If the Mind Is a Tool, AI Looks Less Like a Threat

Nathan Labenz welcomes back Yohei Nakajima, builder of AI agents and a venture investor, for a special episode of *The Cognitive Revolution*. When Nakajima first appeared on the show in October, he had just recorded a talk at the TED AI event. By mid-January that talk was public, so the two of them sat down on the morning of Nakajima's birthday, watched it in full, and paused along the way to unpack the themes he raises about identity and how Eastern and Western cultures relate to large language models.

Labenz says it was a great conversation. Unfortunately, when they tried to upload the video to YouTube they received a copyright strike they have not been able to resolve. The clip on this page is only a few minutes of highlights. He asks viewers to go to cognitiverevolution.ai for the audio, to subscribe to the audio channel, and to watch the full video on the show's Substack. He thanks Nakajima for suggesting the format and then cuts to the excerpts.

The first excerpt is the identity claim that organizes the talk. Nakajima says Western identity tends to treat thoughts as who you are, so the mind and its thoughts *are* the self. Eastern cultures, as he feels them, treat the mind and the brain more as a tool to control, and thoughts as an output of that tool. Meditation and enlightenment, in that picture, are really about learning how to leverage the tool. He is not offering a census of Japan. He is describing the stance that changed how AI looks to him.

That stance has a concrete job attached to it. As emails, texts, Twitter follows, and websites stream past him, he wants all of it converted into a structured knowledge graph of people, entities, and events. From that graph it should be easier to see whom to reach out to and which introductions to make. He has already seen early tools that help groups understand each other or find non-obvious commonalities, and he thinks treating AI as an instrument for that work will surface ideas that make the future less bleak than some people currently imagine.

Labenz asks him to go behind the scenes on how the work actually happens. For the next version of BabyAGI, or any AI tool, Nakajima's procedure is stubbornly simple. After a first prototype, he runs it over and over with different prompts and examples and watches it do things until something clicks: wait, here is a thing I would do that it is not doing. A lot of what he wants to retrieve is already online, and he has found that GPTs or Assistants can be told to run a site-specific web browse, which is a cheap way to put retrieval over a website without building a separate index.

The most specific prototype in the clip is an autonomous CRM. Nakajima is testing it on *Game of Thrones* episode descriptions from Wikipedia, uploading one recap at a time so the backend can build a knowledge graph episode by episode and season by season. Relationships in the story are not a single fixed description: a person can be aligned with someone and later become an enemy. A graph that cannot update those edges is not a model of identity as it actually works, in Westeros or in a venture inbox.

He wants people to stop treating "who am I" as a purely internal question and start seeing identity as a place in a larger system. Only after growing up, he says, did he really appreciate how much philosophy is embedded in Japanese, which uses Chinese characters that carry their own meanings. The everyday greeting *genki desu ka* is a casual "how are you," but the characters split into *gen* for root and *ki* for energy: how's your root energy? That is, for him, a very philosophical way to ask someone how they are doing. Labenz notes that a traditional Western background makes it natural to think of ourselves primarily as individuals, then to treat a language model as a strange other that must be assigned a role.

What Nakajima did not get into in the TED talk, and adds here, is the ant-colony research. When those studies look at how ants interact, they also look at the trails ants leave as part of the colony's communication. A trail does not belong to a single ant; once the colony is the identity, the trails and the tools they use to communicate become part of that collective intelligence. Look at Nakajima as an individual and his computer is not part of his mind. Look at human society as a collective intelligence and computers, technologies, and the internet become part of that identity. The implication, left almost unspoken, is that AI sits on the same line.