最新报道：According to Crypto.news, Karia Samaroo, founder and CEO of xTAO, the only publicly listed company dedicated to the Bittensor ecosystem, explained why AI needs to be decentralized. Artificial intelligence has captured the public's imagination at an unprecedented rate. However, beneath its potential lie serious concerns about the concentration of power and control. Currently, the most popular AI models are the exclusive property of a few large tech companies, which fully control their design and use. Crypto.news interviewed Karia Samaroo, founder and CEO of xTAO, a publicly listed company dedicated to the Bittensor (TAO) decentralized AI ecosystem. Samaroo explained why an alternative model is needed to make AI more open, decentralized, and aligned with user needs. crypto.news: What does blockchain bring to AI, and what role does Bittensor play? Karia Samaroo: Centralization is AI's biggest problem. As AI develops into the most powerful tool humanity has ever created, having it controlled by only a few companies creates enormous concentration risks. I often compare Bittensor to Bitcoin. Bitcoin solves the centralization problem surrounding currency: it's inflated, anyone can access it, and there are no gatekeepers. Bittensor applies the same idea to AI. With centralized AI, like OpenAI, a single institution decides how models are trained, what data is used, what biases exist, and what is censored. They can also cut off access at any time. This is a major problem. Bittensor uses the Bitcoin model to solve this problem in AI. CN: How can companies introduce decentralization into AI? KS: There are some great examples of decentralized AI solutions. Grass incentivizes data collection, although it focuses on one part of the AI stack. Render is a decentralized computing network, which is also very important. Bittensor has a broader scope. I would call it a "global network for AI." It doesn't just focus on one area, like data or compute. It has multiple subnetworks, each addressing a different problem in the AI stack, and they're all interconnected. CN: Why would companies build on Bittensor instead of using more established models like OpenAI? KS: I think there are several reasons. One is philosophical. Many people building on Bittensor see the value in contributing to the decentralized web and the mission of decentralized AI. This definitely has many attractions. Another is technical. In a decentralized network, scalability offers advantages. For example, Bitcoin, through its incentive structure, has created the world's largest computer. It's so widely distributed that it can never be shut down, as it has so many nodes running in different locations, with different networks and power sources. Then there's the concept of open innovation. Anyone can experiment, iterate, and monetize their models without gatekeepers. If you're an AI engineer, typically you have to apply for a job, interview, get hired, and then ultimately work on a very specific task within that company. On Bittensor, you can choose a subnet you want to mine on, build your model, compete with others, and get paid instantly. CN: AI models run by large tech companies benefit from vast amounts of data, like Grok owning Twitter. How can decentralized AI compete? KS: I think Grass is a great example, and there are similar projects on Bittensor. The idea is to crowdsource data and incentivize people to collect and manage it. The network has grown remarkably. This is how decentralized networks can introduce datasets of equal or even higher quality. Large tech companies control the most abundant data today, but with the right incentives, decentralized systems can compete. Another big problem is that when Meta or Twitter owns your data, you get nothing in return. As a contributor, you don't receive rewards. Decentralized networks flip this on its head. They align incentives with creators and contributors, which is the way it should be. If you take a photo, you should be credited. If you publish an article, you should benefit from it. CN: How does decentralized AI address the security and societal impact of its models? KS: There are several aspects to security. One is the training data. If it's biased, toxic, or contains sensitive information, that's a problem, and this applies to both centralized and decentralized systems. It's an issue people grapple with every day. Another is the model's output. How do you prevent harmful outputs? In Bittensor, this is handled by validators. They're responsible for detecting harmful or low-quality outputs, and the better they do, the more rewards they receive. It's built into the network design. The foundation also has some monitoring policies, but the goal is to phase them out. Over time, security and governance really become the job of the validator. CN: Are you worried about these models being scrutinized in the future, either from governments or in response to biased outputs? KS: That's a good question. I would compare it to centralized or state-owned media, where a single decision-maker can choose what to show or not show. If they're under pressure or just making decisions internally, they can change what the output looks like. That's a big problem. We've seen this with social media. If a meta wants to push a certain narrative, they'll do it. It's not necessarily evil - it's just the way the incentives work.Decentralized AI is more representative of the population. It's not perfect, but if a subnet or product on Bittensor becomes too biased, participants in the network can vote and adjust incentives. This means poor performance gets less rewards. The idea is that if the system reflects the population, people will support products that feel fair and transparent. And it's easier to audit—you can see the incentive structure, you can see the code. You can't do that with closed systems. This is why people are worried about centralized AI.