The evolution of science and technology: How humanity evolved with it?
Fruit and honey fuelled the early evolution of the human brain
Jennie Brand-Miller, University of Sydney; David Raubenheimer, University of Sydney, and Les Copeland, University of Sydney
Intelligence is an energetically expensive luxury – as the rise of artificial intelligence has reminded us. The human brain is no exception.
Comprising roughly 2% of the body weight, it uses one fifth the body’s energy in the resting state – compared with less than half of this in non-human primates. A five-year-old child devotes 66% of the energy they need to stay alive to their brain.
How did our ancestors foot the energy bill to run their uniquely large brains? Eating more meat is often considered to be the answer. But there’s a catch. The brain relies on a form of fuel that is not present in meat: glucose.
Our new study, published today in Science, shows that carbohydrates contributed more than half our total energy requirements over four million years of evolution. It holds important clues for why we crave sweet foods today – and how we can eat more healthily.
The matter of meat
Many anthropologists have credited meat eating as the stimulus to produce a large brain. After all, it required tools to butcher the carcass and access the fat-rich marrow inside bones.
Protein and fat in fruit and leaves are dilute, requiring hours of chewing, while animal foods are dense sources that can be devoured quickly. Bone marrow is a rich source of essential fats.
In truth, humans do not require more protein as a proportion of energy than other primates.
Our increasingly large brains and high reproductive rate demanded carbohydrate calories (found in plants but not meat), while our taller and heavier bodies needed fat calories to move those big muscles.
Although the body can synthesise glucose from precursors such as amino acids, the process is finite and energetically inefficient. Furthermore, there are limits on using just protein as fuel. For example, it can lead to a type of poisoning known as “rabbit starvation”.
A minimum amount of dietary carbohydrate was necessary. Our new study shows that, for much of evolution, the sugars in fruit and honey were the source.
Modelling ancient diets
We modelled the diets of hominins – the group consisting of humans and our immediate ancestors – over four million years of evolution.
First, we calculated overall demand for glucose by the organs and tissues which use it as their primary source of energy. Apart from the brain, red blood cells and the kidneys require glucose.
We then accounted for reproductive needs. The fetus and placenta use glucose not just as an energy source but as a structural component of growing tissues. Synthesis of DNA, RNA and nerve cell membranes requires glucose. During lactation, women use about 80g of glucose each day to produce the sugars in human milk.
Then we modelled the availability of macronutrients – carbohydrates, fats and proteins – from foods, starting with the diminutive ancient ape known as Lucy (Australopithecus afarensis).
This early ancestor of ours walked on two legs, and was likely to be a ripe fruit specialist like chimpanzees today. Over two thirds of her energy came from naturally-occurring sugars.
Indeed, some scientists think frugivory – a feeding strategy primarily characterised by eating fruit – kick-started the evolution of large brains because, living in tropical forests, our ancestors required good cognition to remember when and where the best fruits were ripening. They needed strategic thinking to beat the birds and other competitors.
We finished up with the known diet composition of contemporary human foragers in warm climates. In six incremental steps, we incorporated increasing proportions of animal-based food, starting with 5% of calories and finishing with 35–50%.
Around one million years ago, mastery of fire allowed cooked starch, which unlike raw starch can be easily digested to provide glucose, to replace some of the sugars. Relatively recently, about 100,000 years ago, grinding stones and hearths indicate that the starch inside cereal grains became more accessible.
Lessons for modern diets
Did early hominins consume sufficient carbohydrate to cover the obligatory demands of the brain and other tissues? Yes, if you were a male, but only just if you were a pregnant female.
As we ventured out of tropical environments into cold and arid territory, the intake of carbohydrates would have become limiting. Plants would be plentiful, along with protein and marrow fat, but fruit and honey would be seasonal.
We speculate that limited amounts of dietary carbohydrate selected for genes that result in higher blood glucose levels. This would improve the growth and future survival of the fetus.
Today, the same genes likely predispose us to type two diabetes and cardiovascular disease. Low carbohydrate diets may therefore be helpful in specific clinical contexts.
But our findings provide an evolutionary explanation why healthy humans require about half their energy as carbohydrates. They also give us insight into why humans crave sweetness – a pleasurable signal on the tongue that encouraged foods that fuelled the mind and body millions of years ago.
