What can blockchain do for AI? Not what you’ve heard.
Industries all over the place are asking “What can AI do for us?”
However the blockchain trade, identified for difficult norms, can be asking the other query: “What can blockchain do for AI?”
Whereas there are some compelling solutions, three narratives have emerged round this query which are often deceptive and, in a single case, doubtlessly even hazardous.
Narrative #1: Blockchain can fight misinformation attributable to generative AI
An professional panel at a current Coinbase occasion concluded that “blockchain can counter misinformation with cryptographic digital signatures and timestamps, making it clear what’s genuine and what’s been manipulated.”
That is true solely in a really slim sense.
Blockchains can document digital-media creation in a tamper-proof means, i.e., in order that modification of particular pictures is detectible. However it is a far cry from clarifying authenticity.
Contemplate a photograph of a flying saucer hovering above the Washington Monument. Suppose that somebody has registered its creation in, say, block 20,000,000 of the Ethereum blockchain. This truth tells you one factor: The flying saucer picture was created earlier than block 20,000,000. Moreover, whoever posted the picture to the blockchain — let’s name her Alice — did so by digitally signing a transaction. Assuming that Alice’s signing key wasn’t stolen, it’s clear that Alice registered the picture on the blockchain.
None of this, nevertheless, tells you how the picture was created. It may be a photograph that Alice snapped along with her personal digicam. Or Alice might need gotten the picture from Bob, who Photoshopped it. Or possibly Carol created it with a generative AI device. In brief, the blockchain tells you nothing about whether or not aliens had been touring Washington, D.C.—except you already belief Alice to start with.
Some cameras can digitally signal photographs to authenticate them (assuming their sensors can’t be fooled, which is an enormous if), however this isn’t blockchain expertise.
Narrative #2: Blockchain can carry privateness to AI
Mannequin coaching is a node=”[object Object]” goal=”_blank” rel=”nofollow”>trumpeting blockchain applied sciences as an answer.
Blockchains, nevertheless, are designed for transparency — a property at odds with confidentiality.
Proponents level to privacy-enhancing applied sciences superior by the blockchain trade to handle this pressure — particularly zero-knowledge proofs. Zero-knowledge proofs, nevertheless, don’t resolve the issue of privateness in AI mannequin coaching. That’s as a result of a zero-knowledge proof doesn’t conceal secrets and techniques from whoever is developing the proof. Zero-knowledge proofs are useful if I need to conceal my transaction information from you. However they don’t allow me to compute privately over your information.
There are different, extra related cryptographic and safety instruments with esoteric names, together with absolutely homomorphic encryption (FHE), safe multiparty computation (MPC) and safe enclaves. These can in precept assist privacy-preserving AI (particularly, “federated studying”). Every has essential caveats, although. And claiming them as blockchain-specific applied sciences can be a stretch.
Narrative #3: Blockchains can empower AI bots with cash — and that’s a great factor
Jeremy Allaire, CEO of Circle, has famous that bots are already performing transactions utilizing cryptocurrency and tweeted that “AI and Blockchains are made for one another.” That is true within the sense that cryptocurrency is an effective match for the capabilities of AI brokers. But it surely’s additionally worrisome.
Many individuals fret about AI brokers escaping human management. Traditional nightmare situations contain autonomous autos killing folks or AI-powered autonomous weapons going rogue. However there’s one other vector of escape: The monetary system. Cash equals energy. Give that energy to an AI agent and it may possibly do actual injury.
This downside is the subject of a analysis paper that I co-authored in 2015/6. My colleagues and I examined the potential of good contracts, applications that autonomously intermediate transactions on Ethereum, getting used to facilitate crime. Utilizing the strategies in that paper and a blockchain oracle system with entry to LLMs (Giant Language Fashions) equivalent to ChatGPT, unhealthy actors may in precept launch “rogue” good contracts that routinely pay bounties for committing critical crimes.
Learn extra from our opinion part: How a wise contract will get away with homicide: A evaluation of ‘The Oracle’
Fortunately, rogue good contracts of this type aren’t but potential in right this moment’s blockchains — however the blockchain trade and crypto lovers might want to take AI security critically as a future concern. They might want to contemplate mitigations, equivalent to community-driven interventions or guardrails in oracles to assist implement AI security.
The mixing of blockchains and AI does maintain clear promise. AI could add unprecedented flexibility to blockchain programs by creating pure language interfaces to them. Blockchains could present new monetary and transparency frameworks for mannequin coaching and information sourcing and put the ability of AI within the arms of communities, not simply enterprises.
It’s nonetheless early days, although, and as we wax lyrical about AI and blockchain as an attractive mixture of buzzwords and applied sciences, we have to actually assume — and see — issues by.
Ari Juels is the Weill Household Basis and Joan and Sanford I. Weill Professor within the Jacobs Technion-Cornell Institute at Cornell Tech and the Technion and a Laptop Science school member at Cornell College. He’s a Co-Director of the Initiative for CryptoCurrencies and Contracts (IC3). He’s additionally Chief Scientist at Chainlink Labs. He’s the creator of crypto thriller novel The Oracle (Talos Press), which was launched on 20 February 2024.
