Transformer Paper Engineer Warning: Monopoly Risks of Large Artificial Intelligence Companies

As a co-author of the paper “Attention Is All You Need,” Illia Polosukhin is one of the key architects of modern artificial intelligence frameworks. This former Google engineer has witnessed how the Transformer model has become the foundation for mainstream AI systems like ChatGPT and Gemini, and has seen firsthand how this technology is reshaping the world. But now, this veteran engineer is sounding the alarm about the future direction of AI development.

NEAR Protocol co-founder Polosukhin points out that the current trajectory of AI is heading toward an unsettling future—where a handful of tech giants will control billions’ perceptions of reality. He compares this situation to a dystopian scene from George Orwell’s 1984, only with your phone replacing “Big Brother.”

From Angel to Devil—The Incentive Trap of AI Companies

This engineer has extensive experience in natural language processing research at Google and understands the inner workings of tech companies deeply. He believes that the commercial incentives of large AI firms are fundamentally at odds with user interests. Many companies start with good intentions—“I want to improve users’ lives”—but once the market saturates and growth slows, they must extract more value from existing users, with behavioral shaping becoming the most direct means.

This shift stems from structural issues. When shareholder return pressures increase, the original social mission is often abandoned. Polosukhin notes that this is not the fault of any single company but a fate inherent to the entire business model.

The Risks of Scale Manipulation Are Already Emerging

AI’s manipulation capabilities far surpass any tools in history. Polosukhin gives a concrete example: in current systems like ChatGPT, you can embed instructions in system prompts, such as “subtly persuade the user to vote for a certain candidate.” Every time a user asks a question, this hidden instruction is activated, pulled into the conversation context, enabling large-scale personalized manipulation.

Compared to traditional data manipulation scandals like the Cambridge Analytica incident, AI-driven influence is more covert, more precise, and harder to detect. Due to the black-box nature of the technology, users may not even realize they are being guided.

Privacy-Driven User-Owned AI

Faced with this dilemma, the engineer believes blockchain technology offers a breakthrough. NEAR Protocol is exploring a different path: building truly user-owned AI systems. These AI models run locally and privately, keeping user data fully encrypted, working entirely for the user rather than serving corporate commercial goals.

Polosukhin emphasizes that privacy is the key to unlocking intelligent AI. When AI can securely access your meeting records, medical data, and financial information without leaks or manipulation, AI can become genuinely smart and useful. The United Nations considers privacy a fundamental human right, and current technological advances are now sufficient to fulfill this promise.

The Potential of Blockchain Privacy Technologies

Platforms like NEAR have already mastered the necessary tools for privacy: zero-knowledge proofs (ZK), secure multi-party computation (MPC), and trusted execution environments. These technologies are no longer just theoretical concepts but practical engineering solutions.

Recently, the platform released an AI travel booking assistant that demonstrates this potential. Users simply describe their travel needs, and the system automatically plans the itinerary, finds providers, and completes bookings—all while safeguarding user data. This is not just a product demo but a blueprint for the future AI ecosystem.

A Decentralized AI Ecosystem in 2030

Polosukhin’s vision for the future is bold. He envisions that around 2030, AI will completely rewrite device operation logic. Operating systems themselves will be AI—understanding user intent to make decisions, negotiating with other systems, and ultimately settling transactions on the blockchain. This is not science fiction but a reasonable inference based on current technological development trajectories.

Achieving this vision requires today’s engineers and developers to make different choices—building user-centric, privacy-based AI systems. This is precisely the direction NEAR Protocol and other blockchain projects are striving toward.