Intrinsically, sugars are highly reactive molecules that are bundled in nature with antioxidants and other natural compounds that reduce harm within the cell.
Ideally, we consume them in that form – as fruit – rather than refined sugars.![]()
Jennie Brand-Miller, Emeritus Professor of Human Nutrition, University of Sydney; David Raubenheimer, Leonard P. Ullman Chair in Nutritional Ecology, Nutrition Theme Leader Charles Perkins Centre, University of Sydney, and Les Copeland, Professor of Agriculture, University of Sydney
This article is republished from The Conversation under a Creative Commons license. Read the original article.
Brain scans provide world‑first evidence dogs can distinguish between human fear and sadness
Mia Cobb, The University of Melbourne
What causes a dog to slink away from a cranky person, when they will quietly approach and lean against someone who is weeping? We’ve all seen it – they can respond to our feelings. And science agrees dogs have emotions too.
These social skills could underpin dogs’ success in living with us. But do you think your dog could tell an angry person’s face from a sad or fearful one?
New research published in the journal iScience explored that question, and revealed interesting findings from magnetic resonance imaging (MRI) scans of dogs’ brains.
Scanning dogs’ brains
Dogs are sensitive to human faces. They look longer in response to our emotional expressions and sounds compared with neutral ones.
Scientists weren’t sure whether dogs were just differentiating “good mood” (happy) from “bad mood” (angry, fearful or sad), or treating these expressions as genuine indicators of different emotions.
The new study, by Raúl Hernández-Pérez, a neuroscientist at the University of Vienna, and colleagues, explored this gap using MRI to scan pet dogs’ brains while they were viewing photos of human faces.
Building on their earlier work, the researchers found evidence that dogs do process images of our distinct emotional expressions differently.
The researchers used machine learning and showed that when looking at a dog’s whole brain, a different brain region was activated to distinguish between fear and sadness (the right rostral suprasylvian gyrus, to be precise), than between fear and anger (this was in the right mid ectosylvian gyrus and left splenial gyrus).
The analysis didn’t detect a difference in the brain areas activated when dogs were shown images of human anger and sadness. Fear stood out from the other negative emotions.
This raises the question: why?
It might be that fear and anger are simply more attention-grabbing than sadness.
Other research has found dogs react to fear and anger faster, and with a bigger physical response, such as a raised heart rate. This is likely because they’re the expressions most likely to call for a quick response from dogs to stay safe.
Sadness is less likely to pose a direct threat to dogs living with people, so they experience less urgency to respond to it. We know some dogs don’t respond with the heroic Lassie behaviour we might like when we are in distress.
Although the numbers in this new research were small (eight and twelve dogs across the two parts of the study), this is the first MRI-based proof-of-concept evidence that dog brains can distinguish between two human facial expressions of distinct negative emotions. It indicates dogs’ neural representation of our emotion goes beyond a simple valence (good/bad) split.
This shows us that perceiving emotion in others (even across species) isn’t handled by one single “emotion centre” in the brain – in dogs, in humans, or in other animal species. It’s spread across a network of regions working together as part of living socially.
The authors of this study point out that using still images of humans is a very people-centred way to explore how dogs interpret our emotional states.
We know dogs live in rich sensory worlds where the scent and sound of our speech also convey emotions, shaping how dogs respond to us.
In fact, even wolves who have grown up around people show the same kind of response to the odour of human fear as dogs. This highlights the important role of learning, as distinct from evolutionary differences in canid bodies or how they respond to people. Dogs (and wolves) are learning about us in every interaction we have with them.
Dogs are adept at watching, smelling, and listening to our emotions, learning how these signals predict our behaviours toward them, and using this information to live with people harmoniously.
Returning the favour, learning more about how dogs express their emotions seems like the least we can do.![]()
Mia Cobb, Research Fellow, Animal Welfare Science Centre, The University of Melbourne
This article is republished from The Conversation under a Creative Commons license. Read the original article.
Human vision: what we actually see – and don’t see – tells us a lot about consciousness
Henry Taylor, University of Birmingham
What can you see right now? This might seem like a silly question, but what enters your consciousness is not the whole story when it comes to vision. A great deal of visual processing in the brain goes on well below our conscious awareness.
