Most of us think of artificial intelligence as something that can easily out-perform a human brain—but can it?


Believe it or not, scientists are still finding it challenging to create an artificial intelligence (AI) that can perform as well as the human brain. To learn more, researchers at the Imperial College London recently conducted a study (published in the journal Nature) to explore why AI sometimes underperforms. Dr Dan Goodman said the study produced some interesting results:

The most striking difference between humans and AI is that our learning seems to be much more robust.

Human brains learn quickly (even babies can quickly identify images or generalize their learning in one area and apply it to another). 

If you train an AI with one dataset and then test it on slightly different data, it often fails spectacularly—whereas people tend to do much better at adjusting to the new data.

Humans can learn with much less data than is required for AI systems. For instance, humans can recognize a new animal from just one or two pictures, whereas an AI system may need hundreds or thousands of examples.

Researchers asked the question: ‘What would happen if we created a neural network where the cells were all slightly different, just like we see in biological systems?’ To do this, researchers changed the time constant for cells in the neural network—in other words, how quickly the cell reacted to other cells connected to it. The resulting neural network was more robust and able to learn faster. Hopefully, these results should help build better-performing AI systems.


Dec 2020 Forbes Artificial Intelligence (AI): What’s In Store For 2021?

Mar 2021 HBR AI Should Augment Human Intelligence, Not Replace It

Jul 2021 ZDNet What is AI? Here's everything you need to know about artificial intelligence

Jul 2021 ABC Australia Artificial intelligence can now be recognised as an inventor after historic Australian court decision

Oct 2021 The Guardian The truth about artificial intelligence? It isn’t that honest