Cryptonews Podcast host Matt Zahab recently sat down for an exclusive interview with Dominic Williams, the Founder and Chief Scientist of the DFINTIY Foundation, a major contributor to the Internet Computer (ICP) blockchain and avid AI enthusiast.
Williams discussed the future of AI, its ‘perfect partnership’ with smart contracts and blockchain, and how AI can help average Joe and Jane perform complex trades with verbal instructions.
He further explained what differentiates Internet Computer as a crypto cloud from other blockchains, and why it’s currently the only one in the “third lane.”
In this interview, Williams discussed:
- decentralized AI (DeAI);
- why AI projects should be decentralized;
- why dapps should get into AI;
- building out a world-class content studio;
- hosting AI models on the blockchain;
- multichain DeFi;
- decentralized cloud computing.
Dominic Williams gave a wide-ranging exclusive interview, which you can watch above or read below.
The Future of AI is Smart Contracts
Sitting in a new studio in the basement of Zurich HQ, Williams discussed the promising future of AI.
The Internet Computer (ICP) blockchain aims to function as a decentralized cloud. It would enable AI models to operate entirely on the blockchain without centralized cloud providers like Amazon Web Services.
Williams opined that “the long tail of AI, where people run their own models, is probably one where AI runs smart contracts on blockchains.”
And this is possible today on the Internet Computer blockchain.
Everyone in crypto is currently talking about AI: blockchains for AI, tokens for AI, ChatGPT integration that people can use to put chat into their Web3 service, and much more.
And all that is great, said Williams. However, DFINITY is more focused on the fundamentals.
The project’s mission is to reinvent computing on blockchain. This means that they see blockchain as it is today on the Internet Computer.
Smart contracts on the Internet Computer are a bundle of web assembly bytecode and persistent memory pages, and they run in parallel. They directly serve user experiences in end users’ browsers.
This blockchain specifically, said Williams, is capable of acting like a “crypto cloud where you can build anything with smart contracts.”
This allows it to process some 330,000-350,000 Ethereum-equivalent transactions a second.
Moreover, some of these smart contracts are now packing up to 400 gigabytes of persistent memory, said Williams, providing enough memory and compute power to run an AI.
Earlier this weekend the Internet Computer #ICP passed 3 billion blocks 💪 It’s probably now processed substantially more blocks than all the other blockchains combined (?) It’s a beast.
— dom.icp ∞ (@dominic_w) March 24, 2024
“And what we’ve done at DFINITY,” Williams added, “is worked on technology that’s made it possible for the Internet Computer blockchain to run computation at scale. Today, it hosts many orders of magnitude more data and computation than traditional blockchains. And that’s why it’s able to run AI smart contracts.”
AI and Blockchain are Perfect Partners
It’s old news that OpenAI took the world by storm. However, per Williams, it and similar projects are “centralization on steroids.”
However, he said, the good news is that, while these large language models with huge amounts of money do grab a lot of attention, “in the future we’re going to see millions of individual AI models running.” These will focus on specific tasks.
Actually, we’re already interacting with AIs on a daily basis. A common example are services like TikTok or Instagram. This kind of content recommendation neural networks, for example, will be very prevalent in the Web3 field, Williams remarked.
There are some key things to pay attention to, the founder warned.
There will be standalone large language models that coordinate organizations. An example is Delphi AI, which connects to all aspects of a technology company running it: the calendar, Slack, GitHub – any and all apps.
This means that the AI model will have access to all of this organization’s most sensitive data. And it’s going to become the number one target for hackers.
Furthermore, a malicious actor could infiltrate the company and feed the AI misinformation. The AI can’t tell the difference between fake and real, and it becomes corrupted.
Therefore, employers within the organization should interact with this type of AIs only with a strong blockchain-based authentication.
Also, the organization would be fully dependent on the AI. Should this AI run on traditional infrastructure and something mundane goes wrong – a piece of equipment breaks, for example – and the AI goes offline, it would be “incredibly disruptive.”
All this serves to show that AIs must be tamper-proof. Williams said this is achieved through smart contracts.
When AI is accessed via strong authentication through blockchain, “that the AI is unstoppable.”
Therefore, said Williams, “when you think about it, AI and blockchain fit together perfectly. And blockchain is the only way of creating a tamper-proof, unstoppable compute platform.”
All this said, if you want the absolute most efficient AI, you’d run it on a dedicated AI cloud, the founder added.
Power of AI-Based DeFi
Blockchain is complex. Crypto, decentralized finance (DeFi), decentralized apps (dapps), and other sectors of the industry are difficult for the everyday user to understand on the tech level.
“Things start getting complex if you really have to enable people to create almost arbitrary instructions,” Williams argued.
However, this is where AI can come in. Users can just tell it what they want.
Should a project catch a user’s eye, they can get their phone out and verbally provide instructions to an AI crypto wallet that will then carry out a transaction.
Williams provided another example. A user can instruct the AI to sell 5 WBTC if WBTC on Ethereum reaches $100,000 and only if the slippage is less than X.
The AI will then display a concise pseudocode – a simple one that layman can understand – and, if the users likes it, “they press confirm and boom, it’s done.”