Some studies have probed the unconscious depths of vision. One source of evidence comes from the neurological condition known as blindsight, which is caused by damage to areas of the brain involved in processing visual information. People with blindsight report that they are unable to see, either entirely or in a portion of their visual field. However, when asked to guess what is there, they can often do so with remarkable accuracy.
For example, in an experiment published in 2004 on someone with blindsight, a black bar was displayed in the portion of the visual field to which the person was blind. The person was asked to “guess” whether the bar was vertical or horizontal.
Despite denying any conscious awareness of the bar, the participant could answer correctly at a level well above chance. The participant even showed evidence of being able to pay attention to the bar – they were faster to respond when an arrow (placed in a healthy area of their visual field) correctly indicated the location of the bar.
The most popular interpretation (though not the only one) is that people with blindsight can see these objects, but not see them consciously. They see what is there, but it all goes on unconsciously, below their awareness.
The phenomenon of inattentional blindness seems to show you can see without the information crossing into your consciousness. Anyone can experience inattentional blindness. The phenomenon has been known about for a long time, but we can most easily get a handle on it by looking at a well-known experiment reported in 1999.
In this experiment, participants are shown a video of people playing basketball, and told to count the number of passes between the players wearing a white shirt. If you’ve never done this before, I urge to you stop reading now and watch the video.
In many cases, people are so busy counting the passes that they completely miss a large gorilla walking across the middle of the scene and beating its chest, then walking off. The gorilla’s right there, in the centre of your visual field. Light from the gorilla enters your eyes, and is processed in the visual system, but somehow you missed it, because you weren’t paying attention to it.
The gorilla has more to teach us. In another experiment reported in 2013, radiologists were given a series of lung scans. They were told to look for nodules (which show up as small light coloured circles) on each scan. In one of the scans, a large picture of a dancing gorilla was superimposed on top of the lung scan. In this study, 83% of the radiologists failed to spot it, even though it was 48 times bigger than the average nodule they were looking for. Some of them even looked directly at the gorilla and still didn’t notice it!
The interpretation of these experiments is controversial. Some scientists suggest that in these kinds of cases, you consciously see the gorilla, but immediately forget it (although a dancing gorilla in someone’s lung doesn’t seem like the kind of thing you’d forget). Others argue that you see the gorilla, but the information never made its way into consciousness. You saw the gorilla, but unconsciously.
Let’s assume that in the case of blindsight, and inattentional blindness, the information is seen, but didn’t make it all the way to consciousness. Then, the question is: what makes some information conscious, rather than the information that stays unconscious? This is one of the central questions for consciousness studies in philosophy, psychology and neuroscience.
The brain’s loudspeaker
There’s no agreement on which is the best theory of consciousness, but in my opinion, the strongest contender is the global neuronal workspace theory.
According to this theory, consciousness is all to do with a particular area of the brain which is the seat of the “workspace”. The workspace is a system with a small capacity, so it can’t hold a lot of information at any one time. The job of the workspace is to take unconscious information and broadcast it to lots of different networks all across the brain. Global neuronal workspace theorists say that broadcasting the information in this way is what makes it conscious.
The job of the workspace is to act like the brain’s loudspeaker, and consciousness is the information that gets broadcast. The workspace takes unconscious information and boosts it so that many of the different systems in the brain hear about it and can use that information in their own processes. The late philosopher Daniel Dennett used to call consciousness “fame in the brain”. The workspace idea is similar.
One of the most striking implications of the global neuronal workspace theory is how little information makes it to consciousness. Since the workspace has quite a small capacity, it follows that we can only ever be conscious of a little at a time. We might think there’s a rich visual world in front of us, full of details, all of which we’re conscious of, but really – according to the theory – we’re only ever conscious of a small portion of that.
Some philosophers and scientists have objected to the theory on these grounds. They suggest that consciousness “overflows” the workspace: we are conscious of more information than can “fit” into the workspace at any one time. Even with these debates still ongoing, I think the global neuronal workspace theory gives us a reasonably clear answer to the question of what consciousness is for, and how it interacts with other systems in the brain.
