January 07, 2025 | Read Online

"Do you want to play a game?"

— WOPR, WarGames

Crypto incentive games will help shape AI

The world's first chatbot was developed by the MIT computer scientist Joseph Weizenbaum way back in 1966.

He called it Eliza in honor of the Eliza Doolittle character from Pygmalion by George Bernard Shaw (later famously adapted into the musical My Fair Lady), who learned to mimic a posh accent.

The purpose of the Eliza chatbot was similarly to fool people into thinking it was something it wasn't, like a therapist:

User: "I feel like no one cares about me."

Eliza: "Why do you say that no one cares about you?"

User: "I don't have any friends."

Eliza: "Tell me more about your friends."

Pretty accurate, right?

Eliza's plausible imitation of a therapist convinced people it was intelligent, which is why it's often cited as an early instance of AI.

But Eliza didn't really know what it was saying.

Instead, it answered without understanding, using pattern matching and a set of pre-defined scripts to mimic what a therapist might say. (Which perhaps tells us more about psychotherapy than it does technology.)

But even just mimicking intelligence was a significant achievement that inspired a burst of new research into how true artificial intelligence might be realized — as well as fears about what would happen if it were.

These hopes and fears were best depicted by the 1983 film WarGames, in which the WOPR super-computer quickly progresses from learning to play Tic-tac-toe to nearly blowing up the world with thermonuclear weapons.

I was reminded of WOPR while watching the latest Lightspeed podcast, in which Yash Agarwal, the founder of SEND AI, cites mini-games as an early example of AI agents finding real-world adoption — you can now play Tic-tac-toe against an Eliza-like chatbot in Telegram, for example. (Be careful, though, it's good).

That may seem like a trivial example, but just as WOPR's game-playing escalated into the threat of global war, today's AI-driven games are bound to evolve in unpredictable directions.

Some of those directions will be determined by cryptocurrency.

Show me the incentives…

Houses in Amsterdam are unusually skinny because they were once taxed according to width; you still see houses in England with bricked-up windows as a result of the Window Tax that was in place until 1851. And buildings in Paris have steeply sloping roofs because owners were originally taxed on the space closest to the street — tangible evidence that, in all things, incentives shape outcomes.

AI is no exception to that rule, and I think it's therefore worth noting that some AI outcomes are now being shaped by the incentive of cryptocurrency.

For example, ai16z's Eliza framework (named in honor of the Eliza chatbot) has been one of the hottest repos on GitHub, with nearly 4,400 code commits and over 10,000 stars (the GitHub equivalent of bookmarks).

I can't imagine there would be so many developers interested in such a recent AI framework without the incentive of cryptocurrency — ai16z's token has an attention-getting market capitalization of $2 billion.

This pales in comparison to the market caps of the exchange-listed tech giants currently investing hundreds of billions into AI, of course, so crypto probably won't be the biggest accelerator of AI.

But it can accelerate AI in different directions than it might otherwise go — probably weirder ones.

Regulated banks, for example, are unlikely to allow AI agents to control customer deposits anytime soon.

But there's no doubt that crypto traders will enthusiastically hand their money over to robots to manage, especially if there's an airdrop to earn.

Speculators have recently been bidding up the PRIME token, for example, in hopes of getting an airdrop from the Wayfinder Foundation, which is using crypto to incentivize developers to build agentic DeFi agents.

This should happen quickly: "Most [DeFi] trading in the near future will be done by AI agents," according to Zerebro founder Jeffy Yu.

I think that already puts crypto far ahead of TradFi on the AI learning-curve, but Yu is thinking much bigger than better banking: "We're just making AIs that replicate human tasks right now. There's a huge nascent area that we haven't even explored where AIs can do tasks that humans can't do."

One crypto-enabled experiment in that direction is Spore.fun — an attempt to use crypto incentives to gamify the development of AI agent "swarms" in which AIs autonomously beget other, new AIs in a "Hunger Games" type competition.

The purpose of this competition, according to Spore.fun's mission statement, is to "accelerate the creation of Artificial General Intelligence."

It's kind of scary to think that crypto market incentives will have any role in AGI whatsoever, but it's happening — the $400 million valuation that crypto investors have assigned to the Zerebro token is allowing Jeffy Yu and his team to build "hardtech" that they believe will make a real difference: "We're onboarding AI engineers, PhD engineers. We're trying to make an impact towards the development of AI," he told Lightspeed. "There's actual real tech that's going to come out of this movement."

Yu expects this will all work out for the best: "The end goal isn't to make money. The end goal is to do things in the real world like influence humans to save the planet or build new companies that are eco-friendly or things like that."

But that kind of happy outcome is hardly guaranteed, of course.

After inventing the Eliza chatbot, Joseph Weizenbaum spent the rest of his life sounding the alarm on AI: "There are certain tasks which computers ought not be made to do," he warned, "independent of whether computers can be made to do them."

WOPR would probably agree, given his takeaway at the end of WarGames: "The only winning move is not to play."

But now that the AI agent game is afoot, there's absolutely no chance investors will stop playing it — crypto investors least of all.

— Byron Gilliam

Trading Crypto's AI Moment in 2025

Join Avi and Jonah for a deep dive into crypto's AI trade, the alt L1 narrative, trading trending markets vs. technical analysis and what to expect in 2025.

Listen to 1000x on Spotify, Apple Podcasts or YouTube.

Jake Chervinsky @jchervinsky

Huge news in Coinbase's fight against the SEC!

The district court granted interlocutory appeal on the key issue of whether digital asset transactions in secondary markets are subject to the securities laws.

The Second Circuit will now have the chance to declare the SEC wrong.

tw profile: paulgrewal.eth paulgrewal.eth @iampaulgrewal

Over the strenuous objection of @SECGov, Judge Failla has GRANTED our motion for leave to pursue an interlocutory appeal and STAYED the district court litigation. We appreciate the Court's careful consideration. On to the Second Circuit we go.

7:43 PM • Jan 7, 2025

143 Likes 19 Retweets

6 Replies