AI Is Like a Brilliant Child. But It Is Missing One Thing

Wise AI? Waterloo Scientists Seek Answers

Artificial intelligence is becoming more capable by the day. But wisdom—the rare, deeply human kind acquired over years—remains beyond its reach. Many researchers are now asking whether truly wise AI is possible. A team from the University of Waterloo has set out to offer a concrete answer.

A new study published in Trends in Cognitive Sciences shows how this gap might be bridged. Titled Imagining and Building Wise Machines: The Centrality of AI Metacognition, it is the first paper to propose realistic ways of building wisdom into artificial intelligence.

An International Alliance of Scientists

Behind the publication stands a broad international team led by researchers from the University of Waterloo. The project brought together experts from the Université de Montréal, the Max Planck Institutes, the Santa Fe Institute, Stanford University, the University of Warwick, and Google DeepMind. Psychologists, computer scientists, and engineers all at one table—a rare combination, and a notable one in itself.

What Exactly Is AI Wisdom?

One of the lead authors, psychologist Sam Johnson from Waterloo, explains in a university press release why the idea of wise AI makes sense at all:

Artificial intelligence is getting smarter every day, but it lacks one crucial human skill: wisdom. Wisdom is not just knowledge or intelligence. It is a set of mental tools needed to handle life’s challenges: making difficult decisions, navigating unpredictable social situations.

Where Does AI Fail?

Today’s AI systems are masters of well-defined tasks. They can beat a champion at Go, write an essay, or translate text into dozens of languages. But when the rules stop being clear, when a problem becomes complex, and when a situation turns uncertain, their abilities drop sharply.

The reason, as the researchers explain, is simple: machines lack the full set of strategies humans use to cope with uncertainty. Their idea is to teach AI to think about its own thinking—metacognition. The goal is for machines to recognize the limits of their own knowledge, adapt to changing contexts, weigh multiple perspectives, and remain flexible about how situations may unfold.

How to Teach Machines Wisdom?

Igor Grossmann, the second lead author, also from Waterloo, adds in the same release:

Wisdom has previously seemed too philosophical and human-centered to formalize for machines. But by breaking it down into specific strategies—intellectual humility, perspective-seeking, adaptation to context—we can create a concrete blueprint for building AI that doesn’t just calculate but reasons wisely.

The researchers propose specific solutions: new ways of training large language models, system architectures that could support wise reasoning, and even metrics for measuring such wisdom.

Why Does AI Need Wisdom?

According to the University of Waterloo press release, wise AI systems could:

• handle new problems and unfamiliar environments more effectively,
• collaborate more successfully with humans toward shared goals,
• be more understandable to users,
• and above all, be safer, because they would better understand what humans expect of them.

Prof. Johnson offers a vivid comparison:

If the smartest person in the world were a 2-year-old, we wouldn’t entrust them with nuclear weapon codes. AI increasingly resembles a child prodigy who still needs a healthy dose of wisdom from its human parents.

What’s Next?

The next step, the authors say, will be working with industry to develop computational models of human wisdom. These models would then guide the design of future AI systems.

Will it work? It is hard to say. But the very fact that scientists are now seriously asking how to build wise AI shows that the field is beginning to treat the problem with the weight it deserves. The final answer, however, still lies ahead.

Read this article in Polish: AI jest jak genialne dziecko. Ale jednej rzeczy jej brakuje