Therefore, there is also the convenience aspect:
“The system can make it possible for normal people to create much more complex actions.”
On the Internet Computer, smart contracts can be invoked in a heartbeat, Williams noted. They can be periodically woken up to, for example, check a price oracle and behave according to instructions that have been previously made.
It allows people to buy and sell crypto and make trades conveniently, as well as create complex default actions based on conditions.
Also, it’s very easy to create a simple swap interface.
Witness the groundbreaking fusion of AI and Blockchain that promises to redefine the future
AI running on blockchain as a smart contract 🤯 pic.twitter.com/Y3pU7gxOEu
— DFINITY (@dfinity) March 26, 2024
While the product is not finalized yet, the first AI demo includes image classification. Users can upload a photograph, and AI will identify it.
DFINITY is also working with partners and AI experts, and “they’ve got various things they want to run on the Internet Computer.”
The project’s ideas include face recognition, which would enable fully autonomous KYC.
Hosting AI Models on Blockchain
Running AI as a smart contract involves a certain amount of replication.
The Internet Computer allows users to choose the level of replication. On a standard system subnet, it’s doing 13x replication, Williams said.
In the future, there may be lower trust AI subnets, which only have 4x replication, which still gives very impressive security and resilience, he argued.
And the higher the level of replication, the more queries you can do.
Currently, the Internet Computer’s limitation is the amount of main memory that a smart contract has: 4GB each. That’s because it’s still running 32-bit WebAssembly (wasm).
However, said Williams, “I’m really excited about this. I’m pleased to say that we’re moving to 64-bit wasm soon, which completely removes that restriction and will probably increase the maximum amount of main memory to 32GB or something like that.”
This will enable the blockchain to run large language models. (For explanation purposes, large language models include the GPT-3 and GPT-4 from OpenAI, LLaMA from Meta, and PaLM2 from Google.)
Meanwhile, DFINITY’s pipeline already includes Single Instruction/Multiple Data (SIMD) instructions. They will be put into the smart contract execution environment to “massively” speed up these AI smart contracts.
“And we’ve got a whole lot of other things in the works too,” Williams said. First, they plan to host many AI demos.
Furthermore, they’re working on two major projects: one is Utopia, and the other is Orbit – an open source project that will enable users to create a completely decentralized, institutional-grade, cross-chain crypto custody solution on the Internet Computer.
Creating Utopia
Speaking of future development, Williams noted two main areas the team is focusing on.
One is helping optimize the Internet Computer to run AI so it can run larger and larger models and do so more efficiently.
The other is Utopia, which will enable the enterprise sector, governments, NGOs, and others, to run like private Internet Computers that can integrate with the public Internet Computer and with each other.
People using Utopia to create these private clouds will “just be focused on the fact that online systems and services are tamper-proof and unstoppable.”
Another important aspect to note is that the team behind the Internet Computer is creating advances and milestones all the time – yet there is no “fanfare.”
On other blockchains, there are really trivial features “wrapped up in this big release with a fancy name, and then it’s marketed to hell.”
But Internet Computer “isn’t the best” at marketing itself. Williams noted that it tends to be too tech-focused.
This is also something they’re working to correct.
A great example of this is something called deterministic time slicing. It’s a huge achievement, Williams said. It essentially shows that the Internet Computer is a genuine, decentralized operating system.
And while operating systems like Google’s Android, Apple’s iOS, OSX, Windows, or Linux run on a single computer, the ICP stack “works as a kind of operating system that runs across multiple compute units.”
These “things are so much more further in advance than what people are used to on blockchain,” Williams said.
The Three Lanes of Blockchain Technology
Lastly, in connection with the previous discussion, Williams noted that there are three lanes when it comes to blockchains.
The first comprises blockchains dedicated to hosting a ledger.
The second lane is for blockchains that host smart contracts.
The third lane houses “the world computer blockchain.” This is where Internet Computer is, and it’s difficult to communicate that, Williams says.
#ICP‘s tech is years ahead, and will stay ahead.
“A blockchain that supports full stack decentralization, where you can build social networks fully on-chain, and even run AI as smart contracts.” – @dominic_whttps://t.co/G6JFBDGvKB
— DFINITY (@dfinity) March 28, 2024
People assume it resides in lane two, that it’s one of the traditional blockchains hosting smart contracts with “less compute capacity than a Raspberry Pi.”
However, this blockchain is already a serverless cloud where users create online systems and services using smart contract software.
This is not the same as what the leading blockchains have to offer, Williams said. The difference is relevant to communicate.
Blockchains tend to store “very tiny amounts of data and perform very tiny amounts of computing. […] The difference between where blockchain is today and where it has to go to become a crypto cloud is a long one,” Williams concluded.
__________
About Dominic Williams
Dominic Williams is the Founder and Chief Scientist of the DFINTIY Foundation, a major contributor to the Internet Computer (ICP) blockchain.
He is a crypto theoretician and entrepreneur. He has been involved in the blockchain space since 2013. Prior to that, he was an engineering entrepreneur who created multiple internet technologies and products.
Williams specializes in distributed computing and crypto network theory, having proposed multiple innovations that are in use today.
