What is Intelligence?
“Will you not then ponder”
When prophet Ibrahim AS looked up to the heavens, he pointed out an obvious observation: that the celestial bodies his people worshiped all invariably faded out of sight. He famously inferred that these celestial bodies must therefore not be divine. This ability to infer truths from experiences is a hallmark of the intelligence that characterised prophet Ibrahim AS, something we are repeatedly encouraged to emulate. Put more broadly, an intelligent person can discover patterns and rules that relate past experiences, and use them to infer solutions to problems they’ve never met before. But beyond such an intuition, basic questions about the nature of intelligence remain unanswered: What actually is intelligence? We know what it looks like to be intelligent; flexible behaviour in the face of new unknowns. But what is its essence? How does it come about? Can we relate it to tangible mechanisms in the brain? Recent progress in cognitive neuroscience is bringing us close to an answer.
I study the relationship between neurons, the cells that do much of the processing in the brain, and cognition. In particular, I study the activity of neurons in the Hippocampus and Frontal cortex, brain regions linked to higher cognition in mammals, including humans. A large body of research has established that neurons in the Hippocampus contain a record of past experiences (O’Keefe and Dostrovsky, 1971; Scoville and Milner, 1957). What’s fascinating is that this record is not random, but highly organised. Neurons encoding information about previously visited locations are more coactive with those encoding adjacent locations, while distant locations show less overlap in their corresponding activity. This “map-like” representation of the outside world is the best neural substrate we have for flexible navigation. For example, when faced with a new obstacle, a mouse can immediately take the shortest alternative route to its goal, one that it never took before. Such flexibility is paralleled by flexibility of the neurons mapping the space the animal is in. Remarkable experiments show that neurons in the animal’s Hippocampus “preplay” a path they’re planning to take, that is they are active in the sequence the animal plans to take, even when the animal has yet to experience that path (Ólafsdóttir et al., 2015; Pfeiffer and Foster, 2013). Neurons can infer new routes!
My research aims to address how these mental maps are formed and modified as we learn. Working with a team of researchers at the MRC Brain Network Dynamics Unit, University of Oxford, I found that groups of neurons in the Hippocampus can cooperate to encode not only where the animal is but what it needs to do to reach its goal (El-Gaby et al., 2019). Moreover, by precisely manipulating communication between a defined group of neurons as animals learn, I found that learning these representations depends on a subset of neural connections that are particularly malleable (El-Gaby et al., 2016; 2019). This work adds to growing efforts to identify cellular mechanisms that underlie key cognitive processes.
Moving further forward in the brain, neurons in the Frontal cortex play a complementary role to that of the spatial maps in the Hippocampus. Here, neurons respond less to the details of individual experiences and more to the general rules that are common amongst experiences (Morrissey et al., 2017; Zhou et al., 2020). This has led to the hypothesis that frontal neurons may be the seat of abstract concepts in the brain. However, little is known about how these abstract representations emerge. Part of the problem is the lack of an appropriate experimental paradigm to probe these cellular mechanisms. Humans readily use sophisticated forms of abstraction: knowing the general rules that govern the world around us allows us to learn rapidly and with a level of data efficiency that surpasses that of the most sophisticated artificial intelligence. However, practical considerations mean that we are limited in the level of mechanistic detail we can obtain from human brains. Studies of animals with analogous brain regions therefore become necessary.
Working at the Nuffield Department of Clinical Neurosciences, University of Oxford, I recently found that mice can exhibit behavioural evidence for abstract conceptual knowledge, learning not only individual sequences of actions to obtain rewards but the overall rule governing these sequences. Faced with a new sequence, they can infer the final step without ever experiencing it: a tell-tale sign of abstract knowledge. Given the ease of precisely recording and manipulating neural activity in mice, they are ideally suited for addressing questions about the cellular mechanisms underlying abstraction. My current work aims to harness this experimental tractability and newly found cognitive abilities in mice to understand how the mammalian brain forms abstract representations. The hope is that this work will, in time, translate directly into insights about how the human brain forms abstract concepts, a key step in understanding the essence of intelligence.