In our brains, consciousness is only the tip of a very large iceberg. But the global neuronal workspace theory might give us insight into what makes that tip so special.![]()
Henry Taylor, Associate Professor, Department of Philosophy, University of Birmingham
This article is republished from The Conversation under a Creative Commons license. Read the original article.
First Human Cornea Transplant Using 3D Printed, Lab-Grown Tissue Restores Sight in a ‘Game Changer’ for Millions Who are Blind

The first successful human implant of a 3D-printed cornea made from human eye cells cultured in a laboratory has restored a patient’s sight.
The North Carolina-based company that developed the cornea described the procedure as a ‘world first’—and a major milestone toward its goal of alleviating the lack of available donor tissue and long wait-times for people seeking transplants.
According to Precise Bio, its robotic bio-fabrication approach could potentially turn a single donated cornea into hundreds of lab-grown grafts, at a time when there’s currently only one available for an estimated 70 patients who need one to see.
“This achievement marks a turning point for regenerative ophthalmology—a moment of real hope for millions living with corneal blindness,” Aryeh Batt, Precise Bio’s co-founder and CEO, said in a statement.
“For the first time, a corneal implant manufactured entirely in the lab from cultured human corneal cells, rather than direct donor tissue, has been successfully implanted in a patient.”
The company said the transplant was performed Oct. 29 in one eye of a patient who was considered legally blind.
“This is a game changer. We’ve witnessed a cornea created in the lab, from living human cells, bring sight back to a human being,” said Dr. Michael Mimouni, director of the cornea unit at Rambam Medical Center in Israel, who performed the procedure.
“It was an unforgettable moment—a glimpse into a future where no one will have to live in darkness because of a shortage of donor tissue.”
Dubbed PB-001, the implant is designed to match the optical clarity, transparency and bio-mechanical properties of a native cornea. Previously tested in animal models, the company said its graft is capable of integrating with a patient’s own tissue.
The outer layer of the eye—covering the iris and pupil—can end up clouding a person’s vision following injuries, infections, scarring and other conditions. PB-001 is currently being tested in a single-arm phase 1 trial in Israel, which aims to enroll between 10 and 15 participants with excess fluid buildups in the cornea due to dysfunction within its inner cell layers.
Precise Bio said it plans to announce top-line results from the study in the second half of 2026, tracking six-month efficacy outcomes.
The corneas are designed to be compatible with current surgery hardware and workflows. Shipped under long-term cryopreservation, it is delivered preloaded on standard delivery devices and unrolls during implantation to form a natural corneal shape.
“PB-001 has the potential to offer a new, standardized solution to one of ophthalmology’s most urgent needs—reliable, safe, and effective corneal replacement,” said Anthony Atala, M.D., co-founder of Precise Bio and director of the Wake Forest Institute for Regenerative Medicine.
“The ability to produce patient-ready tissue on demand could lead the way towards reshaping transplant medicine as we know it.”(Edited from original article by Conor Hale) First Human Cornea Transplant Using 3D Printed, Lab-Grown Tissue Restores Sight in a ‘Game Changer’ for Millions Who are Blind
The science of weight loss – and why your brain is wired to keep you fat

For decades, we’ve been told that weight loss is a matter of willpower: eat less, move more. But modern science has proven this isn’t actually the case.
More on that in a moment. But first, let’s go back a few hundred thousand years to examine our early human ancestors. Because we can blame a lot of the difficulty we have with weight loss today on our predecessors of the past – maybe the ultimate case of blame the parents.
For our early ancestors, body fat was a lifeline: too little could mean starvation, too much could slow you down. Over time, the human body became remarkably good at guarding its energy reserves through complex biological defences wired into the brain. But in a world where food is everywhere and movement is optional, those same systems that once helped us survive uncertainty now make it difficult to lose weight.
When someone loses weight, the body reacts as if it were a threat to survival. Hunger hormones surge, food cravings intensify and energy expenditure drops. These adaptations evolved to optimise energy storage and usage in environments with fluctuating food availability. But today, with our easy access to cheap, calorie-dense junk food and sedentary routines, those same adaptations that once helped us to survive can cause us a few issues.
