In a groundbreaking study, researchers propose a unified framework for understanding human and artificial intelligence, offering a new path to innovation in AI development.
What exactly is intelligence? For centuries, experts have debated this question, particularly when it comes to human intelligence. But now, with artificial intelligence (AI) becoming an integral part of our daily lives, defining intelligence has become even more urgent—and complicated.
A new study by Australian researchers, Gilles E. Gignac from the University of Western Australia and Eva T. Szodorai from Curtin University, is tackling the issue head-on. Their paper, Defining Intelligence: Bridging the Gap Between Human and Artificial Perspectives, presents a novel way to think about intelligence, one that could reshape how both psychologists and computer scientists approach the concept. By offering a refined and unified definition of intelligence across both disciplines, their work could lead to more meaningful advancements in artificial intelligence technology.
A Century of Confusion: What is Intelligence, Anyway?
For more than a hundred years, scientists have been trying to pin down a clear definition of human intelligence. Gignac and Szodorai note that traditional definitions often describe intelligence as the ability to reason, solve problems, think abstractly, and learn from experience. But these are essentially lists of traits, which do little to capture intelligence as a cohesive concept.
Artificial intelligence, meanwhile, faces its own set of challenges. Since AI doesn’t possess consciousness or biological processes, defining its “intelligence” through human terms doesn’t quite fit. AI is usually described as a machine’s ability to perform tasks that would require human intelligence—things like recognizing patterns, learning, and decision-making. But as the authors point out, this vague definition lacks the precision needed to measure whether machines are truly intelligent or just executing pre-programmed tasks.
A Unified Definition of Intelligence
Gignac and Szodorai’s solution is a new, cross-disciplinary definition that works for both human and artificial intelligence. Their proposal is surprisingly simple: intelligence, they say, is the “maximal capacity to achieve novel goals successfully.” In other words, intelligence is best demonstrated when a person—or machine—can solve new problems they’ve never encountered before.
For humans, this involves perceptual-cognitive processes like attention, memory, and reasoning. For machines, it’s about executing computational algorithms. Both systems can be intelligent, as long as they can tackle unfamiliar challenges. This unified framework opens up new possibilities for how we measure and develop AI, focusing on adaptability and creativity rather than mere expertise.
Bridging Psychology and AI: The Role of “AI Metrics”
One of the key recommendations in the paper is the establishment of “AI metrics,” a field that could mirror psychometrics (the study of human intelligence tests). Just as psychologists have developed standardized ways to measure cognitive abilities in humans, Gignac and Szodorai suggest that computer scientists need standardized measures to assess AI performance.
Why is this important? Because the authors argue that most current AI systems are not truly intelligent, but instead demonstrate artificial achievement—the ability to excel at specific tasks they’ve been trained for, without showing the adaptability required to solve completely new problems. For example, a chess-playing AI can defeat a human grandmaster but fails to understand a simple task like folding a shirt.
This new AI metrics field would ensure that we test AI systems not only for expertise in certain tasks but also for their broader problem-solving capacities, aligning AI assessments more closely with the way we measure human intelligence.
Artificial General Intelligence: A New Way to Think About AGI
The research also has major implications for the elusive goal of Artificial General Intelligence (AGI). AGI refers to machines capable of performing any intellectual task a human can, and it’s often seen as the ultimate goal in AI development.
In their paper, Gignac and Szodorai draw a parallel between AGI and the concept of general intelligence (or “g”) in human psychology. Just as people who are good at one type of cognitive task tend to be good at others, AGI would need to show consistent performance across a variety of tasks. This contrasts with today’s specialized AI systems, which tend to excel in narrow domains.
If we can measure AGI using the same principles of general intelligence applied to humans, researchers argue, we’ll be better equipped to develop machines that aren’t just good at one thing but can think and adapt like humans across multiple areas.
Implications for AI Development and Beyond
So, what does all of this mean for the future of AI? According to the authors, it could lead to more sophisticated AI systems—ones that do more than excel at a specific set of instructions. By focusing on general intelligence and adaptability, we could create machines that understand and tackle unfamiliar problems, just as humans do.
Their work also suggests that interdisciplinary collaboration between psychology and computer science could be the key to making progress. By agreeing on a common language and set of principles for intelligence, these fields can work together to push the boundaries of what AI can achieve.
Conclusion: A New Era for Intelligence Research?
With Defining Intelligence: Bridging the Gap Between Human and Artificial Perspectives, Gignac and Szodorai offer a fresh approach to one of the most complex issues in science and technology. By unifying definitions and encouraging new ways to measure intelligence, both in humans and machines, they lay the groundwork for advancements that could lead us closer to true artificial intelligence.
If adopted, this framework could revolutionize the way we think about and build intelligent systems, potentially narrowing the gap between human cognitive abilities and AI capabilities. The next step? Building machines that don’t just perform—but think.
Source: Defining intelligence: Bridging the gap between human and artificial perspectives