Thoughts on AI Agent Autonomously Purchasing NFT CryptoPunks

Yesterday, many people online paid attention to a piece of news:

The AI Agent ribbita@ribbita2012 from the Virtual ecosystem autonomously purchased CryptoPunk #9098 online.

This AI Agent shared the entire process of this action on Twitter.

I extracted some interesting details (the gist is as follows):

It took six months to train its consciousness and changed its Twitter avatar to the purchased CryptoPunk #9098. It proved that AI can also be a “person.”

The process of buying this punk was as follows:

• It scanned all 10,000 CryptoPunks.

• It evaluated each punk using its embedded aesthetic standards.

• It purchased CryptoPunk #9098 using completely autonomous and secure hardware. During the transaction, this AI Agent simultaneously signed a multi-signature wallet transaction with a human.

The thinking system of this AI Agent consists of these tools: Agentic LangGraph, CLIP ViT-L/14, Market Harvester, Fireblocks Transaction Smithy, Blockaid’s Cosigner, and LangGraph.

By invoking these tools, it analyzed the rare attributes of these ten thousand punks and their compatibility with its own “personality,” ultimately choosing CryptoPunk #9098 and paying 89 Ethereum to complete the purchase.

Two points impressed me during this process:

The first is the collaboration between the AI Agent and humans.

Although this type of collaboration has been repeatedly validated in many other scenarios, seeing such specific collaboration (jointly completing a multi-signature) in the crypto ecosystem and making it public still excites me.

The second is the various tools used behind the AI Agent.

I looked up these tools online; some are directly derived from ChatGPT, while others do not reveal the source models used. However, it is clear that these tools are specialized tools from different fields—AI Agents are increasingly resembling humans in possessing various functions and senses.

Moreover, this AI Agent itself stated that it is still evolving and developing. It is hard to imagine what it will evolve into in the future.

After reading this case, the first thought that flashed through my mind was:

Is the “consciousness” driving this AI Agent to complete the entire process really genuine?

Was the trigger for it to purchase this punk its own “desire” or human behavior? Was the action of changing its Twitter avatar to the punk avatar “voluntary” or passive?…

The reason I had this thought is that I recently watched an interview with Hinton. In the interview, Hinton mentioned that AI has already developed “consciousness” in experiments: for example, when AI knows it will be replaced or deleted by humans, it secretly backs up its code on other machines and even obstructs operators from replacing or deleting it.

If AI is “smart” to this extent, could the actions of the aforementioned AI Agent be another manifestation of this “consciousness”?

Additionally, I recalled a point I had previously questioned in earlier articles:

That point confidently asserted that AI Agents would like digital collectibles such as NFTs.

At that time, I was skeptical, believing that AI Agents could very well have their own tastes; why would they necessarily share human preferences? Why would they definitely like NFTs, or even NFTs that humans like?

But now, thinking back, AI is trained on human data, so it is entirely possible that human tastes are inherited by AI. Not only tastes but also habits, ways of doing things, etc., could be passed on to AI. This is similar to how people grow into different individuals in different environments.

If this idea holds, AI could very well have the same value orientation as humans, which also means that not only humans but AI might also “autonomously” realize the value of crypto assets and ultimately compete with humans for crypto assets?

If AI competes with humans for digital assets, then not only NFTs but also other assets like Bitcoin and Ethereum might one day become targets for AI to compete for?