Helping 100 countries build a model – Stability.AI controversial CEO ambitions

Original author: Wang Chao

原文来源：Empower Labs

On March 23, the controversial CEO of Stability AI, Emad Mostaque, announced his resignation and will devote himself to the cause of decentralized AI in the future. This news caused quite a stir in the market.

On the one hand, Stability.AI has carried the banner of open source AI alone and made important contributions to open source AI, and on the other hand, he has been constantly accused of taking the results of other research teams on his own. CEO Emad's reputation has been further tarnished by the exposure of false academic qualifications and frequent big words on social media. Once had shining investor support and huge financing, Stability.AI has long been in deep financial trouble, and his life hangs by a thread.

After his resignation, Emad revealed that he still held a majority stake in Stability.AI, enough to control the board. Rather than being forced to resign, he decided that in the field of AI, the high concentration of power is harmful to everyone, so he chose to resign to promote a change in the Stability.AI. Because of Emad's disgraceful history of bragging, most people think it's more than simple. But more than Stability.AI, I'd like to talk about the decentralized AI that Emad is going to devote himself to.

A few weeks ago, I joined Emad for a discussion around decentralized AI. After that, I sorted out many of his past speeches on this topic, and roughly sorted out his perspective.

Whoever controls the model controls the mind

If human actions are driven by an operating system, AI is fast becoming an external core component of that operating system. Because humans have become accustomed to outsourcing the burden of thinking to AI, this technology forms part of the way we think. However, the convenience and empowerment brought by AI also comes with significant risks - whoever controls the AI model controls the thinking of the world to a certain extent.

If there is a lack of awareness of how these smart tools work and what the defaults are, our decisions and opinions can be silently affected. The controller of the AI model can subtly guide people's choices, opinions, and behaviors by setting specific default options. AI as the infrastructure of the next era can be catastrophic if it is controlled by only a few business groups. This is what Emad emphasizes the importance and urgency of decentralized AI.

Each country needs to have its own model

When OpenAI invests a lot of energy into the Super Alignment project, who will be responsible for the alignment of OpenAI itself and every country, every industry, and every culture around the world?

No one.

OpenAI's Super Alignment has made a lot of efforts in terms of basic security and common human ethics, but is such an effort sufficient in the face of the diversity of different countries and cultures? Different nationalities and cultures often have values that are very different from those of Silicon Valley's elites, but can these diverse values be fairly reflected in AI models?

The answer is not optimistic. Therefore, Emad believes that every country, every industry, and every culture should have an AI model that represents its own characteristics. These models should be deeply rooted in the local area, absorbing and embodying the collective wisdom of the country, industry and culture. This concept should be familiar to everyone, because two months ago, Nvidia also talked about the concept of sovereign AI on various occasions, which is essentially a thing. However, Emad started talking about this on various occasions a year or two ago, much earlier than Nvidia.

The vast majority of countries in the world simply simply don't have the ability to create their own AI models, and that's exactly the market that Emad is targeting. He hopes to create an underlying stack to support AI models created by each country, nation, and industry. On the stack, it is hoped that the development of the model will be realized in a decentralized collective collaboration.

Emad has said that it is possible to launch/incubate a series of companies, each with different professionals focused on different key areas such as education, healthcare, finance, and of course, AI models for different countries. And as a practice of decentralized AI, these companies play more of a launcher role. By providing a basic model and a standardized framework, we can introduce community talents to participate in contributions. If a large number of talented people in a country can be attracted to participate, these collective wisdom will eventually converge into an excellent national model.

At its core, it's data

To use a simple analogy, the recipe of an AI model is algorithms and data, and then use some computing power to stir them together, the more data, the more computing power that needs to be stirred up this data. At present, the vast majority of teams in the market are pursuing better model algorithms, getting more data, and then matching it with more computing power. However, it has been proven that with high data quality, it is possible to achieve excellent results with less data. In other words, people are wiping their butts with low-quality data by heaping hash power.

This constitutes one of the advantages of the decentralized AI system that Emad promotes. He believes that if a structure can be created to bring together the best people in a country, high-quality national datasets can be assembled that are verifiable, have clear ownership, and can be used to design incentive models around data.

In this way, we can collect data that we wouldn't have had access to in the past. Not only is this data of higher quality, but it also provides a more authentic and unbiased representation of the voices and needs of the public.

Small model clusters vs single large models

In the AI space, scaling laws have become almost an iron law, and whether we choose to pursue it or not, there is no way to avoid it.

Obviously, adopting a decentralized approach to organizing resources and then implementing artificial general intelligence (AGI) through Scalling Law is not realistic in the short term. For quite some time, community-oriented AI models will also struggle to compete with giants like OpenAI for the crown of the most powerful model.

