Peer to Peer AI: Is This Era Upon Us?

“Please consider the advent of DeepSeek as a historical pivot to the era of ‘Peer to Peer AI’.”

This quote is from Michel Bauwens’ essay “AI and the Advent of the Age of the Brahmin Workers.“ Bauwens is the Founder of the P2P Foundation, a network investigating the impact of peer-to-peer and commons dynamics in our times of civilizational transition.

Bauwens is an interdisciplinary polymath – scholar, writer, researcher, activist, and more – as well as a dear friend. Since 2005, he has been researching, cataloging, and advocating for the potential of P2P and Commons-based approaches to societal and consciousness change.

For Bauwens, and a range of AI connoisseurs, DeepSeek represents a watershed moment in AI with profound consequences. DeepSeek democratizes access to advanced AI, democratizes access to knowledge, and access to knowledge leads to a different paradigm. That is the gist of the argument.

It’s an argument worth considering and analyzing, as much as it’s worth exploring the context and references that Bauwens provides along the way – from world history and transitions to spirituality, AI, and crypto.

DeepSeek, Facebook, and Sputnik

To see the big picture, we can brush aside some of the otherwise important details on DeepSeek. DeepSeek is not the first or the only open-source AI model that rivals Big Tech’s AI models at a fraction of the cost, while what this cost is exactly is disputed.

But then again, neither was ChatGPT the first AI model that had advanced capabilities. It was the one that caught the attention of the public, aided by a barrage of media coverage. ChatGPT was AI’s Facebook moment, and DeepSeek was called AI’s Sputnik moment. Obviously, the latter analogy only works if you see AI as a (US-centric) winner-takes-all arms race.

There are a number of reasons why people might hold that view. It could be because they identify with Big Tech and the industrial–military complex around it, or because they see it as a necessary evil/the lesser of evils. Or because they believe AI is a self-catalyzing technology, meaning whoever reaches the AGI/ASI threshold first wins big over everyone else.

“If there’s one technology that has to be developed openly and transparently, by all of us collectively as humans with a shared stake in its trajectory, it’s AI.

The real choice here isn’t between winning or losing an arms race – it’s about whether we want an AI created by humanity for humanity, or an AI shaped by the cycles of conflict and domination that we need to move beyond. We should all be extremely thankful to Deepseek for enabling the former”.

This Arnaud Bertrand quote summarizes the counter-argument. In the end, it’s not so much about AI per se, but more about world views. But can we really consider this DeepSeek moment as a historical pivot to the era of ‘Peer to Peer AI’? It depends.

Peer to Peer AI

First off, it depends on what we mean by ‘Peer to Peer AI’. Certainly, it’s not P2P-produced AI. DeepSeek’s models, like most AI models, were produced by a closely knit core team working for the same employer in a commercial enterprise – albeit a Chinese one in this case. There are a number of reasons for this, which come down to expertise, supply and demand, and coordination.

The closest thing to a P2P-produced AI model would be BLOOM, developed by over 1,000 researchers from various countries as part of the BigScience project. However that model is 3 years old already, which in today’s AI is a lifetime, and there are no signs of maintenance or evolution of the model or other similarly developed models on the horizon.

But what Bauwens seems to be arguing is that AI is the vehicle to bring knowledge to the masses. First, he notes, the internet created the conditions of an explosion of available knowledge, beyond the capacity of humanity to process. Now with AI, we can re-summarize such knowledge in context.

Through DeepSeek and open source P2P AI, this knowledge can be available to the working classes, which now take the form of commons-based, cosmo-local productive communities, and are no longer the commodity-labor of capitalism.

In other words, the argument is that DeepSeek-like AI can be used to empower P2P production. Bauwens calls this ‘networked organizations with commons’, pointing to crypto communities. He refers to idealistic efforts in the crypto community, undertaken under the name of decentralization.

What crypto has achieved, he writes, is the capacity for mutual coordination of labor and money, and a way to pay for its commons. The new organizational vehicle is no longer the state or the corporation, but ‘networks with commons’, based on open contributory ecosystems, that operate cosmo-locally.

But as Bauwens points out, the workers in crypto are mostly coders and developers. They are worker-technocrats. They need to expand and coalesce with the rest of the working class. This is the role he sees for AI. I am both hopeful and skeptical of the prospect.

Network Effects and Data-Driven Feedback Loops

I don’t have the kind of ties that Bauwens has to these crypto communities that he speaks of. But based on my experience as a worker-technocrat, as well as historical precedents, I wonder whether it’s possible for communities like these to form bonds with the wider population. The misalignment in cultural, economic, and technical background seems deep.

