Castellana said he envisions developers creating applications that make autonomous and data-driven decisions without human intervention, potentially enabling AI-powered oracles without relying on centralized intermediaries. This ability, he argues, will revolutionize everything from insurance contracts to supply chain management.

While Castellana, a serial entrepreneur and crypto industry veteran, acknowledges centralized AI’s advantages, he highlights the lack of transparency which he said undermines trust in their outputs. Decentralized AI, in contrast, allows multiple models and validators to participate therefore reducing bias and manipulation risks.

Castellana also explores the future of AI and blockchain, emphasizing the importance of staying informed about these evolving technologies. Below are the Yeager.ai co-founder’s answers to all the questions sent.

Albert Castellana (AC): The separation between centralized and decentralized AI is defined by three main aspects: who trains the models, who runs the inference, and the nature of the models themselves. When thinking about decentralizing AI: training models is a very complex and expensive task, which makes it difficult for small teams to do, but some companies like Meta have decided to open source them; running the models is really hard in a distributed manner, so tons of work is being put into proof of inference systems which allow to verify that the execution was done correctly.

However, in order to create decentralized applications, it’s not just the compute that needs to be decentralized, it’s also the decision-making. Every model is centralized in itself; with its own biases, features and understanding of the world. Opaque and difficult to audit. So, when making decisions, relying on a single model, means having a centralized model.

At Genlayer, we’re aiming to decentralize decision making by involving multiple models, and leveraging blockchain to reach a consensus that allows them to agree on the result of subjective tasks. Think of it as moving from a system with one judge to one with multiple judges or even a jury, where various perspectives come together to produce a more accurate and fair outcome. This democratizes AI and mitigates the risks of a single point of failure or bias.

AC: Over the years, we’ve seen how most users chose utility over privacy, often without a clear understanding of what they are giving up. Many have accepted the trade-off of surrendering personal data in exchange for the convenience of using services that simplify their everyday life.

With AI, this issue will first get much worse, and then improve. The better AI becomes, the more information we will want to share with it. However, the better-decentralized technology becomes, in part thanks to AI, the more we can regain control over our data.

If we want to speed up the process, we need to educate the public, foster open-source tools and push for decentralized AI.

AC: AI brings a level of autonomy and intelligence that can help blockchain applications (dApps) evolve beyond simple, static contracts. The ability to integrate AI into blockchain allows for much more subjective decision-making, which opens the door to entirely new use cases.

For developers, this means the ability to create applications that make autonomous, data-driven decisions in real time without human intervention. For example, prediction markets could use AI to continuously analyze external data and settle outcomes more accurately. Similarly, decentralized finance (DeFi) applications could benefit from dynamic decision-making, like adjusting interest rates or liquidity pools based on real-world data.

AI also enables intelligent oracles—systems that connect blockchain contracts to data sources outside the blockchain—without the need for a centralized intermediary. This could revolutionize everything from insurance contracts to supply chain management by enabling more accurate and autonomous decision-making.

AC: With GenLayer, we’re focused on expanding what’s possible with decentralized applications (dApps) by integrating AI technology directly into the blockchain, at the consensus level. We’re introducing the concept of Intelligent Contracts, which are a step beyond traditional smart contracts. Traditional smart contracts are static—they can only do what’s been explicitly coded into them, and they require deterministic inputs. They’re also limited in their ability to process subjective tasks like natural language or image recognition.

GenLayer changes that by creating a dynamic consensus mechanism where validators—each connected to a different Large Language Model (LLM)—collaborate to process non-deterministic inputs like data from the web, natural language, or even multimedia. This allows us to build dApps that can truly interact with the outside world.

The potential use cases are broad: Decentralized Autonomous Organizations (DAOs) will operate independently, pulling in real-world data without relying on centralized oracles. Parametric insurance will become cost effective, as the resolution will become fast and cheap. Prediction markets will function in real-time, without human oversight. Performance based contracts will be escrowed, evaluated and paid in a completely automated manner. Fees, liquidation levels, and even emergency protocols will be managed autonomously based on external inputs, making decentralized finance (DeFi) much more robust and adaptable.

AC: While decentralized AI brings many benefits, we must acknowledge that there are still strengths in centralized AI systems, particularly in terms of performance and intelligence. Closed-source models, with the backing of massive corporate resources, are still generally more advanced and faster than open-source alternatives. However, centralized AI lacks transparency, making it difficult to fully trust the outputs.

Decentralized AI, on the other hand, offers greater transparency, security, and diversity of opinion. By allowing multiple models and validators to participate in decision-making, decentralized AI reduces the risk of bias or manipulation. But it can also be slower and less efficient compared to centralized systems. Performance is a trade-off for greater security and reliability in this case.

At GenLayer, we welcome this diversity. Validators in our network can run different AI models—open or closed-source. By combining the perspectives of different models, we create a system that is more secure and reliable. This is important for building trust in decentralized systems, where accuracy and fairness are paramount.

AC: My advice to developers, and really anyone, is twofold: stay adaptable and embrace continuous learning. The pace at which AI and blockchain technologies are evolving is insane. What’s possible today might be outdated within a year, so it’s crucial to keep learning and stay on top of technological advancements.

Also, the role of a developer is shifting. Coding will soon be more about orchestrating systems than writing lines of code. You’ll be managing different AI models, decentralized systems, and interfaces. Think of it like conducting an orchestra—each part of the system needs to work together harmoniously, and your job will be to ensure that happens.

Finally, don’t hesitate to experiment with emerging technologies. The beauty of working in such a fast-evolving space is that there are many opportunities to innovate. Some of these opportunities will come and go quickly, so be ready to seize them when they appear.