In August 2022, Daniel Shorr was scared. In his room in a gray Stanford dorm, the 23-year-old was scrolling through Twitter when he saw someone post a painterly illustration of a woman walking past a neon-lit storefront. It’s “beautiful,” he said wealth.
However, the art was not the work of a man. Its author—inspired by the “Los Angeles at night” prompt—was Stable diffusion, an artificial intelligence model that generates images from text. The artistry of the algorithm gave Shorr a “WTF moment.”
“Intelligent algorithms will become an increasingly large part of our world’s computing diet,” he wrote shortly thereafter in a note on his MacBook. “Cryptography is the only mechanism to hold these powerful algorithms accountable.”
Along with his freshman roommate, Ryan Cao, Shorr decided to put his Post-it notes into action. After completing their masters degrees at Stanford, he and Cao founded Modulus Labs, a startup focused on accountability for machine learning. And to keep the AI algorithms in check, they plan to use a technique that has recently taken crypto by storm: zero-knowledge proofs. So far they have raised $1.3 million and are organizing another round of funding for about $5 million.
“We know, we know — yeah, what if we brought AI to blockchain?” sounds like something a seven-year-old would dream up,” they recently wrote a medium contribution. “But in our experience, seven-year-olds can be surprisingly smart.”
Recent Stanford grads aren’t the only ones who believe zero-knowledge proofs are a validation of AI. You are part of a community of researchers and blockchain entrepreneurs who believe in merging this vibrant cryptographic technique with the field of machine learning. While the technological combination is immediately relevant to the crypto world, it could soon find applications beyond the blockchain.
“Especially as we begin to see in very ugly ways how easy it is to manipulate these machine intelligences,” Shorr said wealth“we will rely on the hard bronze of mathematics.”
Why verify AI?
In just a few months, ChatGPT, the AI language model developed by OpenAI, has become an integral part of national zeitgeist. A future in which AI algorithms, shall we say, diagnose cancer or trade billions on the stock market seems a lot less like science fiction.
“If you look at the rise of AI technology, the growth has been amazing.” Daniel Kana prospective assistant professor of computer science at the University of Illinois Urbana-Champaign, said wealth. “And I certainly think within five years we’re going to want transparency about a lot of these algorithms.”
Like Modulus Labs’ Shorr and Cao, Kang believes that as AI models become more powerful, they need more accountability. As an example, he said, if patients are currently admitted to hospitals and their care goes awry, they can request hospital records to review doctors’ decision-making. “But if you imagine a world where some diagnoses or even some decisions are made by, say, a medical model,” he said, “you probably want the same level of accountability.”
Or imagine a future where AI algorithms read job applicants’ CVs, decide who gets a speeding ticket, or decide whether someone goes to jail, he says Jason Morton, an associate professor of mathematics at Penn State. As deployment grows, so do people’s incentives to modify, influence, or even replace AI algorithms with ones that better serve their interests. Lay people need to have faith in their medical diagnoses and believe that parking tickets or – in the most dystopian sense – jail sentences were rightly handed out.
This is partly why Morton, who is on furlough to build his own startup, and Kang dedicated their time and research to combining zero-knowledge evidence with machine learning. “It’s a way of proving to everyone that whatever model they think works is what works,” Morton said wealth.
‘Own Secret Sauce’
Just as the AI has caused an unspeakable stir, Zero-knowledge proofs have generated their fair share of chatter in the world of crypto. “People in this room are very excited about it,” Morton said.
The cryptographic technique, first articulated by researchers in 1985, has two main advantages: privacy and “conciseness,” or the ability to prove something true without having to analyze every single statement. For example, an auditor can quickly verify that someone has correctly filed a tax return without seeing data from the return.
Cryptocurrencies like Zcash And Monero have used zero-knowledge proofs to obfuscate users’ transactions. More recently, programmers have used concise proofs — of which zero-knowledge proofs are a subset — for scalability.
Blockchains like Ethereum are slow, decentralized computers. The larger a program, the longer it takes for a blockchain to run it. To solve this, developers run programs off-chain or on private servers through concise proof. Then they move the proof to a blockchain, where one only has to solve the proof to verify that the code ran correctly, rather than checking every single line of the program.
