“True AI should understand the relationship between cause and effect,” says Judea Pearl, a pioneer in artificial intelligence.

Let’s teach machines to understand the question “Why?”. While in the 1980s we created programmes that primarily calculated probabilities (for example, Bayesian networks), this is no longer sufficient. Current successes in AI are largely based on neural networks. These do the same thing as the previous generation of neural networks, but with truly vast amounts of data. “All the impressive successes of deep learning amount to mere curves,” he recently remarked.

Judea Pearl describes his vision as follows: “The new AI should be capable of causal reasoning. For instance, instead of merely being able to correlate fever with malaria, machines need the ability to reason. That is, that fever is caused by malaria.” According to Pearl’s expectations, machines could provide human-level intelligence if they can understand causation. They could communicate with humans more effectively and even act as moral beings with free will – even for evil.

Source: https://www.quantamagazine.org/to-build-truly-intelligent…/…