The Future of Crypto is Intelligent: AI’s Role, the Next Frontier | Sreeram Kannan | Open AGI Summit

Sham, founder of Igen and CEO of Igen Labs, discusses the intersection of AI and crypto, emphasizing the potential for creating "Unstoppable" and "Verifiable" AGI through smart contracts. He critiques current on-chain AI agents for lacking these properties and proposes a model where AI code is embedded in smart contracts, allowing for autonomous and verifiable AI operations. Sham explores various methods to ensure verifiability, such as cryptoeconomic security, and highlights the potential of Igen Layer to build a universe of verifiable services, ultimately enhancing the intelligence and trustworthiness of smart contracts.

• The intersection of AI and crypto involves the merging of two technological worlds: AI/AGI development and crypto's creation of unstoppable computers.
• The potential outcome of this intersection is an "Unstoppable AGI," combining AI's capabilities with crypto's verifiability and autonomy.
• The core properties of crypto, autonomy, and verifiability, are crucial in understanding how AI and crypto can interact.

"What is actually happening in this intersection between AI and crypto? You have these two different worlds: one world in which people are building AI AGI and another world where crypto is building unstoppable computers. If these two worlds collide, you may get unstoppable AGI."

• This quote highlights the potential synergy between AI and crypto, suggesting that their combination could lead to the development of an unstoppable AGI.

• Present onchain AI agents resemble traditional API integrations, lacking the unique properties of crypto.
• Current implementations do not leverage the autonomy and verifiability aspects of crypto, making them uninteresting from a technological perspective.

"Most people look at the agents today, and agents on chain today look like this: an AI agent run on a server controlling a wallet. There is no difference between an agent using an Ethereum wallet and an agent using a Stripe API."

• This statement emphasizes the limitations of current AI agents, which do not utilize the transformative properties of crypto, such as autonomy and verifiability.

• Smart contracts, core to Ethereum, are unstoppable and verifiable, allowing code to own money without a private key.
• Theoretically, AI inference can be integrated into smart contracts, resulting in unstoppable, verifiable AI.

"Smart contracts are unstoppable once you write a contract, make it immutable, and throw it on a chain; it runs forever and is verifiable."

• This quote explains the fundamental attributes of smart contracts, which are central to the idea of integrating AI into the blockchain for creating unstoppable, verifiable AI.

• The concept of unstoppable, verifiable AI is theoretically possible by embedding AI code into smart contracts.
• Practical implementations are needed to realize this potential, overcoming constraints like gas and computation limits.

"Unstoppable verifiable AI is possible because you can take AI code and put it into a smart contract. It's already really interesting that you can actually have unstoppable, verifiable AGI."

• This statement underscores the theoretical feasibility of creating an unstoppable, verifiable AI by leveraging smart contracts, highlighting the need for practical applications.

• Smart contracts lack sufficient compute power to run complex AI models, necessitating the use of offchain nodes for computation.
• Offchain nodes execute the code and return results to the blockchain, where the outcomes can trigger further onchain actions.
• Ensures that AI computations are verifiable, which eliminates the need to trust a single computation host.

"You're outsourcing offchain computation for running AI and bring the results on chain."

• This describes the process of using external nodes to handle complex computations that smart contracts cannot manage independently.

"If you can make sure that it is verifiable, I don't need to trust this host for actually running the computation."

• Verifiability ensures trustless execution, allowing any node to perform the computation without relying on a single trusted entity.

• Autonomous inference participation means any node can execute the task without requiring trust.
• Verifiability ensures that the computation results are reliable, making the system robust and unstoppable.

"The two Core Concepts here is in order to build Unstoppable verifiable systems you need... the Outsource computation need to be autonomous and verifiable."

• Highlights the importance of autonomy and verifiability in creating reliable and decentralized AI systems.

