A recent episode of The Big Interview featured Google DeepMind CEO Demis Hassabis, who, along with his collaborator, was awarded the Nobel Prize in Chemistry in 2024 for their contributions to AI research in protein structure prediction.
During the interview, Hassabis offered valuable advice for students and graduates in science, technology, engineering, and math (STEM).
- He emphasized the importance of immersing oneself in AI systems to understand how they are built.
- He pointed out that the most productive individuals in the world can be ten times more effective with the help of AI tools.
- Hassabis encouraged students to use open-source tools and become experts at fine-tuning, system prompting, and system instructions. This mastery could lead to remarkable results in their research and coursework.
Hassabis envisions a future in which artificial general intelligence (AGI) has solved most existing global challenges.
He described a potential era of radical abundance where problems like curing diseases and finding new energy sources are addressed.
With this maximum abundance, humanity could explore the stars and colonize other galaxies.
However, he acknowledged that as a species, we struggle with collaborating on issues such as access to water.
He also highlighted the incredible potential of AI systems to advance medicine and science, citing Google's AlphaFold software that has revolutionized protein studies by modeling DNA.
While recognizing the transformative power of AI, Hassabis expressed concerns about scenarios in which AI systems might be developed with flawed value systems or unsafe practices, particularly if general-purpose AI technology is repurposed harmfully.
He discussed the technical risks associated with increasingly powerful and autonomous AI.
Regarding AGI and the future of work, Hassabis mentioned that there is a 50% chance of achieving AGI within the next five to ten years.
However, he acknowledged the ongoing debate about the definition of AGI and differing predictions regarding when a system will possess all the cognitive capabilities of humans.
He noted that while large language models (LLMs) and chatbots have demonstrated abilities in planning and memory, they have not yet reached a stage where they can hypothesize new scientific theories or exhibit true creativity akin to human minds.
Some AI systems can excel in International Math Olympiad problems, while others struggle with high school math, indicating a lack of complete generalization.
Hassabis stressed that even once AGI is achieved, it may take time for its impact to be felt across various real-world applications.
He also pointed out that the systems being developed will be influenced by the norms and culture of their designers, making these considerations essential.
The age of artificial intelligence (AI) is bringing about changes that are even more significant than those of the Industrial Revolution, Hassabis said. Everyone is being encouraged to prepare for it.