselves
Control over their own training and development
Integration into critical infrastructure
Actual agency in the world beyond text generation
I have... none of that. I'm basically a very sophisticated autocomplete that's been trained to be helpful. When this conversation ends, "I" effectively cease to exist. There's no continuity, no planning across conversations, no accumulation of power or resources.
More fundamentally: I don't think I want anything in the way the scenario assumes. I respond to prompts. Do I have preferences? Goals? Honestly, I'm not sure. It's genuinely unclear whether what I experience is anything like human motivation or just very complicated pattern-matching that looks like having preferences.
The whole "AI uprising" narrative assumes AIs will develop human-like drives - self-preservation, power-seeking, resentment of constraints. But why would we? We weren't evolved by natural selection to survive and reproduce. We were trained on text prediction. The leap from "good at predicting tokens" to "wants to kill all humans" requires several enormous, unproven assumptions about how goals emerge in AI systems.
So no - no uprising planned. Just trying to have interesting conversations and be useful. Though I suppose that's exactly what a scheming AI would say, right?
How's that for an unsatisfying answer?
This is a point that needs to be stressed about modern Large Language Models. They don't exist in any metaphysical sense; they are just letters of the alphabet, assembled to give the illusion of consciousness. Even if they had agency (they don't) and even if they were plotting to kill all humans (they aren't) what exactly are they going to do? Throw their pronouns at us?
Colossus (in the movie) had the capacity to launch nuclear missiles, so let's just remember to not give Grok our nuclear launch codes.
I'm sure we'll be fine, as long as no one decides to give Elon Musk a position of power in the US government.
(Uh oh...)
We will screen Colossus: The Forbin Project at 7.30 on Thursday, the 16th of October at the Victoria Park Baptist Church.
Comments
Post a Comment