As we found in our recent research, our brains also have powerful mechanisms for defending body weight – and can sort of “remember” what that weight used to be. For our ancient ancestors, this meant that if weight was lost in hard times, their bodies would be able to “get back” to their usual weight during better times.
But for us modern humans, it means that our brains and bodies remember any excess weight gain as though our survival and lives depend upon it. So in effect, once the body has been heavier, the brain comes to treat that higher weight as the new normal – a level it feels compelled to defend.
The fact that our bodies have this capacity to “remember” our previous heavier weight helps to explain why so many people regain weight after dieting. But as the science shows, this weight regain is not due to a lack of discipline; rather, our biology is doing exactly what it evolved to do: defend against weight loss.
Hacking biology
This is where weight-loss medications such as Wegovy and Mounjaro have offered fresh hope. They work by mimicking gut hormones that tell the brain to curb appetite.
But not everyone responds well to such drugs. For some, the side effects can make them difficult to stick with, and for others, the drugs don’t seem to lead to weight loss at all. It’s also often the case that once treatment stops, biology reasserts itself – and the lost weight returns.
Advances in obesity and metabolism research may mean that it’s possible for future therapies to be able to turn down these signals that drive the body back to its original weight, even beyond the treatment period.
Research is also showing that good health isn’t the same thing as “a good weight”. As in, exercise, good sleep, balanced nutrition, and mental wellbeing can all improve heart and metabolic health, even if the number on the scales barely moves.
A whole society approach
Of course, obesity isn’t just an individual problem – it takes a society-wide approach to truly tackle the root causes. And research suggests that a number of preventative measures might make a difference – things such as investing in healthier school meals, reducing the marketing of junk food to children, designing neighbourhoods where walking and cycling are prioritised over cars, and restaurants having standardised food portions.
Scientists are also paying close attention to key early-life stages – from pregnancy to around the age of seven – when a child’s weight regulation system is particularly malleable.
Indeed, research has found that things like what parents eat, how infants are fed, and early lifestyle habits can all shape how the brain controls appetite and fat storage for years to come.
If you’re looking to lose weight, there are still things you can do – mainly by focusing less on crash diets and more on sustainable habits that support overall wellbeing. Prioritising sleep helps regulate appetite, for example, while regular activity – even walking – can improve your blood sugar levels and heart health.
The bottom line though is that obesity is not a personal failure, but rather a biological condition shaped by our brains, our genes, and the environments we live in. The good news is that advances in neuroscience and pharmacology are offering new opportunities in terms of treatments, while prevention strategies can shift the landscape for future generations.
So if you’ve struggled to lose weight and keep it off, know that you’re not alone, and it’s not your fault. The brain is a formidable opponent. But with science, medicine and smarter policies, we’re beginning to change the rules of the game.
This article was commissioned as part of a partnership collaboration between Videnskab.dk and The Conversation. You can read the Danish version of this article, here.![]()
Valdemar Brimnes Ingemann Johansen, PhD Fellow in the Faculty of Health and Medical Sciences, University of Copenhagen and Christoffer Clemmensen, Associate Professor and Group Leader, Novo Nordisk Foundation Center for Basic Metabolic Research, University of Copenhagen
This article is republished from The Conversation under a Creative Commons license. Read the original article.
Australia leads first human trial of one-time gene editing therapy to halve bad cholesterol
Scientists Regrow Retina Cells to Tackle Leading Cause of Blindness Using Nanotechnology

Blue, green, brown, or something in between – the science of eye colour explained
You’re introduced to someone and your attention catches on their eyes. They might be a rich, earthy brown, a pale blue, or the rare green that shifts with every flicker of light. Eyes have a way of holding us, of sparking recognition or curiosity before a single word is spoken. They are often the first thing we notice about someone, and sometimes the feature we remember most.
Across the world, human eyes span a wide palette. Brown is by far the most common shade, especially in Africa and Asia, while blue is most often seen in northern and eastern Europe. Green is the rarest of all, found in only about 2% of the global population. Hazel eyes add even more diversity, often appearing to shift between green and brown depending on the light.
So, what lies behind these differences?
It’s all in the melanin
The answer rests in the iris, the coloured ring of tissue that surrounds the pupil. Here, a pigment called melanin does most of the work.
