Several pieces of fiction, for example the Terminator films, have popularised the idea that advances in AI could pose a serious threat to human civilisation. However, many people who are seriously concerned about issues such as the alignment problem complain that these depictions of AI have been counterproductive as they have given people unrealistic images of the potential threats, thereby making similar risks in real life seem implausible and causing the field of AI safety to be neglected. One way to to try to solve this problem might be to create fiction presenting some of the biggest threats posed by advances in AI (e.g. unaligned superintelligence, perpetual dystopia maintained using AI, etc.) in a very "realistic" manner. If stories could be written which reached a large audience and left a significant fraction of that audience believing that the threats presented in the story were serious issues in the real world then perhaps this could improve support for technical AI safety research or AI policy research.

Could such works successfully capture the imaginations of a large audience? This appears to me to be the biggest hurdle facing a project of this kind. On the one hand, it appears that there is significant public appetite for fiction based on existential risks. At the extreme, Terminator 2 grossed about $520 million in box office figures, presumably meaning that tens of millions of people have seen the film and many more have been influenced by it. Similarly, Don't Look Up recently gained a lot of traction by satirising indifference towards climate change. However, the most famous examples of course tend to be those which were the most commercially successful and for every successful attempt, there are presumably many examples that few have heard of. Moreover, constraining the work by insisting that the plot be realistic may reduce the likelihood that it reaches a large audience. For example, it might be much harder to write a compelling and entertaining story with this restriction in place.

Even if fiction of this kind were consumed by a lot of people, would many of them come to take AI risks more seriously? It has been argued, for example, that the Terminator films do not in fact portray AI risks unrealistically and that if viewers were left thinking that it's silly to be concerned about AI then that wasn't because the film depicted the threats inaccurately. However, even if that is true, it might be possible to carefully design a story which makes the issues clearer to the audience than the Terminator films and other mass media have managed to, although this would constrain the fiction even further, potentially making it less likely to reach a large audience.

Finally, even if someone created fiction which made a large number of people take AI risk more seriously, would this help much to reduce AI risk? I'm pretty sure the answer to this question is yes since it would broaden the pool of talented people who would consider working to mitigate these threats and might otherwise increase support for relevant research but whether or not this would make enough of a difference to justify the resources that would be required by such a project is of course questionable.

Overall, I think it is very unlikely that an endeavour to produce fiction realistically depicting AI risks would justify the associated opportunity cost because it would require a lot of time and money and would be unlikely to have much of an effect for the reasons mentioned above. However, I think there is a small chance that such a project could be very successful. I am writing this on the off chance that others have good ideas on if/how this could actually be made to work so I would welcome any comments.

26

New Comment
11 comments, sorted by Click to highlight new comments since: Today at 8:04 PM

One data point here (I'm unsure how much weight to give it, probably not very much) is the 1983 movie The Day After, which is about the aftermath of nuclear war.

He [President Reagan] wrote in his diary that the film [The Day After] was "very effective and left me greatly depressed", and that it changed his mind on the prevailing policy on a "nuclear war". The film was also screened for the Joint Chiefs of Staff. A government advisor who attended the screening, a friend of Meyer's, told him: "If you wanted to draw blood, you did it. Those guys sat there like they were turned to stone." In 1987, Reagan and Soviet leader Mikhail Gorbachev signed the Intermediate-Range Nuclear Forces Treaty, which resulted in the banning and reducing of their nuclear arsenal. In Reagan's memoirs, he drew a direct line from the film to the signing.

(Wikipedia; emphasis added)

I think this Wikipedia claim is from Reagan's autobiography. But according to The Dead Hand, written by a third-party historian, Reagan was already very concerned about nuclear war by this time, and had been at least since his campaign in 1980. It's pretty interesting — apparently this concern led to both his interest in nuclear weapon abolition (which he mostly didn't talk about) and in his unrealistic and harmful missile defense plans.

So according to this book, The Day After wasn't actually any kind of turning point.

and in his unrealistic and harmful missile defense plans.

 

Some people argue that Reagan's "Star Wars" missile defense plan did succeed at it's real goal, convincing the Soviets that they would never be able to keep up with America's R&D, so it was better to make peace.   From this point of view, whether SDI was realistic or not misses the point, that Reagan succeeded in creating the impression that it was realistic.   That is, from this point of view, he succeeded in bluffing the Soviets.  

Other factors contributed, like the collapsing Soviet economy, Russian defeat in Afghanistan, etc.  SDI was a kind of economic war, we can out spend you etc.

Someone posted a short story along these lines to the EA Forum or maybe LessWrong a few months ago. I can't find it—does anyone remember the link?

From memory: a programmer pressed enter and let things run over night... the AI reached some threshhold, then quickly hacked its way onto the internet and computers around the world, then started getting up to no good...

I think you're referring to "It Looks Like You're Trying To Take Over The World" by Gwern: https://www.lesswrong.com/posts/a5e9arCnbDac9Doig/it-looks-like-you-re-trying-to-take-over-the-world

Note that in context of OP's original question, this story demonstrates that discussion of realistic depiction may increase chances of risk!

Gwern elaborates on this idea here .

That's it, thank you!

  1. The probability of any one story being "successful" is very low, and basically up to luck, though connections to people with the power to move stories (ex. publishers, directors) would significantly help. 
  2. Most ex-risk scenarios are perfect material for compelling and entertaining stories. They tap into common tropes (hubris of humans and scientists), are near-future disaster scenarios, and can have opposed hawk and dove characters. I imagine that a successful ex-risk movie could have a narrative shaped like Jurassic Park or The Day After Tomorrow.
  3. My actionable advice is that EA writers and potential EA writers should write EA fiction alongside their other fiction and we should explore connections with publishers.

    As a side-note, I wrote an AI-escapes-the-box story the other week, and have since used Midjourney to illustrate it, as is fitting: https://twitter.com/Ideopunk/status/1553003805091979265. If anybody would like to read the first draft, message me! 

One data point here (I'm unsure how much weight to give it, probably not very much) is the 1983 movie The Day After, which is about the aftermath of nuclear war.

Yes, yes, yes and yes.  That's a great example.   That was an amazing movie.   I'm a nuclear weapons junkie, and I couldn't even finish the movie that last time I watched it a couple of years ago.   Seen it ten times, freaks me out every time.

I think this is the full movie on YouTube:   

Finally, even if someone created fiction which made a large number of people take AI risk more seriously, would this help much to reduce AI risk?

 

Fiction, articles, books etc all exist in the realm of abstraction, which is typically a pretty weak medium for persuasion.

What might work better is an ongoing documentary series which highlights real world problems with AI as they unfold.   You know, if you were to obtain footage of a driverless semi truck crashing in to a family car, some things like that.  

Or, just sit back, wait for these things to happen, and watch them on the news.  Let existing media do the work, they love drama and tragedy etc.

One way to to try to solve this problem might be to create fiction presenting some of the biggest threats posed by advances in AI (e.g. unaligned superintelligence, perpetual dystopia maintained using AI, etc.) in a very "realistic" manner.

 

If  you want to further demolish the audience for your art project, change the subject to the knowledge explosion machinery which is generating AI, and all the other emerging technological threats.  Instead of talking about symptoms like AI, talk about root causes, like the knowledge explosion. 

 People will hate it, your investors will pull out, the box office will collapse, the reviews will be ruthless, your career as a film maker will be over.  :-)

Seriously now, AI doesn't matter.   If AI were to magically vanish and all AI threats were thus removed....

The knowledge explosion will just keep whirring along, putting out ever larger benefits and threats, at an ever faster pace.   If you want to make a realistic monster movie, that could be it.