Science & Technology

Scientists launch 'Humanity's Last Exam' to test the intelligence of AI

By T.K. Randall
March 2, 2026

Image: AI-generated (Midjourney) All the way back in 1950, mathematician and computer scientist Alan Turing introduced the Turing test - a method of determining how intelligent a machine is compared to a human.

In practice, the test involved a human being tasked with evaluating two sides of a conversation between a computer and a human participant.

The test could only be passed if the evaluator was unable to tell which side was human and which was a machine.

These days, of course, various AI chatbots are clearly more than capable of having a convincing natural language conversation with humans, thus rendering the Turing test somewhat obsolete.

Keen to come up with a better way of testing the intelligence level of such systems, researchers from the Center for AI Safety and Scale AI have devised a brand new test known as 'Humanity's Last Exam'.

It is comprised of 2,500 questions covering more than 100 subjects with the answers being "unambiguous and easily verifiable but which cannot be quickly answered by internet retrieval."At launch, even the most advanced AI only managed a score of 8.3%, but since then this number has crept up significantly.

Last month, Google's Gemini 3 Deep Think successfully achieved a score of 48.4%.

Human experts, by contrast, were able to score over 90% in their fields of expertise.

It is inevitable, however, that AI will be able to catch up to this in the near future.

Once it does, we may be looking at the world's first artificial general intelligence - a system that can match or even exceed the capabilities of its human creators.

Where things will go from there, however, remains anyone's guess.

Source: Live Science

Tags: Artificial Intelligence