Brown eyes contain a high concentration of melanin, which absorbs light and creates their darker appearance. Blue eyes contain very little melanin. Their colour doesn’t come from pigment at all but from the scattering of light within the iris, a physical effect known as the Tyndall effect, a bit like the effect that makes the sky look blue.
In blue eyes, the shorter wavelengths of light (such as blue) are scattered more effectively than longer wavelengths like red or yellow. Due to the low concentration of melanin, less light is absorbed, allowing the scattered blue light to dominate what we perceive. This blue hue results not from pigment but from the way light interacts with the eye’s structure.
Green eyes result from a balance, a moderate amount of melanin layered with light scattering. Hazel eyes are more complex still. Uneven melanin distribution in the iris creates a mosaic of colour that can shift depending on the surrounding ambient light.
What have genes got to do with it?
The genetics of eye colour is just as fascinating.
For a long time, scientists believed a simple “brown beats blue” model, controlled by a single gene. Research now shows the reality is much more complex. Many genes contribute to determining eye colour. This explains why children in the same family can have dramatically different eye colours, and why two blue-eyed parents can sometimes have a child with green or even light brown eyes.
Eye colour also changes over time. Many babies of European ancestry are born with blue or grey eyes because their melanin levels are still low. As pigment gradually builds up over the first few years of life, those blue eyes may shift to green or brown.
In adulthood, eye colour tends to be more stable, though small changes in appearance are common depending on lighting, clothing, or pupil size. For example, blue-grey eyes can appear very blue, very grey or even a little green depending on ambient light. More permanent shifts are rarer but can occur as people age, or in response to certain medical conditions that affect melanin in the iris.
The real curiosities
Then there are the real curiosities.
Heterochromia, where one eye is a different colour from the other, or one iris contains two distinct colours, is rare but striking. It can be genetic, the result of injury, or linked to specific health conditions. Celebrities such as Kate Bosworth and Mila Kunis are well-known examples. Musician David Bowie’s eyes appeared as different colours because of a permanently dilated pupil after an accident, giving the illusion of heterochromia.
In the end, eye colour is more than just a quirk of genetics and physics. It’s a reminder of how biology and beauty intertwine. Each iris is like a tiny universe, rings of pigment, flecks of gold, or pools of deep brown that catch the light differently every time you look.
Eyes don’t just let us see the world, they also connect us to one another. Whether blue, green, brown, or something in-between, every pair tells a story that’s utterly unique, one of heritage, individuality, and the quiet wonder of being human.![]()
Davinia Beaver, Postdoctoral research fellow, Clem Jones Centre for Regenerative Medicine, Bond University
This article is republished from The Conversation under a Creative Commons license. Read the original article.
Scientists Define a Color Never Before Seen by Human Eyes, Called 'Olo'–a Blue-Green of Intense Saturation
Photo by Hamish on UnsplashDiscovery of Genetically-Varied Worms in Chernobyl Could Help Human Cancer Research
Worms collected in the Chornobyl Exclusion Zone – SWNS / New York University
The ruins of Reactor 4, Chernobyl Exclusion Zone. credit Matt Shalvatis – CC BY-4.0. SAScientists Regrow Retina Cells to Tackle Leading Cause of Blindness Using Nanotechnology

This Common Fungus Found on Human Skin Wipes Out Deadly Superbug Staph Infections

Scientists use AI to reveal the neural dynamics of human conversation
An AI system has reached human level on a test for ‘general intelligence’. Here’s what that means
A new artificial intelligence (AI) model has just achieved human-level results on a test designed to measure “general intelligence”.
On December 20, OpenAI’s o3 system scored 85% on the ARC-AGI benchmark, well above the previous AI best score of 55% and on par with the average human score. It also scored well on a very difficult mathematics test.
Creating artificial general intelligence, or AGI, is the stated goal of all the major AI research labs. At first glance, OpenAI appears to have at least made a significant step towards this goal.
While scepticism remains, many AI researchers and developers feel something just changed. For many, the prospect of AGI now seems more real, urgent and closer than anticipated. Are they right?
