Learning to Let Go

AI’s biggest lesson for 2024.

When we launched this newsletter in October, I naively planned several months’ worth of topics so that I would know what I was writing about ahead of time. Nice try. In reality, AI has moved so quickly, I had a new (and unexpected) subject to cover every week. So if there is one takeaway from 2023, I’m going to say that AI is here to stay — and that it’s only going to move faster in the year ahead.

The topics that I’ve been most focused on have been regulation (the launch of ChatGPT lit a flame under the EU and US to finally get proposals over the line), building the technology in different ways (such as considering the 30% of the world’s population that doesn’t have a voice in development and regulation), and re-engaging academics in long-term thinking, and not demos.  

Instead of looking backwards with a “year in review,” I’m thinking about figuring out how we can learn from AI moving forward. One of the best examples is how the Go community reacted to AlphaGo, the program introduced by DeepMind Technologies in 2015. For those of you who don’t know, Go is an ancient Chinese board game that is notoriously difficult to master. (Think of it this way: Chess has about 10⁵⁰ different game states. That’s a 10 with 50 zeros after it. Go, incredibly, has 10¹⁷⁰ different game states. To give you a sense of scale: there are only estimated to be 10⁸⁰ atoms in the observable universe…) That said, it’s not only hard for a human to play, the strategic complexity has always historically been tough, even for a computer to master.

That’s why it made news when AlphaGo began beating not only other Go programs but became the first to beat a Go master without a handicap. In 2016, during a match with 18-time world champion Lee Sedol, over 200 million viewers watched as AlphaGo won 4-1. The most notable and surprising part of the game? AlphaGo played a wildly creative move, now known as “Move 37” – which surprised everybody. One commentator even said, “That’s a very strange move.” and, “I thought it was a mistake.” The odds of playing that particular move was only 1 in 10,000 – but, probably because of that move, the program won. 

By training itself first against amateurs to learn how humans play, then against versions of itself thousands of times to learn from its errors, the program became a champion-beater a decade sooner than DeepMind had predicted. It would have been easy for humans to just throw in the towel and say that we’re done with Go – it has been solved – no need to play it any more. Instead, something different happened.

The international Go community has reacted by studying these games. Studying how “creativity” came into play with Move 37. (There was even a documentary made about it.) Rather than reject AI, they have embraced the insights that can be learned from the games. This has led to a deeper exploration of the game and even changed how humans play it. When Sedol retired from professional play in 2019, he said, “Even if I become the number one, there is an entity that cannot be defeated.”

AlphaGo has inspired awe and renewed interest in advancing Go strategies. I think we should take the same approach to AI in general: Drop the reluctance and use it to deepen our knowledge, shake up our creativity, and inform our strategic approach. Always be learning.

Happy New Year, and see you in 2024.

---

Worth the Read

Let the facial-recognition fails begin — in this case, paired with meaningful policy implementation: The FTC banned Rite Aid from using facial-recognition software for five years after it falsely tagged women and people of color as shoplifters in hundreds of stores. The director of the FTC’s Bureau of Consumer Protection, said, “Today’s groundbreaking order makes clear that the Commission will be vigilant in protecting the public from unfair biometric surveillance and unfair data security practices.” 

The FTC had a big week: On Wednesday, it proposed legislation that would put significant guardrails around children’s data on gaming platforms, social media apps, toy retailers and more, effectively “shift[ing] the burden” of online safety from parents to apps. My prediction? We’ll be talking about this a lot more in 2024. (You can read my take here.)

Another landmark: The UK Supreme Court declared that a British computer scientist could not register patents for discoveries made by his AI “creativity machine,” DABUS. The court unanimously rejected his appeal, writing that, under UK patent law, "an inventor must be a natural person."

The Pope has called for an international treaty regulating AI. Reminds me of the fun conversation I had with Bishop Paul Tighe. To not focus on a Christian view, there is a lot that we can learn for AI ethics from Buddhism.

Did you want a year in review? Here’s a few. HAI, Stanford’s Human-Centered Artificial Intelligence department, weighs in with the 13 biggest AI stories of the year. My favorite? “The 2023 State of AI in 14 Charts.” Or maybe it’s the one about the dance animator that can choreograph moves to any music? And The Project Liberty newsletter wraps up the year by asking experts in its global alliance to reflect upon the year in responsible tech and make predictions for the year ahead. Some fun insights here. Finally, Axios did a review in their newsletter, with a few things to watch, such as the implications on immigration policy, how is open source vs. closed source is going to play out, and whether the doomsayers or the realists are going to win hearts and minds.