OK Computer

Creativity in an age of generative artificial intelligence

Back in February 1996, midway through my first year as a grad student at the University of Pennsylvania, I left my high-rise apartment near the Philadelphia Museum of Art, wrapped in a warm coat and scarf, and faced into the bitingly cold wind. A month earlier, a colossal blizzard had dumped a record 31 inches of snow on the city within 24 hours, wreaking havoc. The mayor declared a state of emergency and banned regular traffic from the city while emergency work crews cleared the streets, dumping thousands of tons of snow off bridges into the Schuylkill River.

Abruptly and blissfully pedestrianized, the normally busy area around my apartment became a haven for kids, who spent days joyously throwing snowballs and building snowmen. Even weeks later, going anywhere on foot involved threading one’s way through narrow pathways between towering snowbanks, maneuvering around others hustling in the opposite direction. But on that day, I was excited to be out, despite the wind and cold—I was on my way to the Philadelphia Convention Center to witness the reigning world chess champion Garry Kasparov facing off against the IBM supercomputer Deep Blue.

How we came to see the board game of chess as the ultimate proving ground for human intelligence involves a fascinating story that Argentinian academic Diego Rasskin-Gutman admirably traces in his book Chess Metaphors: Artificial Intelligence and the Human Mind. He shows how chess became entrenched as a powerful cultural metaphor for superior cognitive ability, capacious memory, and developed spatial awareness — associations we see in recent productions such as Netflix’s widely acclaimed The Queen’s Gambit. But this same set of associations had also made chess a perfect metaphorical proxy as the United States and USSR vied for geopolitical supremacy during the Cold War.

By the early 1970s, the Soviets had comfortably dominated the elite echelons of chess throughout the post-war era, winning every World Chess Championship since 1948 — but when Bobby Fischer wrested the world title from Boris Spassky in Reykjavík, Iceland, in 1972, three years after American astronauts had become the first to walk on the moon, public interest in the game intensified across the Western world. In the year after Fischer’s win, membership of the US Chess Federation doubled and chess gained a new generation of impassioned devotees. We have continued to celebrate Fisher’s legacy in film, a Broadway musical, and even rap songs. But the status of chess as a culturally symbolic battlefield made it a key proving ground on another front, too, as technologists sought to challenge the primacy of the human mind as the locus of intelligent thought and even consciousness.

The concept of an automated chess-playing machine long predated the advent of modern electronic computing. In 1769, two hundred years before the moon landing, a Hungarian inventor named Wolfgang von Kempelen created a famous, albeit entirely fraudulent, chess automaton called the Turk to win the favor of Empress Maria Theresa of Austria. An intricate mechanical contraption and optical illusion, the Turk enabled a human chess player to hide within its innards and operate it remotely against unsuspecting rivals.

The Turk fooled crowds as it toured Europe and later the Americas for over 80 years, defeating notable opponents including Napoleon Bonaparte and Benjamin Franklin. Edgar Allan Poe helped expose the fraud in his 1836 essay “Maelzel’s Chess Player” (referencing Johann Nepomuk Maelzel, the German inventor and showman who bought the machine after von Kempelen’s death and toured the Americas with it), but the fantasy lived on of a chess-playing machine that could challenge the human monopoly on higher-order intelligence.

Following rapid advances in computing technology during World War II—including the 30-ton ENIAC built at Penn’s Moore School of Electrical Engineering in the early 1940s—the fantasy of a chess-playing automaton evolved into the idea of using these new computing behemoths as a chess-playing proving ground for nascent artificial intelligence. In 1950, the influential mathematician and computer scientist Claude Shannon published a paper in Philosophical Magazine titled “Programming a Computer for Playing Chess.” Arguing that a successful computerized chess machine would carry profound philosophical implications, he claimed that it would “force us either to admit the possibility of mechanized thinking or to further restrict our concept of ‘thinking’.” Shannon, then, sought to win a philosophical concession that much of what we do under the guise of “thinking” could in fact be replicated by a computer — or if this was not “thinking,” we needed a better philosophical definition.

Shannon proposed two potential approaches to mechanized chess playing, a “Type A” approach that would use brute-force computation to identify the best among all possible moves, or a “Type B” approach that would emulate human intelligence and endow a computer with understanding similar to that of an experienced human player. Today, the search for that “Type B” intelligence remains the Holy Grail for artificial intelligence. Computerized chess has been dominated by the “Type A” approach, boosted by exponential increases in computing power since Shannon’s time. The computers that put Armstrong and Aldrin on the moon in 1969 are significantly less powerful than the smartwatch on your wrist in 2024.

The mathematics of chess are dizzying. At the beginning of any game, White can choose among 20 opening moves—each pawn can move forward either one or two squares, and each knight has two possible moves. But Black can choose one of 20 responses to each possible opening move by White—meaning that after only one move by each player, the board can be in any one of 400 possible positions. After three moves by each player, the board can be in any of 119,060,324 positions. And after each player has made just five moves, close to 70 trillion positions are possible. Shannon calculated a conservative estimate of the number of possible positions in a 40-move chess game at 10¹²⁰. This impossibly large number—known as the Shannon number—dwarfs the estimated number of atoms in the known universe (10⁸⁰).