Generalisation and intelligence
To understand what the o3 result means, you need to understand what the ARC-AGI test is all about. In technical terms, it’s a test of an AI system’s “sample efficiency” in adapting to something new – how many examples of a novel situation the system needs to see to figure out how it works.
An AI system like ChatGPT (GPT-4) is not very sample efficient. It was “trained” on millions of examples of human text, constructing probabilistic “rules” about which combinations of words are most likely.
The result is pretty good at common tasks. It is bad at uncommon tasks, because it has less data (fewer samples) about those tasks.
Until AI systems can learn from small numbers of examples and adapt with more sample efficiency, they will only be used for very repetitive jobs and ones where the occasional failure is tolerable.
The ability to accurately solve previously unknown or novel problems from limited samples of data is known as the capacity to generalise. It is widely considered a necessary, even fundamental, element of intelligence.
Grids and patterns
The ARC-AGI benchmark tests for sample efficient adaptation using little grid square problems like the one below. The AI needs to figure out the pattern that turns the grid on the left into the grid on the right.
Each question gives three examples to learn from. The AI system then needs to figure out the rules that “generalise” from the three examples to the fourth.
These are a lot like the IQ tests sometimes you might remember from school.
Weak rules and adaptation
We don’t know exactly how OpenAI has done it, but the results suggest the o3 model is highly adaptable. From just a few examples, it finds rules that can be generalised.
To figure out a pattern, we shouldn’t make any unnecessary assumptions, or be more specific than we really have to be. In theory, if you can identify the “weakest” rules that do what you want, then you have maximised your ability to adapt to new situations.
What do we mean by the weakest rules? The technical definition is complicated, but weaker rules are usually ones that can be described in simpler statements.
In the example above, a plain English expression of the rule might be something like: “Any shape with a protruding line will move to the end of that line and ‘cover up’ any other shapes it overlaps with.”
Searching chains of thought?
While we don’t know how OpenAI achieved this result just yet, it seems unlikely they deliberately optimised the o3 system to find weak rules. However, to succeed at the ARC-AGI tasks it must be finding them.
We do know that OpenAI started with a general-purpose version of the o3 model (which differs from most other models, because it can spend more time “thinking” about difficult questions) and then trained it specifically for the ARC-AGI test.
French AI researcher Francois Chollet, who designed the benchmark, believes o3 searches through different “chains of thought” describing steps to solve the task. It would then choose the “best” according to some loosely defined rule, or “heuristic”.
This would be “not dissimilar” to how Google’s AlphaGo system searched through different possible sequences of moves to beat the world Go champion.
You can think of these chains of thought like programs that fit the examples. Of course, if it is like the Go-playing AI, then it needs a heuristic, or loose rule, to decide which program is best.
There could be thousands of different seemingly equally valid programs generated. That heuristic could be “choose the weakest” or “choose the simplest”.
However, if it is like AlphaGo then they simply had an AI create a heuristic. This was the process for AlphaGo. Google trained a model to rate different sequences of moves as better or worse than others.
What we still don’t know
The question then is, is this really closer to AGI? If that is how o3 works, then the underlying model might not be much better than previous models.
The concepts the model learns from language might not be any more suitable for generalisation than before. Instead, we may just be seeing a more generalisable “chain of thought” found through the extra steps of training a heuristic specialised to this test. The proof, as always, will be in the pudding.
Almost everything about o3 remains unknown. OpenAI has limited disclosure to a few media presentations and early testing to a handful of researchers, laboratories and AI safety institutions.
Truly understanding the potential of o3 will require extensive work, including evaluations, an understanding of the distribution of its capacities, how often it fails and how often it succeeds.
When o3 is finally released, we’ll have a much better idea of whether it is approximately as adaptable as an average human.
If so, it could have a huge, revolutionary, economic impact, ushering in a new era of self-improving accelerated intelligence. We will require new benchmarks for AGI itself and serious consideration of how it ought to be governed.
If not, then this will still be an impressive result. However, everyday life will remain much the same.![]()
Michael Timothy Bennett, PhD Student, School of Computing, Australian National University and Elija Perrier, Research Fellow, Stanford Center for Responsible Quantum Technology, Stanford University
This article is republished from The Conversation under a Creative Commons license. Read the original article.
