AI and blockchain: the top 4 use cases & the biggest risks

Aptos Labs has just announced a major partnership with Microsoft that will include an AI assistant. But are AI and Web3 really such a good match? We posed this question to Pontem’s leaders Alejo Pinto and Boris Povod, as well as our partners at Concordia, Panther Protocol, and Switchboard.

Can’t put AI in the driver’s seat just yet, says Alejo Pinto

AI is one of the top crypto narratives of 2023, with dozens of projects exploring use cases from data markets to whole AI-optimized blockchains. Among the leading L1s, Aptos has just made headlines with a partnership with none other than Microsoft. The tech giant owns OpenAI, the developer of ChatGPT. Together, Aptos Labs and Microsoft will explore ways to use AI for onboarding users to Web3 and for smart contract development.

Of course, there are still those who are cautious about the use of AI in Web3. For example, Andre Cronje, the creator of Yearn Finance and Solidly, even said that “AI and blockchains don’t mix”.

But that’s not what the CEO of Pontem Alejo Pinto thinks. He says that AI can and will transform the Web3 industry, but that at the current stage it still requires a lot of human supervision:

“We’ll arrive at a point where AI will be everywhere and perhaps will even replace GUIs in apps. In blockchain, AI will write and deploy smart contracts. Web3 will become a network of self-optimizing algorithms that hardly need any human input.

At the same time, now we have to apply a conservative approach in Web3, since so much money is at stake - and also because smart contracts are irreversible. Once you’ve deployed a contract, you can’t change it. So you can’t just hand the keys to the AI and let it drive.”

Indeed, when it comes to implementing AI in Web3 products, mistakes can cost dearly. We’ve approached Pontem CTO Boris Povod and the CTO of Panther Protocol, Dr. Anish Mohammed,  and asked their opinion on the safe use of machine learning algorithms in Web3.

Mitigating the risks of AI

Telling apart the true and the false

AI chatbots and assistants  are the first thing that comes to mind when one thinks ot the use of artificial intelligence in crypto. Indeed, LLMs (large language models) like ChatGPT can help onboard users to Web3. But there is a problem: LLMs struggle to tell the difference between true and false information online. There is simply too much crypto content out there, of various degrees of truthfulness, and AI training datasets don’t prepare algorithms to distinguish between honest praise of a quality project and a paid shill.

Pontem CTO Boris Povod comments,

“An LLM is like a bookworm on steroids. It will consume any content it finds, from scholarly articles to tweets. Unfortunately, AI can’t judge this content critically: if will take a fairy tale for a fact unless it’s trained otherwise. The risk is that a crypto assistant will start feeding users investment advice taken from a Twitter shiller account, for example, leading them to buy risky or worthless tokens.

We can mitigate this risk by using AI as an advisor and not the CEO, figuratively speaking. Let it pitch ideas to you, but make sure to vet those ideas before implementing them.”

Shraddha Agarwal, co-founder and CTO of SwapGPT, points out another LLM risk: improper use of data.

“AI assistants are a great way to optimize UX and workflow - for example, in DEX trading. However, we must take care to preserve user privacy when training these algorithms. As we strive to provide the best possible user experience, it can be all too easy to compromise personal data.”

The cost of putting AI on chain

In this article, we are talking about using AI together with blockchain, whereas AI continues to run on centralized Web2 servers. Would it be possible to create dApps where AI “lives” on the blockchain itself?

Dr. Anish Mohammed, CTO and Chief Scientist at Panther Protocol, explains why it’s not cost-efficient:

“An LLM has to process a huge volume of data. If you tried to put all those operations on chain, the storage and computation cost would be prohibitively high.

Of course, you can still have AI-based financial dApps. After all, in TradFi AI has long been used to power high-frequency trading algorithms with billions of dollars flowing through them. The issue arises when you try to put that sort of AI on a blockchain.”

AI in Web3: experts describe 4 use cases  

Data preferences and usage

In order to build an AI algorithm, you need a vast dataset. In Web2, all that data is controlled by a few and is accessible only to a few. The other side of the problem is that data comes from users and isn’t always utilized with their consent.

Aptos co-founder Avery Ching, who has an extensive background in machine learning and data infrastructure at Meta, posted a Twitter thread describing how blockchain can help in both cases. On the one hand, on-chain data is universally accessible, so anyone can use it to train AI models. It actually has “more verified interactions” than regular Web2 data, so it’s better for training datasets.

On the other hand, fast blockchains like Aptos can serve for recording users’ data preferences in a verifiable way. Companies can even reward users for providing their data for LLM training.

Oracles and code verification

Oracles are key blockchain primitives that provide smart contracts with external data like cryptocurrency prices. Can the industry benefit from an integration of AI into oracles?

Mitch Gildenberg, CEO of the multichain oracle protocol Switchboard, says that it definitely can.

“An important feature that becomes possible with AI is arbitrary code execution and verification. You can teach an oracle to verify if a transaction or any other action has been carried out by a specific algorithm as opposed to some third party. This is achieved by running an LLM and submitting its responses to the blockchain to verify the code checksum.

This will be important in AI-powered asset management, where you need to be sure that it is the AI algorithm that is performing the trades. In Switchboard V3, we’ll implement arbitrary code execution and verification on EVM chains and Solana first, followed by Aptos.”

Content origin

The internet is being flooded with AI-generated content, and many people are concerned that it’s often impossible to say which article or image was created by a human and which by an algorithm. Others worry that LLMs will put content creators out of work.

Blockchain can provide a solution. Thomas Ruble, CEO of Concordia - a multi-chain risk and collateral management protocol on Aptos - explains:

“Blockchains are essentially databases of signed facts, so they are a natural fit for storing public records about data origin. You can use on-chain databases to record the provenance of each piece of content and its author - using signatures that can’t be forged.”

DEX trading

We’ve said that in traditional finance, machine learning algorithms are widely used for high-frequency trading. What about DEX trading on the blockchain? SwapGPT CTO Shraddha Agarwal stresses that while predictive AI tools can greatly help on-chain traders, you have to invest effort in model training.

“Data on the blockchain is accessible to all: if we’re talking about DeFi trading, you can get DEX transaction history from explorers, aggregators, etc. However, it’s less organized than financial data in TradFi. Before you can use it to train AI, you need to organize, structure, and clean it. The training procedure itself is the same, but different data types and parameters are used, specific to tokens. You also have to consider the specific objectives of DeFi traders: optimizing liquidity and minimizing slippage.”

Towards the singularity

Pontem CEO Alejo Pinto concludes:

“With AI, we’re moving towards a technological singularity - a world where AI-powered machines can coordinate with each other and self-organize. It will be a super intelligent artificial organism - far more intelligent than humans.

Blockchain could be the last piece of the puzzle needed to bring this singularity about and we at Pontem will be doing our part to bring it about.”

About Pontem Network

Pontem Network is a blockchain product studio building for Aptos, an L1 blockchain known for its sub-second finality and security. The company’s products include the popular Pontem Wallet for Aptos with its 300,000 installs; Liquidswap DEX; and the first browser code editor for the Move language.

Pontem is currently preparing to release an AI product of its own. It also sponsored the AI track at Aptos’ recent Hack Holland event.