The Internet is only nearly 25 years old. In my mind, the Internet has been normal since Google appeared in 1998. That was 26 years ago. Since it was born, there have been a lot of exciting things that were hard to even imagine until they took off and finally came. In the first 10 years of the Internet, people came up with blockchain and cryptocurrency. They later figured out how neural networks and machine learning could be used effectively to solve complex modern problems. If you don’t know, the neural network finds patterns in arrays of information and builds on these patterns to create new information that we need.

Development of AI model

In general, a neural network is a model—a set of files representing algorithms and tables that together form what, in some sense, can be called artificial intelligence. Development is taking place in several directions, including developing more optimal algorithms and increasing the size of the tables.

The latest existing models are of such a size that it would not take 1000 years to train this on a single video card. However, all models today are centralized, and it is the responsibility of a specific person or organization to train them. So only the TOP players like Amazon, Google, Facebook, and Microsoft can afford to train the biggest models nowadays. Ordinary people have to make do with models that are, on average, 100-1000 times smaller than the top models available to these companies.

The Concept of Decentralization

In 2008, the first decentralized blockchain appeared. The essence was that any user could receive virtual coins to help maintain the virtual network at the expense of computing resources on his or her computer. The properties of this network also allowed the transfer of these coins to each other, which allowed this network to eventually grow into a whole financial system based on mining, decentralization, and encryption.

The pros of this approach primarily represent the possibility of cross-border, uncensored payments that occur with nothing more than the Internet and the processing power of the network's computers.

Think about it—there are only 2 factors—the Internet and computer processing power. That's all it took to create a whole new financial system, and that's just the first 10 years of the Internet.

Decentralization of AI

Decentralization means the absence of the main person or entity. The main thing about decentralization is the consensus among the smallest units of the system. With this consensus, we can build more than a new financial system.

Let's imagine for a moment that mining could create something more than just wasting resources to keep the network running in search of a random number. What if each miner computed a small part of a huge artificial intelligence model on his working machine without having to train the whole model? The essence of this idea is to efficiently distribute the training, using the computing resources of different machines - from video cards to ASICs, to obtain a fully complete model or even for its constant additional training on new and new data sets.

Miners would receive certain coins to train the model, which could be used in the future to use the computing resources of the AI model. At some point, the model would be fully trained, and it would seem that anyone could download it to a computer for free, but they would never have the resources to run it themselves. The same mining is required to run the model. One person requests the network, and the network issues a response. It will be like working with the ChatGPT-4 API. Each request would cost the user a few tokens, depending on the rates of the miners. The miners collectively decentralize these requests and produce the results.

There could be a myriad of models, and they could all exist on a single blockchain in the form of smart contracts. Namely, someone—a person or a group of people—creates an AI as a smart contract in the blockchain network, attracts people to train it, and then earns money together to keep the model running.

Implications of Decentralized AI

If such a system is created, the largest companies will lose the status of monopoly on owning the largest AI models in the world. Very soon, the largest AI models will exist in a decentralized way and in no other way. Joint use of huge amounts of computing resources will very soon produce more useful work in total than separate supercomputers of TOP IT companies. This is the next step in the development of technology and the next step towards full autonomy of everyone, rather than dependence on the strongest.

The flow of technology available today is gradually increasing the capabilities of everyone on the planet. Over time, everyone's autonomy will only increase.

We can get endless amounts of information from the Internet on any topic we want. We can get energy in our homes thanks to solar panels. Decentralized money came unexpectedly to everyone and many people have not even understood its essence yet, but the consequences are huge.

Perhaps in the future, there will even be an analog of the free, decentralized Internet. And the future artificial intelligence will be in everyone's home and will not belong to anyone, helping with household chores or working in your garden. Now we have entered a time when we need to develop these decentralized technologies and just let artificial intelligence grow and grow exponentially larger and larger, just like the computing power of the Bitcoin network is growing.

What do I think about Internet Computer Protocol (ICP)?

ICP is pushing boundaries in decentralized AI with some genuinely unique features that aren't just hype—they're solving real problems.

First is data privacy. In the traditional AI world, data is usually funneled through centralized servers, which can be a goldmine for breaches. ICP changes the game by processing data across a decentralized network of nodes. No central server means your data isn’t sitting in one big vulnerable target. This setup is particularly important for sensitive applications like healthcare AI, where privacy isn’t just nice to have—it’s mandatory.

Another area where ICP shines is in its governance. Most AI platforms are controlled by a handful of big players, leading to decisions that prioritize profit over progress. ICP flips this on its head with a community-driven governance model. Developers and users have a real voice in how the network operates. It’s not just lip service to the idea of decentralization; it’s putting power back into the hands of the people who actually build and use these AI tools.

Then there's the technical side—ICP’s architecture supports what you might call "true decentralization". Many so-called decentralized AI platforms still rely on traditional cloud infrastructure at some level. ICP eliminates the middleman, running AI applications directly on a decentralized web. This isn't just an incremental improvement. It’s a foundational shift that reduces dependency on any single provider and ensures continuous uptime.

And let's talk about speed and efficiency. AI workloads are demanding, and ICP's ability to scale computation independently from storage is a big deal. In essence, ICP is pioneering a new way forward for decentralized AI by addressing the core issues—privacy, governance, true decentralization, and technical flexibility—that have held back the field.