Posted on

Boosting AI’s Intelligence with Metacognitive Primitives

Over the past year or so, AI experts, like Ilya Sutskever in his Neurips 2024 talk, have been raising concerns that AI reasoning might be hitting a wall. It seems that simply throwing more data and computing power at the problem is giving us less and less in return, and models are struggling with complex thinking tasks. Maybe it’s time to explore other facets of human reasoning and intelligence, rather than just relying on sheer computational force.

At its core, a key part of human intelligence is our ability to pick out just the right information from our memories to help us solve the problem at hand. For instance, imagine a toddler seeing a puppy in a park. If they’ve never encountered a puppy before, they might feel a bit scared or unsure. But if they’ve seen their friend playing with their puppy, or watched their neighbors’ dogs, they can draw on those experiences and decide to go ahead and pet the new puppy. As we get older, we start doing this for much more intricate situations – we take ideas from one area and apply them to another when the patterns fit. In essence, we have a vast collection of knowledge (made up of information and experiences), and to solve a problem, we first need to identify the useful subset of that knowledge.

Think of current large language models (LLMs) as having absorbed the entire knowledge base of human-created artifacts – text, images, code, and even elements of audio and video through transcripts. Because they’re essentially predictive engines trained to forecast the next word or “token,” they exhibit a basic level of reasoning that comes from the statistical structures within the data, rather than deliberate thought. What has been truly remarkable about LLMs is that this extensive “knowledge layer” is really good at exhibiting basic reasoning skills just by statistical prediction. 

Beyond this statistical stage of reasoning, prompting techniques, like assigning a specific role to the LLM, improve reasoning abilities even more. Intuitively speaking, they work because they help the LLM focus on the more relevant parts of its network or data, which in turn enhances the quality of the information it uses. More advanced strategies, such as Chain-of-Thought or Tree-of-Thoughts prompting, mirror human reasoning by guiding the LLM to use a more structured, multi-step approach to traverse its knowledge bank in more efficient ways. One way to think about these strategies is as higher-level approaches that dictate how to proceed. A fitting name for this level might be the Executive Strategy Layer – this is where the planning, exploration, self-checking, and control policies reside, much like the executive network in human brains.

However, it seems current research might be missing another layer: a middle layer of metacognitive primitives. Think of these as simple, reusable patterns of thought that can be called upon and combined to boost reasoning, no matter the topic. You could imagine it this way: while the executive strategy layer helps an AI break down a task into smaller steps, the metacognitive primitive layer makes sure each of those mini-steps is solved in the smartest way possible. This layer might involve asking the AI to find similarities or differences between two ideas, move between different levels of abstraction, connect distant concepts, or even look for counter-examples. These strategies go beyond just statistical prediction and offer new ways of thinking that act as building blocks for more complex reasoning. It’s quite likely that building this layer of thinking will significantly improve what the Executive Strategy Layer can achieve.

To understand what these core metacognitive ideas might look like, it’s helpful to consider how we teach human intelligence. In schools, we don’t just teach facts; we also help students develop ways of thinking that they can use across many different subjects. For instance, Bloom’s revised taxonomy outlines levels of thinking, from simply remembering and understanding, all the way up to analyzing, evaluating, and creating. Similarly, Sternberg’s theory of successful intelligence combines analytical, creative, and practical abilities. Within each of these categories, there are simpler thought patterns. For example, smaller cognitive actions like “compare and contrast,” “change the level of abstraction,” or “find an analogy” play an important role in analytical and creative thinking.

The exact position of these thought patterns in a taxonomy is less important than making sure learners acquire these modes of thinking and can combine them in adaptable ways.

As an example, one primitive that is central to creative thinking is associative thinking — connecting two distant or unrelated concepts. In a study last year, we showed that by simply asking an LLM to incorporate a random concept, we could measurably increase the originality of its outputs across tasks like product design, storytelling, and marketing. In other words, by turning on a single primitive, we can actually change the kinds of ideas the model explores and make it more creative. We can make a similar argument for compare–contrast as a primitive that works across different subjects: by looking at important aspects and finding “surprising similarities or differences,” we might get better, more reasoned responses. As we standardize these kinds of primitives, we can combine them within higher-order strategies to achieve reasoning that is both more reliable and easier to understand.

