In the latest panel hosted by Pontem Network, experts from Pontem, Panther Protocol, Alethea AI, and SwapGPT shared some unexpected – even controversial – views on use cases for AI in Web3, from decentralized trading to DAO decision-making.
The abundance engine and the scarcity engine: AI and Web3 collide
Pontem Network is a leading blockchain product studio on Aptos, known for its popular Pontem Wallet, Liquidswap AMM, Move language tooling, and more. As Pontem prepares to release its own AI-powered crypto assistant, PontemAI, Pontem’s Chief Growth Officer Alejo Pinto hosted a Twitter Space with several leading experts on AI in blockchain. The discussion proved eye-opening, as the speakers highlighted obstacles and limitations that the media often ignores.
Alejo Pinto noted that the emergence of Large Language Models like ChatGPT marked a “tipping point”:
“Fiction is now becoming reality. I would compare the popularization of AI to such technological tipping points as the invention of the printing press or the internal combustion engine or the splitting of the atom. Knowing that AI could soon be omnipresent is both exciting and scary.”
As AI and Web3 are the two hottest tech narratives, many view them as a perfect match. But in some ways, the two are polar opposites. Brent Homesley, Head of Partnerships at Alethea AI,an R&D studio specializing in generative AI, couldn’t have put it better:
“AI is an abundance engine. It will create as many ideas or even business projects as you prompt it to, in any area: trading, content, digital companions, and so on. By contrast, Web3 is a scarcity engine, because blockchain allows you to prove ownership.”
Indeed, when someone’s rights to an asset are recorded on the blockchain, that thing can’t be freely used by everyone – making such assets potentially scarce.
Another way in which AI and crypto are opposites is in centralization. Alejo Pinto stresses that “AI is essentially a centralizing force, aggregating data and models, while the power of AI is being accumulated in the hands of centralized corporations. Crypto is the opposite: decentralizing data.”
“AI is getting smarter off the labor of the masses”
As Alejo Pinto noted, AI is a centralizing force, with data accumulated and algorithms controlled by giants like Meta, Microsoft, and Google. Can crypto help counterbalance this centralization of AI? Brent Homesley believes it can:
“With ChatGPT, everyone who uses it is training the model for them. It’s getting smarter off the labor of the masses. But look at what happened with Instagram: creators are leaving it and going to TikTok instead, where they can monetize content better, instead of making money for the centralized company.”
Pinto agreed that data privacy is starting to become a consumer trend as people realize how much power their personal data gives to companies. However, corporations’ vast resources allow them to keep LLM usage cheap. A co-owned Web3 model will cost more, and not everyone will pay more for privacy.
Pinto thinks that we may end up with a two-tiered marketplace, similar to health food versus fast food.
“There may be two different consumer categories: junk food AI models – very fast, efficient, but centralized, those that aggregate your personal data for corporations. Others will be “organic AI” – not as calorie-dense and more expensive, but good for your privacy health. Of course, UX needs to be equivalent, as well.”
“AI predictions are only as reliable as the data”: LLMs, DEX trading, and DAOs
The guests agreed that one of the best ways to use artificial intelligence in crypto is to improve decision-making. In the words of Alejo Pinto, AI is like self-driving cars: fully autonomous vehicles are a big leap from systems that help you drive more safely.
One area where AI-informed decision-making would be beneficial is DEX trading. Shraddha Agarwal, co-Founder and CTO of SwapGPT (a provider of AI liquidity management solutions, explained:
“AI can accurately process vast amounts of data from varied sources like blockchain explorers and graphs, analyze social media sentiment, and find correlations. Add to this support for natural language prompts, and you get AI suggestions that DeFi traders can use even if they don’t know much about technical indicators.
Such models can help minimize losses, avoid emotional trading, and improve executions. Still, such tools are designed to assist traders, not replace them: human analysis remains critical.”
SwapGPT CEO Keshav Saraogi continued:
“The human trader has to provide AI with inputs: how they want the trade to be executed, how much risk they are prepared to take, and so on. Once you’ve informed the AI model, however, you don’t have to monitor it. The AI will make decisions and react much faster than a human – for example, in case of a volatility spike.”
Another promising application for AI in crypto is to help inform DAOs. Maker DAO already has a vision to use AI for risk management. Such an AI would provide suggestions to the lending protocol,for example, to lower the risk on a specific collateral asset. Human DAO members would then vote on the suggestion.
Brent Homesley even foresees the emergence of DAOs around co-owned AI models themselves:
“A group of people with a vested interest in an LLM may want or not want it to be trained on specific points. They could form a DAO and vote to put some things in or take them out – like crowdsourcing info from a DAO. Or, if it’s a trading model, the DAO could vote on how it wants it to trade.”
Before humans can trust AI suggestions, be it for DAO proposals or DeFi trading, you have to make sure that AI doesn’t “hallucinate”: providing confident responses that are in fact completely false. To achieve this, AI models will require more reliable data.
“We have to be very careful and only train models on cleaned, vetted data, says SwapGPT’s Keshav Saraogi. “The task is to avoid hallucinations from day one and to validate each model before use. This is especially important in DeFi. This market fluctuates rapidly and AI models can struggle to adapt. AI predictions are only as reliable as the data.”
“I haven’t seen anything driven by just privacy”: can ZK-powered AI attract users?
The cost of AI decentralization emerged as the biggest issue during the discussion. Popular blockchains don’t have the throughput to support the scale of computation needed for machine learning. This could be solved using very fast and scalable blockchains like Aptos – but for now, computation will have to remain off-chain.
This, in turn, creates another problem: how do you prove that AI computation was done properly if you don’t own the location where the data reside? This becomes possible with zero-knowledge (ZK) technology, as ZK proofs can verify on-chain that the output of an off-chain AI model is trustworthy.
Decentralizing a machine learning model is a challenge in itself,so a ZKAI product is likely to be expensive. Yes, it will provide privacy and confidentiality, but are people willing to pay for that?
Dr. Anish Mohammed, CTO at Panther Protocol, a ZK cross-protocol layer, doubts it:
“For most users, there is no difference between using centralized or decentralized products, as long as the experience is the same and the costs are in the same ballpark. The switch has to be seamless and not impact the price too much, and we’re not at that point yet.
As for the added advantage of privacy, it’s like a smoothie at McDonald’s. Nobody goes to McDonald’s to get a smoothie, but if I’m getting a burger, I may get a smoothie as well. I haven’t seen anything being driven by just privacy – if it was, Facebook and Google would be doing it already.
Current ML pricing mechanisms don’t account for compliance and privacy. If these are taken into consideration, there will be valid economic incentives to do this. Perhaps in the next couple of years we’ll see viable solutions for AI decentralization.”
In conclusion, Pontem’s Alejo Pinto expressed hope that the incentives to adopt decentralized structures for AI will arise, and that affordable and scalable chains like Aptos will speed up the process.
The full recording of the AI and CryptoTwitter Space is available here.
About Pontem Network
Pontem Network is a blockchain product studio building for Aptos, the L1 blockchain known for its sub-second finality and security. Pontem products include:
– Pontem Wallet: the only triple-audited wallet for Aptos with 300,000 installs;
– PontemAI: an AI-powered crypto chatbot (coming soon);
– Liquidswap DEX: one of the most popular DEX on Aptos
– Move Code Playground: the first browser code editor for the Move language.
Pontem is currently preparing to release PontemAI, a crypto chatbot powered by best-available market data.. Pontem also sponsored the AI track at Aptos’s recent Hack Holland event.