• Redundant execution, used by Ethereum and Bitcoin, is not scalable for AI models.
• Zero knowledge proofs offer verifiability but are inefficient due to high computational overhead.
• Optimistic systems have high latency and complexity, while trusted execution environments (TE) face security risks.

"R execution is extremely non-scalable so there's no way to actually scale it to run AI AGI."

• Points out the limitations of redundant execution for advanced AI applications.

"Zero knowledge proofs have a massive overhead relative to raw computation."

• Indicates the inefficiency of using zero knowledge proofs for AI due to their computational demands.

"Optimistic systems have huge latency but also they're very complex and buggy to write these fraud proofs."

• Describes the drawbacks of optimistic systems in terms of latency and complexity.

• Cryptoeconomic security involves staking to ensure service correctness, with penalties for incorrect operations.
• This system uses both onchain and offchain observable events for robust security measures.
• Ion layer's cryptoeconomic approach supports the development of autonomous verifiable services (AVS).

"You stake e and then promise that you're going to run these Services correctly and if you don't run them correctly... you can actually slash them."

• Explains the staking mechanism as a way to enforce service reliability and accountability.

"We have something much more sophisticated in cryptoeconomic security using not just slashing based on onchain observable events but also based on offchain observable events."

• Describes the advanced security measures that extend beyond traditional slashing mechanisms.

• AVS are self-operating services that do not rely on trusted entities, ensuring reliability through verifiability.
• These services can be integrated into the onchain AI economy, enabling diverse applications.

"The broader thesis of I layer is to power autonomous verifiable Services."

• Summarizes the goal of creating independent and reliable services using the ion layer framework.

"These two properties together Define any service that is built on to of wagon layer we call them AVS."

• Defines AVS as services characterized by autonomy and verifiability, essential for the onchain AI economy.

• The integration of offchain intelligence with onchain contracts creates what is termed as "Unstoppable Verifiable AI."
• This system is operational today and is not a futuristic concept; it is achievable by 2025.
• Verifiability is ensured by the Igen layer, with data made public via the Igen DA, a hyperscale data availability layer.

"This combination of offchain intelligence and onchain contract gives you Unstoppable verifiable AI and this is now unlike the theoretical construct that I showed you earlier this is possible today in 2025 you don't have to wait for you know 10 years to do this."

• The speaker emphasizes the current feasibility of creating verifiable AI systems using existing technologies.

• An AI agent is more than just a model; it requires orchestration, running models, and making tool calls.
• Agents are capable of digital actions, such as calling software tools, and require each component to be verifiable.
• Components of an AI agent include the orchestrator, model, and tools, each of which needs to be verifiable.

"An agent is something that can actually go and call make tool calls and take actions because it's a digital agent it can take take digital actions which is called other software tools."

• The description of an agent highlights its ability to perform digital tasks autonomously.

"If you want this agent to be verifiable you need each component to be verifiable you need the orchestrator to be verifiable you need the model to be verifiable you need every tool that you're calling to be verifiable."

• The speaker stresses the importance of verifiability for each component of the AI agent to ensure the system's integrity.

• Each component of an AI agent can be run as an Autonomous Verifiable Service (AVS) on the Igen layer.
• Examples of tool calls include computations, writing Python code, and making requests, all of which can be made verifiable through AVS.

"Each tool call can be made verifiable if it is run as an AVS on Iler."

• The quote illustrates how AVS ensures the verifiability of various digital actions performed by AI agents.

• Building an onchain agent involves several dependencies: inference, training, data, benchmarks, and GPUs, all of which can be made verifiable.
• Each dependency is a specialized service that requires careful consideration and development on the Igen layer.

"What are some like dependencies that if you want to build an onchain agent then you know what is the brain of this agent you need inference you need training you need data you need benchmarks you need gpus all of these can be made verifiable."

• The speaker outlines the essential components required for developing a functional onchain AI agent.

