AI is still in its infancy and yet it is time to look ahead.
We can see where the present path is leading us. For this reason, focus upon every layer of the AI is required. The Web 3.0 world has to take aggressive action to decentralize any aspect it can.
This gets difficult the closer in we get to the base model. LLMs require a lot of money to train. The flipside is that inference compute is even more expensive, likely an order of magnitude.
Here is where decentralization has an advantage. Distributed computing will be crucial to handle all the prompts that come from the generative models. That said, we are looking at a great deal more than just compute.
Image generated by Grok
Decentralized AI: The Next Big Frontier
What is the goal?
When looking at the market, this is what we need: compute services that are trustless, permissionless, private, and competitively priced with current cloud offerings.
Users should not be able to distinguish between something that is operating on a centralized system versus distributed. To the individual, it simply is an application (UI) that is utilized.
All of this leads to an agentic system. As stated repeatedly, we cannot allow this to be under the control of Big Tech. The power that comes from an agentic Internet is going to be mindblowing.
From out perspective, that is near the end of the journey. There is still a lot of development required before that point. In addition to compute, we have to look at data, models, evaluation metrics, and agents. Only then can we start to consider the potentiality of multi-agent systems that span many different networks.
What is at stake?
Hundreds of trillions of dollars.
Economic singularity
We discussed the idea of the economic singularity. This is tied to the technological one that so many futurists discuss. However, one is not necessary for the other.
As with technology, the economic singularity has many definitions. This is why it will be debated for a couple decades. One view is a 10x of the traditional growth rate, thus around 30% annually.
Simple math will tell us that we are talking about an enormous amount of economic productivity. If we estimate the global economy at roughly $100 trillion, we are looking at more than $30 trillion per year. Of course, that number will keep growing.
How long until we have a $500 trillion economy? At a 30% growth rate, it will take a little over 6 years.
This might seem absurd but history has shown what happens when automation takes over on a wide scale. In fact, if we break it down, this was the premise of the Industrial Revolution. At that time, we saw machines taking over.
Between computers and robots, history is going to repeat itself. The difference is the pace which things occur. Since much of this is being driven by the digital world, we are dealing with "bit time" as opposed to "atom time".
As always, do we want this solely in the hands of Big Tech?
Communal Ownership
The solution is communal ownership.
Throughout history, we saw a lot of failed experiments with this structure. Cooperatives and other models of community stake enjoyed some success on small scales. However, when moved to larger applications, things blew up. As we moved from the local, incentives of adherence declined. Here is where outside actors operated in a self-interest outside the rest of the community. The result was others were forced to follow suit.
A prime example of this is international fisheries. A few "bad apples" decided to ignore the agreements and gather what they could. This led others to followsuit, destroying the entire ecosystem.
Fortunately, the digital world has enjoyed a bit more success. We are now in the second decade of blockchain technology being heavily utilized. As time passes, the validity of this network structure is solidified. There is still a lot of work to be done, especially in the area of privacy.
Through the use of cryptocurrency, communal ownership was established. Governance systems are also being experimented with, offering hope that we can evolve something that scales.
Blockchains take care of compute and data as both are decentralized (on public networks). This means we have a foundation for decentralized AI.
The key is going to be communally owned models. Here is where we hit a wall. A lot of this could be attributed to the development path we are following. Simply put, it is still too early to see viable models emerging.
A major component that could be missing is the incentivization to develop pre-trained models. There is a lack of compensation for model development, something that hinders funding. This is the area that centralized entities are dominating.
The Future
AI is a force multiplier.
This is why the economic singularity is a possible outcome. The multiplication factor of this technology is to be determined. As more business implement (read automate) workflows, the economic productivity is bound to skyrocket.
We are going to see reduction in human involvement. Many aspects of society will be automated. This alters the pace which things can operate. We only need to look at certain aspects of the financial world to see how computers altered that pace of things there.
Generative AI is automating cognitive abilities. Humans can only learn at a certain pace. More importantly, the speed which people can put that knowledge to works is limited.
Computers, on the other hand, are already light years ahead. Simply look at how quickly a chatbot can produce a 2,000 word essay. Even the slowest models will generate it in under 20 seconds.
Things really change as we start to incorporate AI into physical beings, such as robots. This becomes a force multiplier in the physical world. While the world of atoms will never rival that of bits with regards to speed, it will see an acceleration over what we experience today.
Which brings us back to the starting point: AI.
The need for decentralization becomes clearer by the day. People like Sam Altman want to dominate, being one of the few entities that controls this technology. The result would be massive power in the hands of a few silicon elites such as him.
To me, this is a undesireable outcome. We are at the point where the choice is life under Big Tech or decentralized AI.
I fail to see any other alternatives with how things are unfolding.