Demis Hassabis, the CEO of AI powerhouse DeepMind, is setting the bar high for true Artificial General Intelligence (AGI) with a series of stringent criteria that could redefine the industry’s standards.
The ‘Einstein Test’
Hassabis wants AI systems to be able to independently derive groundbreaking scientific theories, starting with Albert Einstein’s general relativity, using only knowledge available before 1911. This challenge, dubbed the ‘Einstein test’, is a far cry from current AI capabilities.
Roger Penrose, a mathematician and physicist, and Nobel laureate, is one of the few humans who could potentially pass the ‘Einstein test’ with ease. However, no current AI system comes close to clearing this bar, and it’s unlikely that any of them will in the near future.
The ‘Einstein test’ is not just a intellectual curiosity; it’s a benchmark for truly advanced AGI. If an AI can derive general relativity from scratch, it would demonstrate a level of cognitive flexibility, creativity, and problem-solving ability that is currently unique to humans.
AGI Redefined
Hassabis’s criteria for true AGI go beyond the ‘Einstein test’ and include several other challenges that current AI systems struggle to meet. These include:
- The ability to learn and reason independently, without relying on large amounts of data or human input.
- The capacity to generalize knowledge across different domains and tasks.
- The ability to understand and manipulate complex systems, such as the human body.
What this means
Hassabis’s stringent criteria for AGI have the potential to delay perceived AGI milestones, as current AI systems will need to make significant strides before they can meet even the first of these challenges. However, this could be a good thing, as it forces the industry to focus on developing truly advanced AI capabilities rather than settling for incremental progress. In the long run, this could lead to AGI systems that are more robust, reliable, and useful to society.
For now, Hassabis’s criteria remain a challenge to the AI community, a reminder that true AGI is still a pipe dream – for now.
