Doomsday stories about rogue artificial intelligence focus on the ‘singularity’, an inflexion point at which humans perceive an existential threat because the AI has surpassed their cognitive powers, and they fear the loss of control over themselves and their future. This usually happens in other-worldly situations—in The Terminator, the crisis is in the future, and in 2001: A Space Odyssey, it’s in interplanetary space. Real-world disaster scenarios have AIs taking over nuclear facilities, tripping national power grids, and hijacking telecom systems and satellites using means that brush aside human controls. But who could have imagined that Hollywood writers, actors and people working in broadcast media would be the first to experience the singularity? It’s very appropriate, though, since they are the very people who have brought AI dystopias to life for us.
The US film industry is in the midst of a historic strike led by the Writers Guild of America and the Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA). Its effects are being seen in the biggest venues—the cast of Oppenheimer walked out of the London premiere—and in niche events like the San Diego Comic-Con, which Hollywood A-listers skipped this time. Some of the strikers are no doubt driven by Luddite futility. Movies and media have become hugely technology-dependent, and artists will be no more successful in holding off the newest wave than Canute was, even if it threatens their livelihoods. But they should be successful in negotiating better terms for how the industry appropriates their talent because right now, it’s like the Wild West.
Performing artists are particularly vulnerable because apart from their cultural products, their voices, their bodies, and their very selves are easily appropriated by AIs. In March, an AI researcher used a machine to sing a verse in the manner and voice of Kanye West, and some critics said that it was better than the rapper himself, which must have been pretty crushing. Industry specialists hazard that within a couple of years, every popular singer will have several AI models trained in their work—a Taylor Swift studio version, stadium version, and stadium in the rain version (she does sing in the rain). Anyone could write a song and cut a track in any version of her voice. Without legal curbs on machine-made clones of human work, they would be free to exploit it commercially. Upon which, Ms Swift would have three choices: One, stop expecting revenue from streaming and the sale of albums and earn mainly from concerts like the Grateful Dead did. Two, spend the rest of her life suing people globally, which is a depressing choice. Or three, go on strike to seek terms under which she is properly compensated for third parties using her capital.
The writers and artists on strike in America have chosen option three. While singers can lose control of their voices to AIs, writers can be almost totally removed from the picture, and actors can have their entire bodies taken over. These people have rebelled because they face an existential crisis due to studios and owners of production facilities not clarifying how they are to live with AI—or will they not survive at all?
While the Hollywood strike has all the eyeballs because it’s Hollywood, Meta has quietly upended the AI industry by open-sourcing its AI Llama-2—you can install it on your laptop and take it offline. Open-sourcing is a smart step. Social media is buzzing with AI ‘experts’ trying to sell you an interface or some prompts—verbal instructions to get an AI to do what you want—because they know the gold rush won’t last. Artificial general intelligence (a machine you can talk to) will become the default gateway to the internet. That would make it critical infrastructure, and governments would have to guarantee public access to it. AI will be democratised, as Meta anticipates, and will grow faster because of it.
Following the open-sourcing of Llama-2, Nick Clegg, former UK deputy prime minister and president for global affairs at Meta, has been talking up its advantages in India, which is Meta’s biggest market. It is one way of allaying anxieties about AI. An AI can go rogue only via an unforeseen outcome—for instance, if programmers and trainers have not anticipated an unwanted activity. For instance, it is a good policy to tell chatbots to reply to queries only as an assistant and never as a system user. It prevents machines from playing boss. But unexpected issues can also lurk in the source code of the machine. If the code is open to inspection and manipulation by thousands of programmers and researchers everywhere, the issues are much likelier to be caught and patched than by a handful of bug-chasers employed by a corporation.
The most visible problem with AI chatbots is what the industry calls ‘hallucinations’—like propagandists, machines make up facts to substantiate their thinking and even provide Borgesian references to books which do not exist. Open-sourcing is likely to contain such issues faster, leaving humans to deal with the internet’s real problem, which mass access to AIs is bound to increase—fake news created by humans themselves. The problem is never technology, which is value-neutral. The problem is the rogues at the console, who use technology like Archimedes’ lever to multiply their destructive power.
Pratik Kanjilal
The India CableEditor of
(Tweets @pratik_k)