In summary, giving today’s AI systems a metacognitive-primitives layer—positioned between the knowledge base and the Executive Strategy Layer—might provide a practical way to achieve stronger reasoning. The knowledge layer provides the content; the primitives layer supplies the cognitive moves; and the executive layer plans, sequences, and monitors those moves. This three-part structure mirrors how human expertise develops: it’s not just about knowing more, or only planning better, but about having the right units of thought to analyze, evaluate, and create across various situations. If we give LLMs explicit access to these units, we can expect improvements in their ability to generalize, self-correct, be creative, and be more transparent, moving them from simply predicting text toward truly adaptive intelligence.

Posted on

Can You Teach Dogs To Be Creative? A Pawsitive Experiment

A few years ago we got a puppy and, like everyone else, we started house training him right away. We hung a bell on our patio door, and every time we took him out, we’d help him ring it with his paw. Within a week, he figured out that ringing the bell meant going outside, and he started ringing it himself. And about a month later, he had become a pro at this!

A few months later, a funny thing happened. He likes to sit on the couch in the living room and one day it just so happened that we were sitting on the couch and there wasn’t any space for him. So he came up with a clever idea. He rang the bell, and just as one of us got up to open the door, he ran back and jumped on the couch to claim the open spot. Problem solved!!

We all know dogs and other animals are pretty smart, and this little story shows just how creatively they can solve problems. Being able to use a concept in a new or different way is a key part of creative thinking, and it’s what tests like the Alternate Uses Task (AUT) on the Torrance Test of Creative Thinking try to measure. You know, those questions like “How many ways can you use a paperclip?” that see how well people can think in different ways.

Take our dog, for example. He’s found different ways to use that bell. Early on, he figured out a more obvious extension to the original purpose — he could ring the bell not just for a potty break but anytime he wanted to go out for fresh air or chase after a bunny. Now he also rings it to express his displeasure when we don’t share our snacks. In his case, using the bell to solve a problem that doesn’t involve going outside reflects a higher level of “flexibility” in his alternate uses of the bell. 

As pet owners, we have all seen some creative ways our furry friends solve problems that point to their intelligence. But can they be trained to be creative on demand?

In one study, researchers recruited dog owners to train their dogs to be creative. First, they taught the dogs to associate a specific word “create” with doing something new. The trainers would reward any new behavior the dog did, and then encourage the dog to try even more new things by moving around and using different objects. The only rule was that the dog couldn’t repeat the same trick twice in a session. To figure out how creative the dogs were, the researchers looked at three things: if they repeated tricks, how much energy they used, and how original their tricks were. Basically, they wanted to see if the dogs could come up with lots of different ideas, switch between different types of tricks, and do stuff that was new and unexpected. 

The only required criteria in this experiment was that the dogs don’t repeat the same trick twice. Researchers found that all dogs did more different tricks than repeated ones, which proves that dogs can indeed be trained to be more creative! As a side result, the study also showed that dogs have a good memory –  they can remember what they’ve already done and come up with new ideas. 

The ability to think creatively is a trait we often associate with humans. Yet, as we’ve seen, our canine companions possess a surprising degree of ingenuity that challenges our preconceived notions of animal intelligence. Understanding how dogs learn and apply creative thinking can not only deepen our bond with our furry friends but also provide valuable insights into human cognition and behavior. 

Posted on

Creativity Hack: Combining Unrelated Ideas

One of the most potent ways to find creative ideas is to take two completely unrelated concepts and try to combine them. This ability to associate unrelated ideas is a natural process for our brains but we often underuse this capability in finding novel ideas.  

About The Hack

Associative thinking, the ability to combine unrelated ideas, underlies a lot of innovation we see in the real world. Google search, one of the most well known inventions, is the product of associative thinking. When Sergei Brin and Larry Page were students working on improving search, they hit upon an interesting insight. The problem that they were trying to solve is to point users to high quality web pages that contain information users would find useful But how do you determine which websites are good and which ones are not? Their “aha” moment came when they realized that academic journals have a mechanism to identify high quality papers — the number of times a paper is cited by others. Applying the same concept to web pages, they realized that the more a web page is linked to by others the more authoritative it must be. They used that idea to create their first algorithm to rank web pages and Google was born! 