These mathematics make computerized chess challenging, even in an era of ever more powerful machines, because a pure brute-force approach that requires analyzing and ranking the outcomes of every possible move, including the obviously terrible ones, quickly becomes difficult to sustain. Human intelligence works differently: Our brains instinctively home in on obvious solutions to problems while filtering out tremendous amounts of noise, a trait that evolutionary biologists trace to our hunter-gatherer past. This capacity to filter noise long gave the best human players an indomitable edge over even the best chess-playing computers.

Human players don’t sift through billions of bad moves looking for the best ones; they typically consider only lines that look promising in terms of yielding a material or strategic advantage. This, it turns out, is a very difficult advantage for a computer to counter. Early computers could defeat novice to intermediate opponents simply by playing solidly, being indefatigable, and avoiding glaring blunders, but better players found chess computers predictable and thus easy to defeat. This made facing one of the most dominant world chess champions a seemingly insurmountable challenge for IBM.

On stage at the Philadelphia Conference Center, Kasparov, wearing a brown suit, sat staring intently at the board, often squirming around in his chair, sometimes holding his face in his hands. Across from him sat an IBM representative who tracked the computer’s moves on a monitor, replicated them on the physical board, and then inputted Kasparov’s responses. Mounted on either side of the chess clock were small flags of the United States and Russia—a reminder that despite the collapse of the USSR five years earlier, this encounter had a long geopolitical history.

Still just 32, Kasparov had been world chess champion for over a decade, having won the title at a record age of 22 in 1985. His reputation for obliterating opponents with aggressive, swashbuckling play had given him the same aura of invincibility in chess that Mike Tyson in his prime enjoyed in heavyweight boxing. Kasparov had handily dispatched the world’s previously most sophisticated chess computer, Deep Thought, in a two-game match seven years earlier. Many observers did not believe that the contest with Deep Blue would produce anything other than another decisive win for the Russian.

IBM had other ideas, having beefed up its hardware and programming talent considerably in the interim. It had hired the Carnegie Mellon graduate students behind Deep Thought and then built Deep Blue, a 2,800-pound supercomputer with custom-built chips that could analyze a then-unprecedented 200 million moves per second, or 12 billion moves per minute. These investments in programming talent and powerful hardware would prove decisive in turning the tide in favor of computerized chess.

On that wintry day in February 1996, a computer defeated a reigning world chess champion for the first time in history. Kasparov proved the stronger player in the Philadelphia series overall, winning 4 games to Deep Blue’s 2, but an even more powerful iteration of Deep Blue defeated him in New York the following year, 3½ to 2½. Ten days after Kasparov’s loss, Radiohead released their iconic album OK Computer, which seemed (albeit accidentally) to hail a new era. Indeed, this historic reversal of the man-machine dynamic in chess proved pivotal for the public’s understanding of artificial intelligence overall—even at a time when only around a third of Americans owned a home computer and only a quarter had Internet access. We all knew, even then, that the AI revolution was only getting started.

Following the 1997 rematch—as Vikram Jayanti details in his 2003 documentary Game Over—Kasparov publicly accused IBM of cheating, alleging that it had used a hidden cabal of human grandmasters to bolster Deep Blue’s strategic positioning during the games. Kasparov’s accusation centered on his conviction that Deep Blue had played several unpredictably brilliant moves that, in his view, simply could not have been foreseen by a computer. At the core of Kasparov’s claims (which he later partially retracted) lies a debate about whether brute-forcing solutions through computational power can ever reproduce the flashes of inspiration or genius that we associate with human intelligence. This debate has come back to the fore in recent years, as we now have to consider whether computers can replicate aspects of human thinking that fall outside the realm of brute force calculation.

When we conceived of the Story Rules Project, we imagined ultimately writing a book that would harness the emergent neuroscience of story to define actionable guidelines for writers seeking to make a meaningful social impact with their projects. Naturally, we initially assumed we would be writing for an audience of human writers. But when OpenAI launched ChatGPT in 2022—25 years after Deep Blue defeated Kasparov in New York—the power of generative artificial intelligence began making headlines. Students quickly proved that they could use this new technology to write passable high-school level essays, raising widespread concern about academic integrity. Generative AI also proved adept at taking tests—GPT-4.0 can score 1460 on the SAT, placing it in the 96th percentile of all test-takers—raising further concerns about the potential for cheating.

Analysts predict widespread impact on the job market, with Goldman Sachs estimating that generative AI could replace the equivalent of 300 million full-time jobs globally, with the biggest impact on people in mid-career, mid-ability white-collar fields. It is not difficult to envision how AI could make significant headway in fields such as coding, data analysis, accountancy, or paralegal work—but creative professionals, too, have become concerned that AI could pose a threat to their livelihoods. During last year’s strike, the Writers Guild of America sought to limit studios’ use of generative AI to produce and rewrite creative material. The Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) also sought to limit studios’ use of AI to simulate the likeness of human actors.

Because we anticipate that AI will eventually play a role in creative story development, the Story Rules Project now intends to address not only human storytellers but also programmers working in the nascent field of artificially intelligent creativity. Our burning question is now to articulate how writers can combine human creativity, artificial intelligence, and neuroscience in ways that exponentially advance the potential of impact-oriented storytelling. This is the first of a series of posts that will explore computational creativity, asking how people become creative, whether and how computers can also become creative, and what the future of creativity holds, especially as it pertains to writing, screenwriting, and filmmaking. More soon!