Niraj Pant, Co-founder of Ritual, on The Future of Crypto and AI, Decentralized AI, and Applying ML to DeFi | Ep. 324

Author: CoinSense

In an exclusive interview with Cryptonews, Niraj Pant, co-founder of Ritual, the decentralized execution layer for AI, talked about why blockchain and AI haven’t joined hands yet – and how blockchain is reaching the point where it’s ready to do that.

He told us how the relationship between AI and crypto is mutually beneficial, the lack of AI-enabled dapps and what Revolut is doing about it, the GPU problem and how crypto solves it, why some DeFi aspects are inefficient and how ML can change that, and more!

In this interview, Pant discussed:

  • building autonomous worlds and games;
  • building decentralized AI infrastructure
  • applying ML to DeFi;
  • Smart Agents;
  • Story Protocol working with Ritual to train and track models on-chain.

Niraj Pant gave a wide-ranging interview, which you can see above or read below.

Creating the Foundation for AI On-chain

It is notable that AI crypto is emerging now and not, let’s say, during the last cycle, Pant told Cryptonews Podcast host Matt Zahab.

There has been an idea of decentralized AI for a long time. But it faced a number of challenges. It couldn’t reach the level of what is today a growing sub-industry within the crypto space.

A key challenge is that there simple wasn’t the developed transformer (deep learning) architecture as we see it now.

Yet today, we find machine learning (ML) architecture behind numerous large language models used in massive products, such as ChatGPT.

Additionally, there were not even many consumer AI applications to speak of before.

Following the rise of transformer architecture, as well as the creation of ChatGPT and GPT-3 in 2020, “a new renaissance in AI development” has begun.

“We finally saw a big AI consumer use case,” Pant said.

Now, numerous companies and startups are using AI, developing it, and/or raising money for it.

Before all this had happened, bringing AI to blockchains was pretty much impossible. Blockchains could not efficiently support actions such as training, inference, and tuning.

Blockchain is just now reaching the required level of efficiency. We’re seeing improved speed, better cryptography, Layer-2 blockchains, zero-knowledge proofs, and more.

AI may still be a “pie in the sky,” but these developments are “bringing us to a space where we can actually do AI on blockchains.”

Crypto and AI: Bi-directional Benefits

Crypto and AI can help each other, Pant said. There is a “confluence of the two spaces.”

Therefore, the two gain “bi-directional benefits.”

AI is highly centralized. And this is where crypto can help.

AI is basically just a few major products that people use globally (ChatGPT, Midjourney, and Runway, for example), and only a few companies making the models (OpenAI, Microsoft, Google, and Facebook).

The crypto world, however, is very resistant to the idea of a centralized authority. They realize that hands with too much power in them are bound to abuse it.

It is notable to remember that AI is not going anywhere: nearly all products use it in some way, and “billions of people” will use it daily in a matter of years.

Therefore, having a “decentralized, transparent alternative where you can get privacy, computational integrity, governance rights, ownership amongst the users and the people that contribute to it is really important.”

Meanwhile, AI can also help crypto. There are many interesting use cases, from non-fungible token (NFT) generation, over building customized games, to creating customized movies, and much more.

“We’re really entering this new era where we can combine the infinite abundance of AI with the ownership and self-sovereignty properties you get with crypto,” Pant said.

They are at the opposite ends of the technology spectrum: one is centralizing, and one is decentralizing. Merging them is key.

Missing: AI-enabled Dapps

Before he started working on Ritual, Pant spent six years as a General Partner at Polychain.

He got increasingly interested in AI, while also staying “very close to the crypto infrastructure side of things.” That led him to research the crypto-AI intersection.

While many of the teams he worked with at the time had great ideas, he started noticing that no one was building AI-enabled decentralized applications (dapps).

“That feels like a massive opportunity,” Pant commented. “That’s going to grow in size a lot.”

Therefore, their company, while it started small, is now growing “quite rapidly.” The amount of attention the Ritual team has seen in the space “has skyrocketed.”

The team behind it focused on enabling developers to easily use AI in their smart contracts and on enabling the above-mentioned use cases.

To accomplish this, they built out Ritual in two phases.

The first phase is a system called Infernet.

This is a lightweight library to bridge off-chain compute on-chain. It’s a decentralized oracle network that enables smart contract developers to request computation to be done off-chain from Infernet Nodes and delivered to their consuming on-chain smart contracts via the Infernet SDK.

For example, if a developer wants to create a new NFT mint based on user input, they could build a smart contract that relays the information to the Infernet off-chain compute system, which does the work inside a container, and then returns the result, optionally with proofs or privacy.

The focus today is on EVM-compatible blockchains, but in the future, it will be “really anything.”