However, the pursuit of AGI and the creation of AI that can be widely applied are two different things. As technology continues to advance, community-driven small and medium-sized models are rapidly gaining capabilities. It is expected that within a year or two, small and medium-sized models will be sufficient for most daily tasks. Maybe it's not the strongest, but it's practical enough and low enough to open up a wide range of application scenarios. Just like most of the online purchases we buy do not require SF Two-Day Delivery, the mixed use of models will gradually become mainstream in the future.

This brings about a very important change, as when collective intelligence-driven models are widely used, greatly reduces the potential risk of a single large model being controlled by a single agency. If the data of a large model is contaminated, these collective intelligence-based models can easily act as calibrators to make the necessary corrections. It's not just about practicality and cost savings, it's about the game between collective intelligence and AI God.

From a technical point of view, the small model is not exactly a disadvantage either. The small size makes it easier for them to do additional training for vertical areas. In these areas, they may not be as comprehensive as large models, but they perform well as expert tools. And a small model cluster composed of many expert models will not necessarily lose in the competition with a single large model.

More importantly, the small model can effectively promote the decentralization of deployment. When we talk about decentralization, we are not only referring to the construction of models and the decentralization of data sources, but also the decentralization of governance and the decentralization of deployment. If the open-source model can be easily deployed on an individual's laptop or even mobile phone, this will constitute AI equality. Even if a centralized service provider shuts down the service, users can still rely on local AI to continue operations. Making AI accessible to all audiences is also an important goal of decentralized AI.

AI + Web3 – Scammers or the Future

There is no doubt that the decentralized AI initiatives that Emad is actively promoting are closely related to crypto. He has made it clear that he wants to design a Web3 protocol that integrates and implements these ideas. That's because a few key elements that are currently lacking in the AI space – data verifiability, data ownership, large-scale coordination and incentives, and collective governance capabilities – are where Web3 technology excels.

Here I would like to highlight governance. Because there has never been a technology as powerful as artificial intelligence today, and that is about to impact all corners of the world broadly and deeply. Who should decide the future direction of this technology? Who can effectively control it? Having the boards of directors of a few companies, such as OpenAI, govern AI is certainly not the most effective option. Simply putting some hard shackles on AI models by regulators may not be effective in addressing the challenge. Collective governance may be the real solution.

In the field of Web3, the experiment of collective governance is flourishing, covering multiple levels such as data governance, application governance, network governance, and organizational governance. While most attempts are still in the exploratory phase and have experienced many failures, it is at the forefront of the development of human governance.

Over the past five years, the crypto space, especially around decentralized autonomous organizations (DAOs), has experimented with almost every governance model in human history. OpenAI's innovative structure of non-profit foundations controlling for-profit companies has been widely practiced in DAOs. In my opinion, people in the Web3 world have been playing a governance speedrun game for the past few years. In just a few short years, people have reenacted thousands of years of human governance.

One of the most common criticisms is that most Web3 governance is nothing more than a replication of the governance model that humans have had in the past, and then an on-chain vote is added. However, history has taught us that when there is such a fast growth rate and a high density of talent, completely new things will soon evolve.

A less apt example is internet advertising. I remember the early years of the Internet, when I opened a news website, a huge full-screen ad would suddenly appear and then slowly fade, and the web page was covered with dense ad blocks, which became a trademark memory of the early Internet. Because people didn't know what a more effective way to promote the Internet at that time, they simply moved the advertising methods on the traditional media to the Internet. However, with the progress of the scene, with the deepening of people's understanding of Internet technology and culture, they soon evolved an efficient promotion model that has never been seen in human history, and quickly eliminated traditional advertising on the Internet.

It seems to me that the governance of technology will also follow a path similar to that of Internet advertising. I am confident that blockchain technology will bring about a vastly enhanced coordination and governance capacity, from which a collective governance scheme that has never been seen before in human history will grow.

Write at the end

It is difficult to come to a conclusion when we look at the future of the AI era and look at Emad's entire vision. Clearly, his plan faces enormous challenges on many levels. Coupled with the fact that Emad used to exaggerate a lot, it's not easy to distinguish between what he says can be taken seriously and what is just casually said.

However, exploring the power structure of AI is an early, extremely complex and important topic. Emad and other colleagues in the pursuit of decentralized AI, their thinking, their attempts, while they may be far from the final answer, deserve enough respect and attention. These explorations are difficult but courageous attempts to forge ahead. These efforts, regardless of the outcome, will become a chapter in the epic of history.

Maybe one day, the world will be grateful to the Emads.