It started with a noisy desk. The desk was a wooden cubicle in a lab at Northumbria University, in northern England, where a young AI researcher began his PhD track. This was in 2015. The researcher was Ben Fielding, who had built a large machine stuffed with early GPUs to develop AI. The machine was so loud it annoyed Fielding’s lab-mates. Fielding crammed the machine beneath the desk, but it was so big he had to awkwardly stick his legs to the side.
Fielding had some unorthodox ideas. He explored how “swarms” of AI — clusters of many different models — could talk to each other and learn from each other, which might improve the collective whole. There was just one problem: He was handcuffed by the realities of that noisy machine beneath his desk. And he knew he was outgunned. “Google was doing this research as well,” Fielding says now. “And they had thousands [of GPUs] in a data center. The things they were doing weren’t crazy. I knew the methods… I had lots of proposals, but I couldn’t run them.”
Ben Fielding, CEO of Gensyn, is a speaker at Consensus 2025 in Toronto.
Jeff Wilser is the host of The People’s AI: The Decentralized AI Podcast and will host The AI Summit at Consensus 2025.
So a decade ago, it dawned on Fielding: Compute constraints would always be an issue. In 2015, he knew that if compute was a hard constraint in academia, it would absolutely be a hard constraint when AI went mainstream.
The solution?
Decentralized AI.
Fielding co-founded Gensyn (along with Harry Grieve) in 2020, or years before Decentralized AI became fashionable. The project was initially known for building decentralized compute – and I’ve spoken with Fielding about this for CoinDesk and on panel after panel at conferences – but the vision is actually something wider: “The network for machine intelligence.” They’re building solutions up and down the tech stack.
And now, a decade after Fielding’s noisy desk annoyed his lab-mates, the early tools of Gensyn are out in the wild. Gensyn recently released its “RL Swarms” protocol (a descendant of Fielding’s PhD work) and just launched its Testnet — which brings blockchain into the fold.
In this conversation leading up to the AI Summit, at Consensus in Toronto, Fielding gives a primer on AI Swarms, explains how blockchain snaps into the puzzle, and shares why all innovators — not just tech giants — “should have the right to build machine learning technologies.”
This interview has been condensed and lightly edited for clarity.
Congrats on the testnet launch. What’s the gist of what it is?
Ben Fielding: It’s the addition of the first MVP features of blockchain integration with what we’ve launched so far.
What were those original features, pre-blockchain?
So we launched RL [Reinforcement Learning] Swarm a few weeks ago, which is reinforcement learning, post-training, as a peer-to-peer network.
Here’s the easiest way to think about it. When a pre-trained model goes through reasoning training – like DeepSeek-R1 – it learns to critique its own thinking and recursively improve against the task. It can then improve its own answer.
We take that process one step further and say, “It’s great for models to critique their own thinking and recursively improve. What if they can talk to other models and critique each other’s thinking?” If you get many models together in a group that can all talk to each other, they can start learning how to send information to the other models… with the overall goal of improving the entire swarm itself.
Gotcha, which explains the name “Swarm.”
Right. It’s this training method which allows many models to kind of combine, in parallel, to improve the outcome of a final meta-model that you could create from those models. But at the same time, you have every single individual model just improving on its own. So if you were to come along with a model on a MacBook, join a swarm for an hour and then drop back out again, you would have an improved local model based on the knowledge in the swarm, and you would have also improved the other models in the swarm. It’s this collaborative training process that any model can join and any model can do. So that’s what RL Swarm is.
Okay, so that’s what you released a few weeks ago. Now where does blockchain come in?
So the blockchain is us moving forward some of the lower-level primitives into the system.
Let’s just pretend that someone doesn’t understand the phrase “lower-level primitives.” What do you mean by that?
Yeah, so I mean, very close to the resource itself. So if you think about the software stack, you’ve got a GPU stack in a data center. You’ve got drivers on top of the GPU. You’ve got operating systems, virtual machines. You’ve got all this stuff going up.
So a lower-level primitive is the closest to the bottom foundation in the tech stack. Am I getting that right?
Yes, exactly. And the RL Swarm is a demonstration of what’s possible, basically. It’s just a somewhat hacky demo of doing really interesting large-scale, scalable machine learning. But what Gensyn’s been doing for the past four-plus years, realistically, is building infrastructure. And so we’re in this period now where the infrastructure is all at that v0.1 sort of beta level. It’s all done. It’s ready to go. We have to figure out how to show the world what’s possible when it’s quite a big shift to the way people think of machine learning.
It sounds like you guys are doing a lot more than decentralized compute, or even infrastructure?
We have three main components that sit underneath our infrastructure. Execution – we have consistent execution libraries. We have our own compiler. We have reproducible libraries for any hardware target.
The second piece is communication. So assume you can just run a model on any device in the world that’s compatible, can you get them to talk to each other? If everybody opts into the same standard, everybody can communicate like TCP/IP from the internet, basically. So we build those libraries and RL Swarm is an example of that communication.
And then, finally, verification.
Ah, and I’m guessing this is where blockchain comes in…
Imagine a scenario where every device in the world is executing consistently. They could link models together. But can they trust each other? If I connected my MacBook to yours, yes, they could execute the same tasks. Yes, they could send tensors back and forth, but do they know that what they send to the other device is actually happening on the other device or not?
In the current world, you and I would probably sign a contract to say, yes, we agree that we’ll make sure our devices do the right thing. In the machine world, it needs to happen programmatically. So that’s the final piece we build, cryptographic proofs, probabilistic proofs, game theoretic proofs to make that process entirely programmatic.
So that’s where the blockchain comes in. It gives us all of the benefits of blockchain you can imagine, like persistent identity, payments, consensus, etc. And so what we’re doing with the testnet now is taking RL Swarm and the primitives of the other infrastructure and we’re adding in the blockchain components and saying, ‘Hey, when you join a swarm now, you have a persistent identity, which exists out there on a decentralized ledger.’
In the future you’ll have the ability to make payments, but right now, you have that trust consensus mechanism where we can terminate disputes. So, it’s kind of an MVP of the future Gensyn infrastructure, where we’re going to add in components as we go.
Give us a tease of what’s coming down the pipeline?
When we reach main-net, all of the software and infrastructure is live against blockchain as the source of trust, payments, consensus, etc., identity. This is the first step of that. It’s adding identity in and saying when you join a swarm, you can register as the same person. Everyone knows who you are without having to check some centralized server or website somewhere.
Now let’s get wild and talk further in the future. What does this look like one year from now, two years from now, five years from now? What’s your North Star?
Sure. The ultimate vision is to take all of the resources that sit under machine learning and make them instantaneously programmatically accessible to everyone. Machine learning is heavily constrained by its core resources. This creates this huge moat for centralized AI companies, but it doesn’t need to exist. It can be open-sourced if we can build the right software. So our view is Gensyn builds all of the low-level infrastructure to allow that to get as close to open-source as it possibly can. People should have the right to build machine learning technologies.