Also, I remain unconvinced that the crypto community will be the vanguard driving us toward a better, more evenly distributed, sustainable future. I do, however, share the notion that “AI for the masses” is both essential and possible, and that worker-technocrats have a crucial role to play.

There are common threads in Big Tech effective monopolies such as Google or Facebook: network effects and data-driven feedback loops. The value of their products increases with more users, both to other users (network effect) and to providers (data-driven feedback loops).

More users equals more data, and more data equals better products. The “data is the new oil” adage iconified by the Economist a few years back remains relevant today, but open-source AI could possibly undermine the oil rigs. Unlike social media, there is no network effect in AI – at least not yet. If we reach a point where autonomous AI agents interact, that may change.

At this point, each user interaction with an AI model is a 1-1 between the user and the model. How many other users the model has is largely irrelevant to the value the user gets out of the model. In fact, the fewer users an AI service has, the faster response times may get.

But AI service providers are striving to grow their user bases for reasons that go beyond market capture. With available public data to train on having been purportedly exhausted, high-quality private data as well as interaction and feedback can be invaluable to build better models.

The track record of said AI service providers should leave no doubt that anything shared with them can and will be used as they please. In other words, the more people interact with these AI services providing them with data and feedback, the better they help AI services get. And that applies equally to OpenAI’s models and DeepSeek as a service.

Open-Source AI and Brahmin Workers

But there is another way. Rather than using AI as a service, open-source AI models enable the option of having local, personalized AI models that are private, secure, and tailored to user needs. This is not an easy feat to accomplish currently, as it requires lots of expertise and work, as well as a minimum level of hardware capabilities.

This is something that some organizations are experimenting with, for reasons having to do with privacy, control, and competitive advantage. In addition to being able to deploy and fine-tune AI models, this option also requires a structured approach to data management and governance.

Over time, we can expect the know-how and techniques to trickle down to individual power users, and eventually to a larger user base. This is something I am exploring as well, aiming to integrate it as an advanced feature of the Pragmatic AI training.

Whether that leads to a different paradigm remains to be seen. Even though Bauwens mentions the internet and the World Wide Web as an example of the democratization of access to knowledge, at the same, he acknowledges its co-option as a cautionary tale. Plus, access does not necessarily mean productive use.

That’s all fine and well, but what on Earth are Brahmin workers and what does P2P or open-source AI have to do with it? Bauwens refers to a spiritual worldview according to which the timeline of the world unfolds in repeating cycles.

The first stage of a full cycle is dominated by the spiritual class (the Brahmins), followed by a second one in which the warrior class takes over, a third one where the merchant class rises to power, concluding with a fourth one where the working class rules. Bauwens elaborates on the evolution of this worldview and argues that we are entering another transition.

What the world needs, he writes, is not just a transition to a new cycle, but a synthesis to a self-aware, spiritually driven human group. Neither Brahmins nor workers, (and not warriors or merchants either) but Brahmin-workers. I encourage you to read the essay, as it’s full of references that will expand your thinking.

Towards a Different Transition

The thesis is that for humanity to transcend its eternal cycles of ascent and descent and live in harmony with nature, we need to manage the complexity of our usage of natural resources, and our relationship with all other interdependent beings. There are two scenarios of how to do that, but Bauwens is suggesting a third one.

The first scenario is that of the Down-wingers, which want to bring down material and energy usage to levels compatible with their sustainable capacity. The second is the scenario of the Up-wingers, who want to augment the availability of matter and energy, believing technology will eventually make it compatible with natural limits, even if we have to go beyond Earth to achieve it.

Bauwens’ third scenario, which I’d call the Mid-wingers, is one in which smart mutualization keeps us as ‘modern’ and ‘complex’ as possible. For this, he argues, we need human-friendly AI, available autonomously to the productive communities, the commoners, which are the new agents of human history. The Brahmin workers.

“This transition is different: we have nowhere to go, we have exhausted the planet. The conclusion is therefore stark: we can’t just continue the cycle of extraction, collapse, dark age, regeneration, and starting over. We have to ‘transcend’ the cycle of extraction and arrive at a steady state relation with the natural world, to which we integrally belong”, Bauwens writes, and I wholeheartedly agree.

Whether crypto communities along with open-source AI are the agents of evolution that will take us there, I don’t really know. But it’s a scenario that offers some hope in an otherwise bleak-looking world, and one I’d like to see unfold and help bring about.

On the one hand, I will be sharing more on how to use and configure open-source AI. On the other hand, we are collaborating with Bauwens on a syllabus for a joint in-person workshop that will explore and try to bridge everything from world history and transitions to spirituality, local culture, and modern technology. Stay tuned for more details.