Gaming

Victories in AI games aren’t just for machines

Victories in AI games aren't just for machines

Shortly after dropping the entire human race in October, Takuma Miyazono explained his defeat. The artificial intelligence of his opponent, said the defeated world champion of the motor racing simulator Gran Turismo, had been surprisingly aggressive. The AI ​​driver, built by Sony engineers, had risked moves no human would contemplate.

If Miyazono’s apology sounds familiar, that’s because it is. The past quarter-century has been punctuated with times when some champions of Homo sapiens was forced to yield — in the realms of checkers, chess, and go — to the superior skills of a machine. Still, there’s a note of amazement at how innovative the AI ​​was to secure victory.

But the crash of Miyazono and his teammates, which was announced last week by Sony to coincide with the publication of a report in Nature, belongs to a slightly different category. For gamers like me, this has far more exciting implications. While those previous clashes were staged on game boards that long predated the machines brought in to humble us, this was fought on something much closer to the machines’ home ground.

Sony’s artificial intelligence agent, known as GT Sophy, was designed to beat, through tens of thousands of hours of learning, the world’s best human competitors in a computer game of extraordinary but, critically, devoid of trickery. Game makers have always been able to defeat a human by cheating (they can simply rig the opponent to be incredibly faster or stronger): Sophy had to outplay humans with only technique, an assessment of risk and, most breathtaking of all, panache.

I found this personally exciting because of how I’ve always played and thought about games throughout my 40 year love affair with them. Specifically, the single player type where you play to beat the software. Where others may have been content with the idea of ​​impersonal conflict, I always imagined myself playing the creators of the games – the once-mysterious (often Japanese) names that appeared in the final credits. I imagine these people, on whose addictive altars I have sacrificed thousands of hours, furious at having killed their ghouls, solved their puzzles or beaten their left backs. After meeting many of them in person, I realize that they are actually thrilled.

Counterintuitively, even though games have become the product of huge teams of developers, I’ve always found the tendency to humanize them intensified by large human vs. machine showdowns.

When these defeats have been covered as news events, the tone and analysis has generally been alarmist: critical human strongholds falling into the hands of an ever-improving mechanical enemy who knows what a dystopia that foreshadows. ? When Garry Kasparov lost to IBM’s Deep Blue supercomputer in 1997, the lingering image was of the bewildered grandmaster rising from the table and, with the wound to human pride seemingly too vast for words, shrugging his shoulders. to his mother in the audience.

Nearly 20 years later, cameras have captured every grimace of 18-time world Go champion Lee Sedol’s 2016 defeat by Google’s AlphaGo program. His protracted, ever-reviewable descent into disbelief has been lazily presented by some as a concession of humanity that should piss us all off.

But such interpretations are undermined by a constant characteristic of these defining moments: the flesh-and-blood reactions of the engineers. Instantly visible from the live footage of the programming teams during the tournaments, and from several excellent documentaries produced afterwards, one sees how intensely emotional and human the “machine” sides of these man versus machine clashes are.

Kasparov, Lee and now Miyazono have adopted computers designed, programmed and trained by people who have dedicated decades to the effort. Along the way, these engineers have been consumed by disappointment, frustration, tension, self-doubt, and a desire for victory just as powerful as the demons that propel grandmasters to greatness.

The joy for me is twofold. The future offspring of Sophy’s algorithm, according to Sony, will eventually populate more and more games: when I play them, I will now have very specific people in mind when engaging in their creations. But even more glorious is the psychological step forward it takes the industry: towards a place where gamers like me can compete against true champions, rather than cheaters.

Leo Lewis is the FT’s commercial publisher for Asia. Janan Ganesh is missing

Best of the FT weekend

The Premier League at 30, an English success story?

Since top clubs parted company with their peers in 1992, foreign players and money have flowed in

Inside Peloton’s Epic Series of Missed Calls and Bad Luck

The American exercise company has found itself the darling of investors and clients. What went wrong?

To follow @ftweekend on Twitter to hear our latest stories first