Who Really Controls Your AI? Illia Polosukhin on Privacy, Ownership, and the Future of Intelligence In this episode of When Shift Happens, I sit down with @ilblackdragon, co-founder of @NEARProtocol and CEO of the @NEARFoundation, to discuss his journey from helping develop the transformer architecture behind ChatGPT and Claude to building a blockchain ecosystem focused on user-owned AI. We explore what happens to our personal information when we use AI, why concentrating intelligence in the hands of a few companies could be dangerous, and how cryptography could give individuals greater control over their data and digital lives. From Growing Up in Ukraine to Helping Build Modern AI Growing up in Ukraine after the collapse of the Soviet Union, Illia experienced firsthand what happens when financial institutions fail. His grandparents' savings became practically worthless, hyperinflation made everyday necessities increasingly expensive, and his family had to grow much of its own food. Technology offered him a different path. He started programming at age ten, eventually pursuing competitive programming and joining Google as an AI researcher. At Google, Illia contributed to the groundbreaking 2017 paper Attention Is All You Need, which introduced the transformer architecture that would become foundational to modern large language models. Before transformers, many language models processed text sequentially, analyzing one word after another. Transformers introduced an approach that could process words in parallel, using a mechanism called attention to understand relationships between them. This made it possible to train increasingly capable language models much more efficiently. For Illia, the achievement was part of a much older ambition. At fourteen, he had already imagined computers that could understand human language and program themselves. Yet despite Google's resources, he found that developing disruptive products within a large organization was difficult. He eventually left to build NEAR AI, taking a significant risk to pursue his ideas independently. How a Payment Problem Led Him Into Crypto Illia’s AI startup relied on students across Eastern Europe, Southeast Asia, and China to complete small data-labeling tasks, which sometimes earned as little as fifteen cents per assignment, making traditional international payments impractical. Unfortunately, existing blockchain networks weren't much better, with transaction fees that could exceed the amount being paid. Illia saw an opportunity to build financial infrastructure that made moving money globally faster, cheaper, and more accessible. But as he explored blockchain technology, he began to recognize its potential to protect individuals from the institutional failures he had witnessed growing up. This inspired NEAR's broader mission of enabling people to control their data, assets, and freedom of choice. However, building NEAR wasn't straightforward. The team initially expected to develop its blockchain in six to eight months, but the process took approximately two years. They also navigated difficult fundraising conditions and multiple crypto market cycles. What sustained the project was a belief that the infrastructure solved a genuine problem, regardless of market sentiment. What Really Happens to Your Data When You Use AI? Today, Illia sees another form of institutional dependence emerging through artificial intelligence. Every time people interact with an AI assistant, they share information about their businesses, finances, health, relationships, and decisions. This is even worse for companies using AI agents that access internal documents, emails, proprietary workflows, and strategic discussions. Illia argues that when this information is processed through centralized AI providers, users must trust those companies' policies and security practices. Even when providers offer settings that restrict model training, he questions whether users have sufficient visibility into how their information is processed, retained, or otherwise used. There is also the risk of unauthorized access and legal disclosure, particularly for sensitive financial, medical, legal, and proprietary business information. But his concern extends beyond privacy. As AI becomes the primary interface through which people access information, the companies operating these systems could gain enormous influence over what users see, believe, and decide. The deeper problem, Illia argues, is the lack of independent verification over how these systems operate. Building AI That Works for, not Against, You Illia's proposed alternative is user-owned AI. “AI should be on your side,” he explains. His argument is that conventional AI companies ultimately answer to their business interests and incentives that may not align with what benefits individual users. User-owned AI would instead prioritize privacy, individual control, and alignment with the user's interests. To make this possible, NEAR is combining blockchain infrastructure with privacy-preserving AI technology. One component is encrypted AI processing, designed to prevent providers from accessing users' prompts and responses in readable form. Another is cryptographic verification, which can provide evidence about the model, software, and computing environment used to process a request. Illia compares the intended user experience to the security indicator in a web browser. People shouldn't need to understand the underlying cryptography to know whether an application provides verified privacy protections. Blockchain also provides a potential economic foundation for this system, allowing participants to hold tokens and access services rather than relying entirely on a single company. Still, it’s concerning that more private AI could also make it harder to monitor harmful uses. Illia acknowledges safety as one of the strongest arguments against his approach, but believes privacy and security can coexist through better technical systems. The Next Shift Could Be Bigger Than AI Looking ahead, Illia is particularly excited about personalized medicine. He imagines a future where AI can analyze an individual's medical information and help develop treatments tailored to their specific needs, including therapies that might otherwise be commercially unattractive to large pharmaceutical companies. He also expects AI to automate more activities where people primarily care about the outcome, while human experiences such as sports, music, and art retain their value because people care about the process itself. Ultimately, Illia's believes that technology should expand what individuals can do without requiring them to surrender control over their lives. As AI becomes more powerful and deeply integrated into everyday decisions, the bigger question won’t be how intelligent these systems can become, but who they ultimately serve. 👉If you enjoyed reading the summary, head over to When Shift Happens on YouTube or your favorite podcast platform to access the full convo. Retweeted by @NEARProtocol from @KevinWSHPod