Introduction.
Very often, I find myself writing articles a bit too early. Sometimes a couple of years ahead, sometimes even more. But I think that is entirely normal when you are dealing with and researching emerging technologies. For me, it is fine. I am a practical futurist, scientist, and visionary. Also, apparently, a Dexterity Leader. I hope it is fine for you as well.
So, as you are reading what I write, think ahead and keep an open mind, please.
The majority of technically “weird” things are actually disruptors – and they are already happening. You can often see them, sometimes you even used them already; the rest may be hidden under the bonnet of the digital space ecosystem.
In this article, I am going to talk about Federated Learning – and why decentralised AI may shape the future of intelligent systems.
The Shift from Centralised to Decentralised AI.
Let’s start with a simple idea: the majority of AI models and systems we know today mostly live in a super gigantic servers and the cloud. They are trained on massive datasets collected and stored in central locations. I mean really big data centres owned by big companies.
This is working solution, but it comes with challenges. I mean privacy issues, huge energy demands, and a bottleneck on innovation that is tightly controlled by a handful of so called “actors” and “players”.
Now try to picture a different approach.
What if AI could be trained collaboratively, without moving all that data to one place? What if the intelligence was distributed by learning at the edge, closer to the user, on devices like laptops, phones, sensors, or even satellites? Furthermore your private data will stay private?
Am I dreaming or “Hallucinating”, I do not think so!
That is where Federated Learning (FL) comes in.
What Is Federated Learning?
Federated Learning is a machine learning technique that allows models to be trained across multiple devices or servers holding local data samples – without ever sharing that data.
Simply we do not centralise sensitive data!
Instead of uploading everything from edge devices to the cloud, the model is “sent” to the data. Each device trains the model on its own data and sends back only the updates. These updates are then aggregated and used to improve the global model.
In simple terms: your phone helps train AI without ever giving away your private data. Multiply that by millions of devices, and you get a collective intelligence that is both smart and privacy-conscious.
In other words, Training is orchestrated across distributed nodes while keeping your data localised. Such an idea dealing with serious privacy concerns and regulatory requirements (GDPR etc). It still delivers all benefits of collective intelligence and distributed datasets.
You may say “Decentralised Collaboration”, and you are right!
Win-win IMO, especially for a bit smaller “players” like you and me.
One step further: Why not to use Blockchain-Based Federated AI Model Training?
For sure, we can integrate AI into blockchain. It may raise the questions of how to manage AI models and training data in a decentralised environment?
We need to remember that current “traditional” AI model training process is centralised. That means a single platform gathers a very large datasets and trains a model, or models on super-powerful servers as I mentioned before.
Blockchain offers a new way for collaborative model training and uncompromisable data governance.
Instead of using one central repository, data owners (which could be individuals or organisations) can contribute to training an AI model without directly sharing their raw data.
Blockchain can coordinate this using smart contracts (SC). Partners might train a model on their local data and submit only model updates in form of “gradients” or “parameters” to the blockchain.
Such a blockchain based approach can than aggregate these partial contributions in order to improve a ultimate model. It is done by using a technique named federated averaging.
By doing so, the distributed ledger keeps an immutable record of who contributed, when, and how.
Federated-AI model can be trained on multiple participants data without needs to showing others “raw data”. It sounds like win-win again for everyone involved. And that is a beauty of Decentralised AI idea.
Also, one more benefit from using the blockchain is fact that AI training trust is managed by consensus of the particular blockchain. That means no single party or player can cheat or compromise the records, because every update is verified by nodes operators, and agreed on the truth before permanently recorded inside a block of blockchain.
Similar like with crypto and blockchain based transactions, I guess. I believe that some of you are very familiar with that topic. For many of you it is not new at all.
Why This Matters.
Federated Learning tackles some of the biggest challenges in AI development today:
Privacy: Data stays where it belongs. That is a big win in industries like healthcare, finance, and education where data sensitivity is critical.
Security: Less data transmission means fewer chances for interception or breaches.
Personalisation: Models can be fine-tuned locally, learning from individual user behaviour to offer smarter, more relevant outputs.
Scalability: Training happens at the edge, massively distributing the workload and reducing reliance on central infrastructure. It also reduces the “ping” and latency in response.
All of these points lead to a much more democratic, fair, and decentralised AI ecosystem. There one where intelligence can be cultivated and emerge from almost anywhere. I mean from me and you, whatever base in Enugu, New York, or Krakow, not just a server farm in Silicon Valley or Big Tech.
From FedAvg to Swarm.
I will risk adding some more detailed and weird terms and hope you will not “hate” me for that. I just like to keep you aware of what is happening and learn this interesting stuff just as I am learning.
Federated learning has seen some big improvements lately.
It started with FedAvg, the foundational method that lets devices train models locally and then combine their progress. It is efficient but with some limitations.
Then, FedProx built on this by making sure updates from each device do not drift too far from the global model, which is especially useful when data varies a lot between users.
After that FedNova came in to normalise those updates, helping models converge faster even when devices train for different lengths.
Finally, FedMA pushed the boundary further by matching similar features across neural networks, leading to more accurate shared models.
A major shift has been toward personalised federated learning.
Since data across devices is rarely identical, global models often do not perform as well as local ones.
To solve this approaches like FedPRL uses smart client selection and tailored strategies to better adapt to individual data.
PeFLL goes even further, generating ready-to-use personalised models with minimal extra training, saving time and computing power.
Infrastructure-wise, federated learning is moving beyond central servers. Peer-to-peer and swarm architectures offer more privacy and resilience, with Swarm Learning using blockchain to securely share insights without sharing raw data.
All actions and approaches lead to a more decentralised, privacy-preserving AI future.
I hope I do not overload this part with too many technical terms. My intention is to highlight the progress which is happening beyond our sight.
It is good to know and be fully aware IMO.
Real-World Use Cases.
You may already be benefiting from Federated Learning without even realising it.
Google uses it to improve predictive typing and auto-correct on Android devices.
Apple applies similar techniques to power Siri and keyboard suggestions.
I have read that some hospitals are starting to explore Federated Learning (FL) to build diagnostic models across different medical centres locations. All without sharing strictly private patient records. Awesome idea, I fully support that way of thinking.
As I mentioned before this is not a “sci-fi” or likely story of mine. It is fact happening around us, and now.
And it is not limited to tech giants only. Smaller startups and researchers are exploring federated learning for applications in health care, agriculture, smart cities, autonomous vehicles, and even space tech.
I am working on a project named LedgerMed and I am proposing such a solution in my project.
Decentralised AI & my Broader Vision.
Federated Learning is a big step toward Decentralised AI, but it is just one piece of the Lego, or you may think puzzle. We are slowly moving toward a world where AI may not be controlled by a super central power but exists as a network of intelligent agents-distributed, resilient, and adaptive.
This decentralised vision aligns with Web3, DePIN, edge computing, blockchain, and other decentralisation trends that are pioneering and reshaping digital infrastructure.
Together, they point to a future where AI is more collaborative, ethical, and embedded in the real world. Our world, the world we are living and proudly are still active part of it.
It is not trapped behind a high level of closed doors and so called “The Corridors of Power”.
Let make a toast for Decentralised AI and Federated Learning.
Someone said the next revolution will be decentralised.
Cheers, The Author.
Oryginally published at: