From the question “What does artificial intelligence know?” to the deeper question: “How does it think? And how does it solve problems?”

Can we give artificial intelligence a test similar to the human IQ test?

At first glance the question seems simple, but it opens one of the most exciting files in the future of artificial intelligence: how do we measure machine intelligence?

A person is not described as intelligent only because they memorise a great deal of information, but because they can understand a problem, reason, discover patterns, plan, use knowledge and make decisions.

As artificial intelligence moves from systems that merely answer questions to systems that can search, plan, use tools and carry out multi-step tasks, it has become necessary to think about new ways of measuring its abilities.

Here Dr. Doaa Mohey Eldin, faculty member at the Faculty of Computers and Information and AI and Data Science consultant, explains the importance of tests such as GAIA, which represent a different direction in evaluating general-purpose AI systems.

GAIA: When the Question Is Not “Do You Know?” but “Can You Get It Done?”

Traditional AI tests may measure a model’s ability to answer questions in mathematics, language or general knowledge.

GAIA, however, looks at the matter from a more complex angle.

The core idea is to put the system in front of tasks that require several abilities at once, such as understanding the question, reasoning, searching, using tools, handling data or files, and carrying out several steps to reach the correct result.

In this way, evaluation moves from:

“How much information does the system know?”

to:

“What can the system do with this knowledge?”

And this is where the real comparison with human intelligence begins.

Is GAIA an IQ for Artificial Intelligence?

The precise scientific answer is: no.

GAIA is not an IQ test in the familiar psychological sense, and its result cannot simply be turned into a number such as 120 or 140.

But GAIA’s philosophy comes close to an important question in human intelligence testing: measuring the ability to deal with problems, not just to memorise information.

An intelligent person does not rely on knowledge alone; they know how to search, how to analyse, how to connect information and how to make a plan to reach a solution.

These same abilities have now become part of the evaluation of modern intelligent systems.

From “a Single Intelligence” to “a Cognitive Fingerprint”

Dr. Doaa Mohey Eldin believes that the future of measuring artificial intelligence may lie not in searching for a single number like the human IQ, but in building a Cognitive Fingerprint that shows a system’s strengths and weaknesses.

One model may be very strong at reasoning but less capable at planning.

Another may be excellent at using tools but need improvement in verifying its results.

A third may excel at understanding images and multimodal data.

Reducing all these abilities to a single number may therefore hide important information.

The more expressive alternative is to build a multi-dimensional cognitive profile.

What Could This Fingerprint Look Like?

Artificial intelligence can be pictured as a system made up of a set of abilities, such as:

Reasoning – problem solving – planning – tool use – knowledge integration – working with data – verifying results – multimodal understanding.

The system’s performance is then measured on each axis separately.

This gives us a more accurate picture of its abilities.

Instead of saying:

“This system has an IQ of 150”

we can say:

“This system has a high level of reasoning and tool use, and a medium level of planning and verification.”

This is far more useful to the researcher, the developer and the decision-maker.

Why Has This Idea Become Important Now?

Because artificial intelligence is no longer just a chatbot.

We are gradually entering the age of Agentic AI, in which a system can understand a goal, make a plan, use tools, carry out a chain of actions and then review the results.

Here the old way of measuring is no longer enough.

If a system can carry out a task that takes a person hours, the question is no longer only about its ability to give the right answer, but about:

How did it plan?

Which tools did it use?

Did it choose the right strategy?

Did it discover its own mistakes?

Was it able to correct its course?

And did it reach the result in a reliable way?

These questions come much closer to the concept of practical and cognitive intelligence.

Human and Machine: Can They Be Compared?

Comparing humans and machines is not easy.

Humans have aspects that AI tests cannot easily measure, such as life experience, social awareness, emotion, moral judgement and the ability to understand human context.

On the other hand, artificial intelligence can process huge amounts of information faster than humans in many tasks.

The aim, therefore, is not to prove that the machine is “smarter than the human” or the other way round.

The aim is to understand:

What kind of intelligence does the machine have?

Towards a “Cognitive Intelligence Index” for Artificial Intelligence

From here a new research direction can be developed, based on building a cognitive index for intelligent systems that combines the results of a set of tests and benchmarks, instead of relying on a single test.

The index could include dimensions such as:

Reasoning.

Problem Solving.

Planning.

Tool Use.

Knowledge Integration.

Execution.

Verification.

Multimodal Understanding.

We would then have something like a cognitive profile of artificial intelligence that can be used to compare models, rather than just a single number.

From the Laboratory to Decision-Making

The importance of this direction does not stop at scientific research.

In universities, such indices can be used to evaluate educational AI tools.

In organisations, they can be used to choose the right systems for specific tasks.

In scientific research, they can help compare the abilities of models more objectively.

As for the decision-maker, they need to know not only “which model scored highest?”, but:

Highest at what?

Better at which kind of task?

Where can it be trusted?

And where does it need human oversight?

Here, measuring artificial intelligence turns from a technical issue into a managerial and strategic one.

Intelligence Is Not Measured by a Number Alone

The importance of this vision is that it reframes the question from the ground up.

Perhaps the future is not in saying that artificial intelligence has an IQ like a human.

Perhaps the more precise question is:

What is the map of this system’s cognitive abilities?

What can it do better?

Where does it make mistakes?

When does it need a human?

Can it learn from its mistakes?

And can its decisions be trusted?

These questions may be more important than any single number.

Artificial intelligence does not necessarily need a “human IQ”… it needs a measure that describes its intelligence as it really is.

This is the importance of moving from an intelligence test to an intelligence fingerprint, and from the question “What score did the system get?” to a deeper one:

“What can this system think about, plan, carry out, learn from and justify?”

And perhaps this is the path towards the next generation of AI evaluation standards: a generation that does not measure the machine only by how much it resembles a human, but first tries to understand the nature of its own intelligence.