DeepMind's Kohli: Embracing AI for Humanity's Benefit, Not Just Bigger Models

Image Source: Internet

At the Hindustan Times Leadership Summit 2025, Pushmeet Kohli, Vice President of Science and Strategic Initiatives at Google DeepMind, emphasized the importance of building artificial intelligence responsibly to benefit humanity. Highlighting the organization's 15-year mission, Kohli stressed that AI can push the boundaries of human knowledge when approached as a scientific problem. DeepMind's focus on science-driven AI has led to breakthroughs like AlphaFold, an AI program that predicts protein structure. This achievement has far-reaching implications, such as engineering more resilient crops and understanding heat diseases. Kohli noted that the conventional approach to AI focuses on large language models, but DeepMind's narrow approach tailors AI to specific domains. Kohli emphasized that the organization's core focus is on developing the most powerful and competent AI models. He believes that less data and less supervision will be key to success, and that fundamental breakthroughs will drive AI efficiency. The Gemini family of models, developed by Google, has shown competence in various tasks, including answering complex questions. Regarding the trustworthiness of AI outputs, Kohli acknowledged that AI still makes mistakes, but highlighted the importance of identifying and addressing these errors. He also emphasized the need to build tools that can detect instances of AI hallucinations and warn users. Looking ahead, Kohli predicts a significant emphasis on structural biology, driven by the democratisation of AI technology. He believes that acceleration of science and the development of agentic systems will be key themes for the next year, with a focus on healthcare, drug discovery, and leveraging AI in countries like India. With 180,000 researchers and students already using AlphaFold in India, Kohli sees immense potential for AI-driven research and innovation in the country.