NEAR Protocol Co-Founder Illia Polosukhin Delivers Keynote at 2024 Hong Kong Web3 Festival on Why AI needs Web3

Hong Kong, 2024 — At the recent Hong Kong Web3 Festival, Illia Polosukhin, co-founder of NEAR Protocol, delivered a keynote on the role of Web3 in the evolution and democratization of AI. Drawing from his extensive background in AI and contributions to Google’s TensorFlow and the pioneering Transformers model, Illia shared insights into the origins of NEAR Protocol and its intrinsic connection to AI.

Illia recounted his journey from Google Research, where he specialized in natural language understanding, to the inception of NEAR Protocol. The protocol was born out of a necessity to find a scalable, low-cost, and accessible payment solution for a global workforce involved in data tagging for AI training, many of whom lacked traditional banking services. This challenge led to the exploration of blockchain technology as a viable solution, culminating in the development of the NEAR Protocol.

Highlighting the evolution of language models since the 1950s, Illia emphasized the nature of word embeddings introduced in 2013. This innovation, he noted, significantly enhanced the capability of deep learning models by enabling the mathematical manipulation of textual data, thereby laying the groundwork for contemporary AI advancements.

The keynote further delved into the limitations of Recurrent Neural Networks (RNNs) and the subsequent development of the Transformers model to overcome these hurdles by leveraging computational parallelism. This approach, according to Illia, was instrumental in scaling AI models, leading to the creation of transformative tools like ChatGPT and Gemini.

Illia also touched upon the specialized infrastructure required for training sophisticated AI models, highlighting the immense computational power and connectivity of modern supercomputers. He raised concerns about the rise of closed AI models and the potential for manipulation and bias, underscoring the importance of transparency and decentralization in AI development.

Illia emphasized the importance of developing a comprehensive suite of AI tools across data, infrastructure, and application layers within the Web3 framework. He stressed that as language models gain the ability to interact directly with society, they hold the potential to manipulate information on a broad scale. However, Illia pointed out that the issue isn’t with AI itself — similar challenges have existed long before AI’s advent. The key, he argued, lies in leveraging cryptographic techniques and blockchain-based reputation systems to ensure authenticity and accountability, making the source and community consensus around information paramount.

Further elaborating on the diversity within AI applications, Illia differentiated between tools and autonomous agents, which can be either centralized, like ChatGPT, or decentralized, like open-source Llama models. He advocated for a future where decentralized models could run on user devices, ensuring full autonomy and alignment with user expectations without necessarily relying on blockchain technology. This, he suggested, could lead to a new era of AI governance, where models make crucial decisions based on verifiable criteria.

Illia also explored various specialization types within AI, from zero-shot learning and fine-tuning to retrieval augmentation, emphasizing that outputs need not be limited to text but could also include rich UI components and direct actions on the blockchain. He discussed the potential for models to be tools, autonomous planners, continuous jobs, or even engage in reinforcement learning optimization, pushing the boundaries of AI’s capabilities.

Addressing infrastructure, Illia highlighted the continuum from centralized systems, like those used by OpenAI, to decentralized inference mechanisms, pointing to an intriguing use case: the transition from programmable money to intelligent assets. These assets, defined and potentially acted upon through natural language, could interact with the real world and other digital entities in novel ways. However, he cautioned that current language models are susceptible to adversarial influences, underscoring the need for robustness in AI development.

As Illia concluded his keynote, he presented a stark choice between a closed AI world, prone to manipulation and control by a few large entities, and an open, decentralized future powered by Web3 and open-source principles. In this envisioned future, crypto-economic incentives could foster the creation of competitive open-source AI models, ensuring a more equitable and transparent digital landscape.

NEAR Protocol, as part of its ongoing mission, is dedicated to realizing this vision within the broader ecosystem. “AI is NEAR,” Illia remarked, inviting the community to engage with NEAR’s upcoming initiatives.

For more information and updates, follow Illia on Twitter and NEAR social platforms.