There is a new idea that tours on how to organize our LLM models and our new technologies to better take advantage of their power. It is not yet very common, because people always get used to the idea of AI in general, but it appears in certain areas of the academic world and business.
One of these places is Mit Media Lab, where researchers continue the work of grown -ups like Marvin Minsky, Seymour Papert and others. This is actually the 40th anniversary of the laboratory this year, and what we see is that this research institution offers very good ideas on how to go forward at the time of AI .
One of these ideas is to decentralize AI platforms. This is something I heard about in the context of planning the next generation of services and deployments. What is decentralized AI, and why is it important?
The power of decentralized AI
Some of the best defenders of decentralized AI stress that in current centralized and monolithic models, some large companies dominate the use of data to supply systems.
One of these analysts is my friend and colleague Ramesh Raskar. By pleading for a decentralized AI, he emphasizes that companies centralize data, calculation and even governance, in a worrying way, and that they often do not want to share data in a constructive manner.
He quotes “suspicious, disconnected and disinterested” parts without a real incentive to work together.
Against this idea, supporters of the decentralized AI presented the possibility of a “mixture of experts” where the interaction between learning to strengthen and the supervised fine adjustment creates a workflow for success. This, some suggest (including In this Venturebeat article) is the place where Deepseek has arrived – using the brighter aspect of RL to jump on LLM efficiency and bypassing SFT.
But beyond the model of Deepseek model, there is the idea that companies at the forefront of innovation are themselves vulnerable. This gray level article Underlines how, just after Deepseek’s announcement, the company was the victim of a hack which represents the kind of thing that decentralized AI could potentially resolve.
Four decentralized AI pillars
When you listen to Ramesh or someone else who talks about the principle of decentralized AI, they indicate four global objectives for these processes, as follows:
Confidentiality – How do you keep the data of a safe and private individual?
Incitations – What are the incentives to the different parties to work together?
Check – How do you check if someone is a good actor or not?
Dashboard – There must be a kind of interface for this type of collaboration.
I also met this interesting analogy for the decentralized AI that has to do with the development of the Internet. “Web One” was this series of websites and pages that make up global internet. “Web Two” was social media. “Web three” is the blockchain, a truly decentralized system in which nodes and components interact with sovereignty.
What’s going on in a decentralized design?
Decentralization and history of AI theory
Another way of thinking about it is to return to the old days when people first theorize the concept of AI itself.
For example, the idea of ”decentralized AI” echoes part of Marvin Minsky’s work in his book, “Society of Mind”. In this document, he suggests that the best AI system should work like the human brain, which is essentially not a single computer, but a series of interconnected computers working together.
Regarding risks, those who try to prepare for decentralized systems emphasize that some of these constructions can be vulnerable to accidents or hostile controls in the event of an attack of 51%.
On the other hand, they can accelerate complex membership if the system is intelligent enough to control agents to ensure that they are trustworthy.
And it is, in large part, what it comes from: confidence. We may each have our own individual AI agents acting in our name – or we could simply have large networks of non -playing characters in real life, so to speak. We must be able to have a framework to establish confidence, and the rest will be set up.
“There are billions of agents. They all want to talk to each other, but they don’t trust each other. They don’t even know where they are. They don’t even know what they are doing. As if I meet you today…. If I meet someone else, I don’t know … their names, I don’t know their expertise and how to determine whether we should collaborate or not, right? – Ramesh Raskar
Think about the possibility of decentralized AI systems for all that we now design – from recommendation engines to autonomous driving, insurance and loan systems, or AI planners for smart cities. With artificial intelligence, we operate at a level of view of the bird, including the complete image and all the details at the same time. It is a powerful thing, but it must be exploited and deployed correctly, which is why many of these experts have another suggestion:
It is because we must first deploy our pioneer systems in low -challenges. Then we continue to work on them until we can trust them, and it is only then that they can be put in critical mission systems. So let’s remember this while we are moving forward.