Now startups are pushing to use zero-knowledge machine learning proofs, a fusion they call zkML.
Zero-knowledge evidence allows outside observers to prove that a company or developer used a promised AI algorithm. For example, OpenAI, the juggernaut that developed ChatGPT, can prove that its chatbot wrote a poem without revealing the algorithm’s “weights” or what an AI model learns after training on vast amounts of data.
Twitter Open-source their recommendation algorithm, but the weights remain hidden! How can we trust it? We show how to verify the Twitter algorithm with zkml!
📝 Blog post: https://t.co/Sjhkl7fjjK
GitHub: https://t.co/DwtT9uidFB
1/6 pic.twitter.com/RUmRRYzusR— Daniel Kang (@daniel_d_kang) April 17, 2023
Kang, an Illinois academic, recently demonstrated the above when he created a zero-knowledge proof to validate Twitter’s AI-powered recommendation algorithm that ranks the tweets that appear in a user’s timeline. While Elon Musk released the code behind the ranking algorithm, the embattled CEO of the social media site and Tesla has not released the weights of the AI model that powers it. “Twitter has a lot of really good reasons for not posting the weights,” Kang said wealth. “There is a proprietary secret sauce to begin with, but it also contains a lot of private data.”
But with zero-knowledge evidence, Kang can prove that Twitter’s ranking algorithm ran without manipulation. “Even today, companies can – and do – lie and accountants come under fire for it,” he said. “But with zero-knowledge evidence, examiners can actually say with certainty that Twitter, for example, did the right thing.”
Immediate applications
Kang’s task as a researcher is to “develop technology that will have an impact in five to ten years”. But there are entrepreneurs who believe that they can now build zkML-based companies.
One of the more immediate uses is scalability. It’s too expensive to run AI models on decentralized computing platforms like Ethereum, so developers need to run them off-chain, which goes against crypto’s ethos of transparency. To convince users that they used the right AI model, developers can use zero-knowledge proofs to demonstrate that they ran the right algorithm.
For example, Modulus Labs is collaborating AI arena, a video game where AI fighters learn from human players. People pit the fighters against each other with the victor winning the crypto. Because of the financial stakes, players must trust that their opponents have not unfairly manipulated or influenced the fighters. To give users that confidence, Modulus Labs develops custom zero-knowledge proofs to verify that the AI fighter trained by a player is the same one deployed in a fight.
“We strongly believe that this technology will ultimately be more than just crypto itself.”
Daniel Shorr, Modulus Labs
Venture capitalists are taking notice. “We are definitely interested in investing in this category because we believe it will be very important,” said Ali Yahya, general partner at venture capital firm a16z crypto wealth. (However, he stated that as an investor he was focused on applying zkML to crypto, not beyond.)
Worldcoin, an initiative founded by OpenAI’s Sam Altman, is also an influential player in the emerging space. The project that should convince billions Scan her iris to prove they aren’t robots, uses zero-knowledge proofs with machine learning as part of its iris-scanning technology. One of its developers has organized a Telegram group dedicated to this area, and its foundation intends to give grants to zkML projects and researchers. “This is a huge area of research for us, one that shows great promise,” said Steven Smith, head of protocol at Tools for Humanity, the company developing Worldcoin wealth.
And smaller companies are also snooping around on zkML. “Sometime in the next — whether today, six months, four months — it’s probably time for a deep tech investor to take an early position,” said Tom Walton-Pocock, founder of deep tech VC firm Geometry wealth.
Making zero-knowledge proofs a daily part of our AI verification diet has a number of logistical hurdles, including the magnitude of the computational power required to generate proofs for complex AI algorithms and the mechanics of how the proofs grow leaked to a layman who can search Twitter or Bluesky or which social media platform will implement the technology first. Additionally, Kang says they’re not a panacea and AI accountability is “multi-pronged.”[ed] Approach.”
But for Shorr and Cao, who work at a Silicon Valley WeWork with whiteboards covered in equations, sticky notes and charts, they’re taking a calculated risk.
“We really believe in it,” Shorr said wealth“that eventually this technology will transcend cryptography itself.”
—————————————————-
Source link