Author: Johnson Lai
Compiled by: TechFlow
Recently, we have witnessed the craze for AI agent tokens and AI meme coins on Crypto Twitter. But this also raises an important question:
We are living in a strange era - an era where we are struggling to prove that something is AI and not human. Think about it: we spent decades proving we are human through CAPTCHAs and Turing tests, and now we are doing the exact opposite.
As more and more AI agents emerge every day, we start to question, "Wait, is this really AI or is someone pretending?" This reversal of the situation is both ironic and fascinating, especially as these AI agents become more deeply embedded in our daily digital lives.
We also noticed that Apple has adopted a blockchain technology to enhance the transparency of its Apple Intelligence.
What are AI agents?
In simple terms, AI agents are artificial intelligences with tool and reasoning capabilities. In the crypto space, these agents may be able to access tweets, posts, and even have crypto wallets to perform on-chain transactions.
It all started with the ReAct paper, which demonstrated that AI can handle complex tasks when equipped with the right tools. Subsequently, the emergence of Langchain made these tools more accessible to engineers, leading to the rapid proliferation of AI agent tools.
Image:ReAct Prompting
The unique aspect of AI agents is their ability to:
Reason about specific contexts
Formulate action plans
Execute actions using existing tools
Adjust based on results
Actual applications like the terminal of Virtual's Luna (https://terminal.virtuals.io/) demonstrate these capabilities, allowing users to observe the agent's reasoning and decision-making process in real-time.
The evolution of AI agents
Achieving truly transparent AI agents requires following these steps:
Open-source code
Open data
Decentralized reasoning
Autonomous AI agents
I believe the current state of AI agents has reached at most stage 3: open reasoning.
Stage 0: "Trust me, I'm AI"
Currently, most AI agent projects are still at this stage. Users interact with systems claiming to be AI agents, but cannot verify:
Whether the responses are from genuine AI reasoning
The actual decision-making process
Whether there is human intervention
This leads to the "black box" problem, where users can only rely on trust to accept the claimed AI capabilities.
Stage 1: Open-source code
Open-sourcing the code of AI agents is a key step in establishing transparency and trust.
By opening the code, developers can audit and contribute to the project, fostering collaboration and innovation.
Stage 2: Open data
Example 1: Virtual's Terminal
Example 2: Using Chromia to record large language model data on-chain
Simply open-sourcing the code is not enough; we also need to ensure the agent's "thought" process is transparent.
This critical stage introduces transparency in the agent's thought process and decision-making. Projects at this stage typically:
Open their training data
Provide access to the agent's reasoning logs
Allow external auditing and verification
Stage 3: Decentralized reasoning
Open data alone is not enough to ensure genuine large language model reasoning. Decentralized reasoning solves this by:
Using zkML/opML to verify model execution
Providing a transparent and auditable reasoning process
Allowing community verification
Stage 4: Decentralized AI operating system
The ultimate vision is to have autonomous AI agents running in a transparent ecosystem, which includes:
Independent runtime or operating system
Crypto wallets for the agents
Autonomous decision-making capabilities
Comprehensive observability of resource management
Auditable decision paths
As AI capabilities and autonomy continue to grow, this transparency becomes increasingly important, allowing us to understand and verify their behavior at all levels. The market is still exploring how to achieve this goal.
Ongoing progress
We are just getting started, but some notable projects have emerged:
$GOAT: Building an AI meme culture community
$LUNA: Creating AI agents on the blockchain
$GMIKA: Personalized AI companion
$CAI: Running a decentralized AI network via opML
$AI16Z: DAO for AI funds
Conclusion
Imagesource
We are at the beginning of a major transformation. While Vitalik's article perfectly captures the synergistic operation of crypto and AI, who would have thought that this fusion would start from an AI meme coin called $GOAT? This unexpected beginning showcases how unpredictable and vibrant the future of Crypto x AI can be.