• AI agents need to perceive external environments beyond the blockchain, such as the internet, applications, and the real world.
• Different kinds of oracles are necessary to provide the agent with information about these external environments.
• Agents can also act in the real world by issuing bounties and require privacy tools like trusted execution environments.

"If you want to perceive what's going on on the internet what is going on inside your app what is going on inside a web server what is going on in the real world you need different kinds of oracles that'll do each one of these things."

• The need for oracles is emphasized to enable AI agents to gather information from various external sources.

"The agent needs to have its own privacy which is it it can have tools like trusted execution environments or like multi-party computation to actually create privacy."

• The importance of privacy tools for AI agents is highlighted to ensure secure interactions with the external world.

• Onchain agents are being developed within the ion layer economy, offering tools to create powerful, expressive, and unstoppable agents.
• These agents are characterized by their verifiability and autonomy, allowing for the creation of decentralized and trustless systems.

"One of these blue boxes is an ABS already being built in the ion layer economy so what you can start doing is building tooling around taking these pieces and mix and match them to build very powerful very expressive onchain agents which are unstoppable and verifiable."

• The quote highlights the potential of onchain agents to be both powerful and verifiable, emphasizing their role in creating decentralized systems.

• Unstoppability and verifiability are key properties of AI models in this context, allowing for the creation of robust and reliable systems.
• These properties enable the integration of various verifiable services to create rich AI agents.

"AI ex scrypto is interesting because you have these two properties unsto ability and verifiability models can be made uh you know Unstoppable and verifiable using an AVS."

• This emphasizes the importance of unstoppability and verifiability in developing AI models, ensuring they are both reliable and secure.

• IG Layer is focused on building a universe of verifiable services, allowing for the creation of powerful AI agents.
• The goal is to enable anyone to build and mix verifiable services to achieve desired functionalities.

"What we're doing at IG layer is trying to build a universe of verifiable services and by letting anybody build these verifiable Services you can mix and match them to express very powerful very rich AI agents."

• The quote underscores IG Layer's mission to democratize the creation of verifiable services, empowering users to develop sophisticated AI agents.

• Verifiable inference is achievable through tools like AVS called opacity, which verifies API calls.
• This allows for verifiable interactions with APIs like OpenAI, ensuring transparency and trust.

"Today we already have like uh uh this tool AVS called opacity which allows you to verify an API call an arbitary API call and we actually have Integrations into to actually say oh I called openai with this query and you got this response from open a that can be verified today."

• This quote illustrates the capability of current tools to verify API calls, ensuring that interactions with external services are transparent and accountable.

• Running agents on a blockchain provides autonomy and verifiability, enabling trustless pooling of resources.
• This contrasts with centralized systems, where trust is necessary, highlighting the benefits of blockchain-based solutions.

"The diff between running it on your own on a centralized server and putting in top on top of iay is autonomy and verifiability but autonomy and verifiability give you trust which basically means you know you can just delegate a lot of actions without having to trust anybody."

• The quote emphasizes the advantages of blockchain-based systems in providing autonomy and verifiability, reducing the need for trust in centralized entities.

• The demand for verifiability is linked to the need for trustless systems, particularly in collective agents like smart contracts.
• Verifiability enhances the intelligence and functionality of smart contracts, expanding their potential applications.

"Think of I stakers come in and put in capital because they don't need to know who I am or trust us or anything and they can actually take because these smart contracts are agents of the collective."

• This quote highlights the role of verifiability in enabling trustless interactions in smart contracts, allowing for collective investment without personal trust.

• AI can transform smart contracts from simple code into intelligent agents, expanding their action space and capabilities.
• This evolution creates a significant market opportunity for more intelligent and functional smart contracts.

"With AI you actually make smart contracts smart and what is the addressable Market from that so that's that's one way to think about is there demand for verifiability is there demand for smart contracts and should smart contracts become more intelligent."

• The quote discusses the potential for AI to revolutionize smart contracts, making them more intelligent and expanding their applicability in various markets.