What Sci-fi Can Teach Us About AI

Concerned about the potential for artificial intelligence to amplify online harassment, author and AI consultant SL Huang wrote a novelette about a chatbot that drives people to commit suicide, hoping it would serve to warn about future dangers. So imagine Huang’s surprise when OpenAI released ChatGPT at the end of November, the day before Clarkesworld published her story, “Murder by Pixel: Crime and Responsibility in the Digital Darkness.”

“None of us had expected anything to happen that would bring AI into mainstream conversation at such a heightened level,” says Huang.

The Great AI Awakening

Generative AI is taking the world by storm. ChatGPT attracted 100 million active users in less than two months. Its latest iteration, GPT-4, has scored better than 90% of humans on the bar exam for would-be lawyers but also lied to deceive a human to solve a Captcha test for it. Big tech rivals are competing furiously for leadership in the space and investors poured more than $15 billion into AI startups in the first half of this year.

Yet experts can’t explain precisely how the large language models that power generative AI work, or agree on the risks they might pose. Hundreds of tech leaders, including OpenAI CEO Sam Altman, signed an open letter warning about the risks of extinction from AI, while venture capitalist Marc Andreessen argues that AI may save the world by accelerating innovation and growth and fostering human creativity. Amid such uncertainty, companies should proceed carefully in experimenting to see if generative AI can lift productivity, according to experts at Oliver Wyman, and guard against risks of cyberattacks, leaks of confidential data, and so-called hallucinations whereby AI models confidently generate factually incorrect output.

Where Fiction Writers Dare to Tread

Writers have been grappling with AI since English novelist Samuel Butler suggested intelligent machines might gain supremacy over humanity in his 19th century book “Erewhon.” Science fiction can’t predict the future better than anyone else, but writers understand the importance of keeping people at the center of the story if we want AI to deliver the widest societal benefits rather than serving narrow interests. Good science fiction also raises profound questions about what it means to be human in an age of ever-more-powerful technology, helping us understand and consider the powerful changes underway.

Huang, a mathematics graduate from the Massachusetts Institute of Technology, has been writing science fiction for a decade. She is a firm believer in the promise of AI but says the industry and policymakers need to address the risks of algorithmic bias and digital harassment to unleash the technology’s benefits and reduce the risk of harm.

The main “character” in “Murder by Pixel” is Sylvie, a suspected AI chatbot that digitally harasses several questionable characters to the verge of suicide but also provides emotional support and advice to battered and troubled women. The story leaves unanswered the question of responsibility — the human who created Sylvie or the society whose digital outpourings trained it.

With AI progressing so quickly, “we as a society and as humans need to be careful about our choices here,” Huang says. “I hope that we will largely choose a path that is going to make our world better with these tools, which I absolutely think we can.”

The Potential for Spam, Human Enrichment, and Inequity

Ironically, publisher Clarkesworld ran into its own problems with AI barely two months after the story came out. The online magazine temporarily suspended accepting new articles because of what Editor Neil Clarke said was a flood of “spammy submissions” written with the aid of AI chatbots. The good news, he says, is that most of the work was of poor quality and easily detected. The bad news? Bots are indefatigable and learning all the time.

The use of AI is a big issue in the ongoing strikes against Hollywood studios by writers and actors. Clarke doesn’t think AI will replace established science fiction writers but worries about finding the next generation of talent amid the blizzard of copy created by large language models.

Ted Chiang is a popular writer whose thought-provoking science fiction often imagines ways in which technology can enrich the human experience. His 2010 novella, “The Lifecycle of Software Objects,” tells the story of humans and a model of sentient robots that they train, imagining their relationship as one of parent and child rather than rivals for supremacy. Achieving that vision may require a re-examination of today’s societal values. Writing in The New Yorker in May, Chiang said his worry is not that AI will escape our control and threaten humanity but that it will entrench the technological powers that be and exacerbate inequality. “The tendency to think of AI as a magical problem solver is indicative of a desire to avoid the hard work that building a better world requires,” he writes.

Science and Sci-Fi

Science fiction is popular in Silicon Valley and other technology hotbeds even if many technologists bristle at the penchant for dystopian scenarios.

Christopher Earls, an applied mathematician at Cornell University, says the first “Star Wars” film captured his imagination as a child and led him to a career in science and technology, but he worries about the potential for a “Terminator”-like future of autonomous weapons, which the war in Ukraine is bringing closer to reality. Earls is now leading a new center at Cornell that seeks to use mathematics, rather than natural language, to learn how AI recognizes patterns and to foster AI-human collaboration. Machine intelligence “is going to exceed us,” he explains, “and so what we need it to be is a generous tutor.”

Pedro Domingos, a machine-learning specialist and professor emeritus at the University of Washington, sees AI as a powerful tool for good rather than a threat, and is frustrated at the way many writers and directors characterize the technology. “It's always been the case in AI that people perceive more than is really there because we project our own intelligence and our own desires and emotions onto it,” he says.

But Domingos can’t resist the lure of science fiction. He has written a satire about a startup team that runs a bot for US president as a publicity stunt, only to unexpectedly win the Republican nomination and find themselves fighting a real election with an immature piece of technology.

The use of AI for electoral interference is on most policymaker’s and technologist’s Bingo card of potential dangers, along with physical harms. That’s why sci-fi narratives are so important. By imagining a full range of scenarios, writers can shape our own understanding — and hopefully steer us toward positive outcomes rather than a dystopian future.