To use associative thinking in product design, find random objects or concepts and try to connect it to your central problem. For example, suppose you are tasked with making a new kind of mug. You then think of different objects or attributes, not typically associated with a mug, and see if there are ways to combine it. Suppose you picked a ball to combine with a mug. The simplest way to combine would be a ball-shaped mug. But, you could go further and use an attribute of the ball in a more meaningful way. Let’s say you pick “inflatable” as an aspect to incorporate. That leads you to creating an inflatable mug that is easy to pack on trips and provides good thermal insulation thanks to the layer of air in between. 

Summary

Finally, here is a quick summary of the creativity hack and how to use it in product design or with students.

DescriptionTo find a creative idea for product improvement, try to combine a random object or attribute with the product. 
ExampleIn designing a new kind of mug, you combine it with a ball. One attribute of the ball is “inflatable” which leads to the idea of an inflatable mug. The mug is useful because it packs more easily for hiking trips and also provides better thermal insulation due to the layer of air in between.  
Tips Instead of combining objects directly, use an attribute of the random object to combine. That often leads to more interesting and novel ideas  
ExtensionsTo build associative thinking in students, ask them to incorporate other famous characters (fictional or otherwise) into their stories, or do a project that combines their hobbies with a subject they are learning (e.g. music and math)
Creativity Hack: Combining Unrelated Ideas
Posted on

What Neuroscience Tells Us About Learning

Students today spend more time on academic learning than generations before. They cover more ground – learning things like programming or environmental science that their parents didn’t have to fret about – and spend more hours doing homework after school. One study found that in the sixteen-year period from 1981 to 1997, there was a 25% decrease in time spent playing outside and a 145% increase in time spent doing homework. 

As our society advances even more, students will have to cover more and more content, not just during their K1-2 school years but throughout their careers. By some estimates, students growing up today will have to learn entirely new domains and reinvent their careers every few years. Learning is no longer limited to younger ages but is becoming a lifelong journey. 

What does this really imply?

Students have to learn to learn –  acquire knowledge and master concepts faster – without which they will find it harder to stay abreast of new developments coming their way. But it’s not just about superficially memorizing things. Students will have to understand how to apply their newfound knowledge to problem-solving. In other words, learning has to become a more efficient process in terms of speed, depth, and understanding.

Thankfully, advances in neuroscience are giving us clues on how to make learning more efficient. Understanding how the brain processes information can help students take charge of their own learning, not just in their student years but throughout their life.

Neuroscience Of How Our Brain Learns

At a high level, we can view learning as a three-step process. When we encounter any new information, our brain first encodes this information and places it in short-term memory. For example, if you come across a new fact, say learning about a new breed of dog, the information first goes into your short-term memory. The next day, you might recall that your childhood friend had a similar-looking dog, and now you start to remember other details about the dog – how friendly it was, how it played, and so on.

At this stage, your memory is in long-term storage; it continuously consolidates other pieces of information that you already had. Over time you might add more connections to this piece of information, maybe a joke you heard about it, and it starts to get more and more enmeshed with other pieces of memory. 

After a few days, you might forget the name of the breed and try to recall it. You struggle a bit and then remember your friend’s dog, the joke, and other bits of memory that were tied to it. And then the name suddenly comes back to you, and you get a sudden burst of relief! 

A few days later, as you share a story about your childhood friend, her dog and the name of the breed come to your mind effortlessly, and you marvel at how well you remember this now. 

The picture above encapsulates how our memory works. Once we consolidate information into our long-term memory, subsequent retrieval and reconsolidation help to strengthen the memory traces and make it easier to recall information in the future. 

Forgetting Is The Path To Learning

Over the last couple of decades, neuroscientists have discovered interesting things about how our memory works, and counterintuitive as it sounds, forgetting information is an important aspect of remembering! Our brain is constantly pruning information that it thinks it doesn’t need so that it can serve the really important bits of information faster. 

Imagine if your brain stored every little nugget of information that it receives – the color of the shirt a passenger wore in the subway, or the name of the street your friend in another state lives on – it would make it much harder to find the useful information that you really need. So if you don’t need any piece of information, its retrieval strength starts to get weaker. However, when you try to recall something that you have forgotten, i.e., when you have to struggle a bit to remember it, that’s when the brain gets a cue that this particular memory is important and might be needed again.

So, with the process of retrieval, it starts to reconsolidate the information – find newer connections to other traces of memory so the memory is stored more strongly. As a result, this process of forgetting and remembering actually helps you learn better. 

