DFINITY Founder and Chief Scientist Dominic Williams recently had a fireside chat with Tracy Trachsler, who joined DFINITY in May as Head of Institutional Relations.
The topic was artificial intelligence, specifically what Dom thinks about the current discussion around the intersection of blockchain and AI, what the Internet Computer has achieved so far, and what to expect next. Below is a summary of the conversation.
The GPU Dilemma
One of the most important decisions when running AI models is where to host them, and in the past few years, user-adopted large language models (LLMs), such as OpenAI’s ChatGPT and Meta’s Llama 3, have run on servers controlled by their developers.
As we explore in DeAI: Uncovering the Black Box Problem of Artificial Intelligence, the problem with relying on centralized servers is that they are vulnerable to cyberattacks, such as lurking agents, which can be programmed to produce inaccurate outputs in certain situations.
At DFINITY, we believe that AI on the blockchain is critical to ensuring society fully benefits from this game-changing technology, but not all decentralized models are created equal.
One approach that is gaining traction is using GPUs, which are required for training models and running inference, from a decentralized physical infrastructure network (DePIN). However, DePIN will have trouble competing with centralized cloud providers like Amazon Web Services, which can leverage their infrastructure (co-located GPUs in large data centers) to offer richer capabilities.
Centralized cloud providers also benefit from bulk buying power when investing in new hardware (Elon Musk recently purchased $500 million worth of chips from Nvidia) and can pass those savings on to customers, making it difficult for DePIN to match its pricing.
AI as Smart Contracts
Running AI in smart contracts overcomes many of the limitations of centralized models, becoming:
Tamper-proof - Because the Internet computer is highly decentralized and its consensus mechanism uses chain key encryption, smart contracts are immune to network attacks and there is no single point of failure;
Unstoppable - Once a smart contract is deployed, it is guaranteed to run, the Internet Computer never goes offline, never needs to be restarted, and does not rely on data centers that may experience outages;
Autonomous — Smart contracts can be managed by a DAO without even a control structure, in which case no one can circumvent its rules or change its code.
Most blockchains cannot run AI completely on-chain because their smart contracts can only access limited memory, which also explains why some blockchains rely on centralized cloud servers, so they use blockchain to run AI.
The Internet Computer is different because our smart contracts use a 32-bit WebAssembly (WASM) virtual machine, so they can access 4GB of memory, which means we run AI on the blockchain.
Of course, running LLM in smart contracts is the ultimate goal, and while the computing power required to achieve this presents considerable technical challenges, we are making progress.
The Internet Computer is moving to 64-bit WASM, which will quadruple our smart contract memory to 16GB, enough to deploy a lightweight version of Llama 3.
The blockchain effect
So how does an Internet Computer version of AI on a blockchain protect against the threats of centralized models?
Let’s start with a real-world example. Let’s say a law firm employs 100 attorneys. Each attorney might take weeks to prepare for a trial, which is costly for their clients and the attorneys might miss important details.
Now imagine that this law firm had an AI oracle that collected all client communications, legal documents, historical precedents, and texts. If the lawyers had access to this oracle, the time required to prepare for trial would be reduced to a few days, and the likelihood of mistakes would be reduced.
If this model runs on a centralized server, it becomes a target for hackers because of the sensitive information stored. It is the equivalent of a hot wallet for client data. Anyone who holds cryptocurrency knows how vulnerable they are. Hackers can use this information to undermine a client’s legal case or life. Or the IT infrastructure may suffer an outage before trial, preventing lawyers from conducting the research they need to build their defense.
However, as described in the previous section, due to the immutable and unstoppable properties of smart contracts on the Internet Computer, neither of these situations will occur in the model running the smart contract.
Likewise, dApps can leverage AI on the blockchain in innovative ways. For example, a digital wallet can be a smart contract, which provides a range of benefits including enhanced security. Since smart contracts are autonomous, no one can modify the code and steal a user’s assets.
These wallets can also include customizable features, such as trained AI models that can receive instructions to execute transactions under certain conditions, such as exchanging half of a user’s Bitcoin holdings for USDC if the BTC price reaches $100,000.
Many use cases can benefit from autonomy. Another example is an AI model that audits Solidity code on Ethereum to detect vulnerabilities and prove its trustworthiness. The legitimacy of the audit depends entirely on the autonomy of the smart contract that contains the model.
There are some big things happening on the horizon…
Dom ended the conversation by revealing a major development that will bring a new dimension to the intersection between AI and blockchain. He didn’t reveal more details to Tracy, but he said the team hopes to launch a beta version in early September, so stay tuned!
To see AI on the blockchain in action on the Internet Computer, check out this facial recognition demo that Dom recently recorded:
IC content you care about
Technology Progress | Project Information | Global Activities
Collect and follow IC Binance Channel
Get the latest news