The second phase is called the Ritual Chain.

This sovereign, Layer-1 chain extends the ideas around Infernet onto an execution layer where users get more direct proofs, privacy, and on-chain semantics that make it easier to build the applications they want.

It will be an execution layer custom-built to support AI-native operations and enable a new class of applications at the intersection of crypto and AI.

“That’s kind of our roadmap for the next year,” Pant said. “We are looking to do devnet in the summer.”

However, developers can build applications on Infernet and move them to the chain if and when they choose.

The GPU Conundrum and How Crypto Solves It

Niraj Pant told Matt that when it comes to the graphics processing unit (GPU) as a service, “there’s a ton of different vectors that you can innovate on.”

This could be on the geographical side, new market side, incentives with tokens, different types of hardware, being able to coordinate those machines, and more.

Therefore, GPU as a service is “one of the most interesting use cases of crypto AI.”

GPUs can be used for “a whole bunch of tasks.” But buying one – unless you’re a massive company – is extremely time-consuming and expensive.  By the time you get one, the tech has advanced, so “you’re constantly behind.”

Another option is using cloud services. However, these are limited in availability and/or “they just charge an insane premium” on top of the actual raw cost.

These scenarios have resulted in GPUs becoming prohibitive for many startups building AI.

There is a third option: Web2 providers that offer a basic service for more technically savvy users.

While these are great, they don’t have “a full market” like, for example, the crypto market does: buyers and sellers on two sides. Instead, the provider is always on the other side.

Therefore, these companies are “running their own supply.” They have limited hardware, and they get to dictate the deals and with whom they make them.

However, the crypto market could expand the range of hardware suppliers. Perhaps certain groups in Europe or Asia can now open up the supply and satiate the demand.

Another key thing is being able to bridge into more types of hardware through novel, unique architectures. Crypto can do this.

“And one of the great things is that crypto uses a ton of GPUs,” Niraj Pant said.

Meanwhile, Ritual has partnered with an external GPU-as-a-service company called Ionet. They have “a massive cluster of GPUs all around the world” that run Ritual nodes and are able to take down Ritual demand requests.

Using ML to Make DeFi More Efficient

Ritual has recently released a toolkit called Infernet ML.

This is a series of ML workflows that the team has pre-built – a “bunch of examples that allow you to do things like that NFT mint, or use an LLM in a smart contract, or really anything else, across a bunch of different ML frameworks.”

That said, ML helping decentralized finance (DeFi), Niraj Pant remarked, is one of the most interesting applications for crypto AI.

Today, when you launch a DeFi protocol, you’re trying to accomplish a task. Some DeFi protocols are very narrow in what they offer, while others are full-featured systems with many different products.

However, building a protocol is not the end of the job. Now, the team has to manage everything related to it, such as the system itself, the treasury, the protocol security, risk, and much more.

Additionally, a big issue in DeFi today is governance, Pant argued. Decentralized organization (DAO) governance is difficult and “laborious in time and in people.”

It takes days to read through proposals, deliberate, do “politics” to get the required votes, and finally vote.

“And this is very inefficient for many tasks,” Pant said.

However, while human governance will likely remain necessary for treasury management for a long time, said Pant, AI and ML may benefit other aspects.

“This might be things like what’s the interest rate parameter, or what’s the liquidation factor or the collateral factor. You can stream in data from different protocols and stream in price feeds from different exchanges and use that to drive the decision-making around those different factors,” the co-founder explained.

So if the price of an asset that’s on a lending market drops significantly within some bound, this would indicate that a project should tighten up the required collateral and make people shore up so that the protocol doesn’t have additional risk.

Therefore, DeFi can be used with AI in many different ways, Niraj Pant noted, from governance proposals to treasury management to risk parameter management and in many different use cases, such as lending, yield generation, portfolio optimization, and others.

Meanwhile, one of Ritual’s advisors is Tarun Chitra, Founder and CEO of Gauntlet. In the future, Gauntlet could create the models and earn a royalty—a more direct form of revenue than the current one. “So it’s a very exciting future. It’s one that we talk about internally quite a lot,” Pant said.

Also, there’s a lot coming down the pipeline for Revolut. They’ll be pushing “tons” of use cases around AIs and PCs, new types of ways to interact with NFTs, making games more personalized, and much more, Pant concluded.

__________

About Niraj Pant

Niraj Pant is co-founder of Ritual, the decentralized execution layer for artificial intelligence (AI).

Prior to founding Ritual, Pant spent six years as a General Partner at Polychain, leading investment rounds in startups like Offchain Labs, EigenLayer, and Compound.

He began his career as a cryptography researcher in the Decentralized Systems Lab at UIUC.