AI Is Stalling. What's Next? | Dr Vaishak Belle

AI is stalling. What’s next?

In our latest episode, Dr Vaishak Belle, Director of Research & Innovation at the Bayes Centre, argues that machine learning is hitting its limits. It's a view American Psychologist and expert AI voice Gary Marcus strongly echoes.

It might look like AI reasons and plans like we do, but it’s really just retrieval. It can spot patterns, sure, but it doesn’t grasp the why behind them: it deals in correlation, not causation.

In biomedicine, deep learning can reveal links like “disease X is more prevalent in this population,” that would otherwise take hours to find manually. But it still can’t reason about why that link exists.

Even in drug discovery, generative models remix what they’ve already seen, not invent new targets or the precise chemical combinations for an optimal treatment.

So where do we go from here?

For truly intelligent systems, Dr Belle argues that we need Neuro-symbolic AI, the "sweet spot" between classical machine learning and knowledge-based reasoning. It's an approach that could help machines move beyond pattern-matching to real understanding.

‍