Reverse Engineering Human Intelligence for Advanced AI and AGI: A Cognitive Science Approach

Josh Tenenbaum, a professor at MIT leading the computational cognitive science group, argues that despite impressive advancements in AI technologies like deep learning, true Artificial General Intelligence (AGI) remains elusive. Current AI excels at pattern recognition and specific tasks, often surpassing human performance in narrow domains like playing Go, but fundamentally lacks common sense, flexible general-purpose intelligence, and the ability to model the world in a rich, explanatory way. He highlights that these systems are engineered for single tasks, requiring vast data and resources, and cannot generalize or understand basic concepts like what a 'game' is, unlike a human child who learns efficiently from minimal experience.

Tenenbaum posits that intelligence is far more than pattern recognition; it involves actively modeling the world to explain, understand, imagine, plan, and solve problems. This includes building and refining mental models, sharing them through communication, and learning from others. He advocates for a 'reverse engineering' approach, where insights from human cognitive science and neuroscience are formalized mathematically and computationally to build intelligent machines. This methodology, he notes, has historical precedent, as many foundational algorithms in deep learning and reinforcement learning originated from theoretical psychology and cognitive science research decades ago, demonstrating scientists thinking like engineers.

The Center for Brains, Minds, and Machines (CBMM) at MIT embodies this vision, aiming to bridge the science of human intelligence with the engineering of machine intelligence. A key focus area is visual intelligence, where humans effortlessly construct a rich, 3D understanding of their surroundings from sparse, foveated visual input, tracking objects, people, and even their mental states. Tenenbaum illustrates the limitations of current AI with examples from image captioning bots, which, despite high accuracy on curated datasets, often produce nonsensical or superficial descriptions when faced with novel or subtly complex real-world images, demonstrating 'data set overfitting' rather than genuine understanding.

Ultimately, Tenenbaum proposes a long-term research roadmap (50+ years) focused on addressing fundamental questions of human intelligence, such as consciousness, meaning in language, real learning, culture, and creativity. He believes that a science-based, reverse-engineering approach, rather than the current industry-driven trajectory optimized for narrow tasks, offers the highest value and most promising route to developing machines that can truly see, learn, and think like people, leading to profound engineering payoffs and a deeper understanding of what it means to be human.

we don't really have any real AI basically we have what I like to call AI technologies which are systems that do things we used to think that only humans could do and now we have machines that do them often quite well maybe even better than any human who's ever lived right like a machine that plays go but none of these systems I would say are truly intelligent none of them have anything like common sense none of them have anything like the flexible general-purpose intelligence that each of you might use to learn every one of these skills or tasks

alphago might beat the worlds best but it can't drive to the match or even tell you that go it what go is it can't even tell you the go is a game because it doesn't even know what a game is right

intelligence is about a lot more in particular it's about modeling the world and think about all the activities that a human does so model the world that that go beyond just say recognizing patterns and data but actually trying to explain and understand what we see for instance

by reverse engineering how intelligence works in the human mind and brain that will give us a route to engineering these abilities in machines

if we approach cognitive science and neuroscience like an engineer where so the output of our science isn't just a description of the brain or the mind in words but in the same terms that an engineer would use to build an intelligence system then that will be both the basis for a much more rigorous and deeply insightful science but also direct translation of those insights into engineering applications

what you see here is already a long history of scientists thinking like engineers these are people who are in psychology or cognitive science departments and publishing in those places but by formalizing even very basic insights about how humans might learn or how you know brains might learn in the right kind of math that led to of course progress on the science side but it led to all the engineering that we see now

when we talk about visual intelligence this is the whole stuff we're talking about and you can start to see how it turns into basic questions I think of not of what we might call the beginnings of consciousness at least our awareness of ourself in the world and of ourselves as a self in the world but also other aspects of higher-level intelligence and cognition that are not just about perception like symbols right to describe even to ourselves what's around us and where we are and what we can do with it you have to go beyond just what we would normally call the stuff of perception to say the thoughts in somebody's head and your own thoughts about that

my sense is no I think that it's not when I say no I don't mean like it can't happen or it won't happen what I mean is the highest value the highest expected route right now is to take this more science-based reverse engineering approach and that if at least if you follow the current trajectory that industry incentives especially optimized for it's not even really trying to take us to these things

what you can see when you really dig into these things is there's often a lot of what I would call data set overfitting it's not overfitting to the training set but it's overfitting to whatever are the particular characteristics of this data set