Demis Hassabis quantifies the scale of AGI's impact with a powerful analogy: it will be ten times as transformative as the industrial revolution but will unfold over a decade instead of a century. This framing underscores the unprecedented speed and magnitude of the societal upheaval and advances he anticipates.
Demis Hassabis provides a concrete and near-term forecast for Artificial General Intelligence (AGI), stating there is a 'very good chance' of it arriving within the next five years. This timeline is consistent with predictions he and his co-founders made when starting DeepMind in 2010.
The gap between the top few AI labs and the rest is growing, not shrinking. Demis Hassabis explains this is because these labs leverage their own superior tools for coding and math to accelerate development of the next generation of models, creating a powerful compounding advantage that makes it harder for others to catch up.
For AI safety, Demis Hassabis advocates for an international regulatory body, similar to the International Atomic Energy Agency. This body would have technical experts who audit frontier models against agreed-upon benchmarks, checking for undesirable properties like deception and ensuring public confidence through independent verification.
Demis Hassabis identifies critical capabilities missing from today's AI systems. The biggest hurdles are continual learning (the ability for a trained model to learn new things without retraining) and hierarchical, long-term planning. This suggests that simply scaling current architectures may not be enough to achieve AGI.
Contrary to the narrative that model performance is plateauing, Demis Hassabis states that while returns from scaling are no longer exponential, they remain 'very substantial.' Frontier labs continue to see significant gains from increasing model size and compute, suggesting the current AI paradigm is not yet exhausted.
Demis Hassabis argues that building DeepMind in London provided a key advantage. Being slightly removed from the Silicon Valley 'maelstrom' and its latest trends is 'very conducive to thinking deeply about things' and being more original, which is critical for long-term, ambitious deep tech projects.
Demis Hassabis presents a paradox: while AI is experiencing peak short-term hype, its revolutionary potential over a ten-year horizon is still vastly underestimated. This suggests that even the most bullish observers may not fully grasp the magnitude of the changes AI will bring to the economy and society.
The long-term strategy for AI in drug discovery is a two-step process. First, create an AI platform to design effective drugs. Second, after a dozen or so AI-designed drugs succeed, use that data to convince regulators to trust AI predictions, potentially allowing future drugs to skip steps like animal testing and accelerate trials.