Neuroscience-based memory models give us clues on how to structure our learning for maximum effectiveness. Here are three ways to boost your learning.

Repurpose Failure

When students don’t remember or don’t apply concepts correctly, it’s a sign that the information has been stored weakly in the brain. However, instead of feeling that they are ‘not cut out’ for this kind of work, students need to understand that their failure is simply a sign for their brain to reconfigure and become more efficient. Human brains are designed to learn through mistakes, so it makes sense to reframe forgetting as what it really is – a trigger that tells us that we need to take additional steps to ensure learning is complete. Students should use the opportunity to review concepts again and try to reconcile the mistakes so their understanding of the subject increases.  

Adopt Active Learning Strategies & Neuroscience

Adding some challenge to the learning process that taps our brain’s natural mechanisms to process, store and understand information can significantly boost learning. Such challenges are ‘desirable difficulties’ because they make learning more efficient. Here are a few strategies that students and teachers can adopt: 

  • Retrieval Practice: When learning new information, periodically quiz yourself about the central ideas and new terms encountered without looking at the text. This forces your brain to fetch the answers from long-term memory, and repeated retrieval is going to strengthen your memory.
  • Spaced Learning: To add more desirable difficulty to learning, practice retrievals after a period of time. When you start forgetting, you exert more effort in trying to remember, which then cues the brain to store the information more deeply. The gap between learning and retrieving can be anything from a day to a week – the key is that the gap should allow for some forgetting to happen.
  • Interleaving: Instead of waiting to thoroughly master one concept before moving on to the next, try mixing up different kinds of problems or concepts once you feel you have gained sufficient understanding in one. Not only does this make good use of spacing, but it also allows you to spot connections or differences between different kinds of problems. 

Research studies show that such strategies can be very effective in the classroom. In one study, students who practiced math problems in three sessions spaced apart by a week performed twice as well on the final test compared to students who did all the practice problems in one session.  In another study, students performed significantly better on their science exam when a practice quiz one month before the exam interleaved concepts on the quiz. 

Associative Learning & Neuroscience

Another useful strategy in learning is to connect the information you are learning to other pieces of knowledge you already possess. If retrieval practice creates deep roots, then associative learning creates more branches that help anchor the information better. To build associative learning

  • Find an analogy: Ask yourself if the new concept is similar to any other piece of information that you already possess. As an example, you might make a connection between gravity and magnetism as both involve a force that they can’t see and attract objects. 
  • Find a personal connection: In some cases, your personal experience can be helpful in finding connections about what you are learning. For example, while learning about the ice age, you might remember an earlier trip to Grand Coulee, where they saw how the Missoula Floods carved out a massive canyon in a very short time. The scale and impact of the event will give you an enhanced perspective of the topic and deepen your understanding. 

Conclusion

By understanding the neuroscience behind learning, students can take charge of their own learning. The key to efficient learning is to add and embrace the right kind of challenges that push our brains to reconfigure themselves. Unless students lack relevant background or specific skills to make sense of the concept in front of them, such challenges should be welcomed instead of dread. 

With a deeper understanding of the learning process, students can try different approaches and customize them to their needs. As an example, for some students, one day of spacing might be enough, whereas, for others, it might be a week. For the latter set, practicing a skill every day might not be as effective because they haven’t forgotten enough for reconsolidation to take place. With some trial and error, students can identify strategies that work best for them and become smart learners. 

This article first appeared on edCircuit

Posted on

Cognitive Underpinnings of Creative Thinking

50,000 years ago humans shared the land with other hominins like the Neanderthals and the Denisovans. But somehow, over the course of the next 30,000 years, every other hominin species went extinct while the modern day humans saw huge growth and advancement.

Some people point to the development of language and tools that gave us a Darwinian edge. But evidence of language and tools, some of which were fairly sophisticated, have been found in Neanderthals and the other hominins. So what made us special?

According to Thomas Suddendorf, professor and author, what set us apart was not language or tools, rudimentary forms of which exist in other animals, but our ability to do open-ended imagination and make connections between different concepts. This enabled us to do mental “time travel”, going back in time and in the future, and allowed us to foresee and plan for our survival. In addition, making connections allowed us to find novel and interesting solutions to problems that we faced.

From a cognitive perspective, our brain allows us to voluntarily think of a concept which then triggers another concept, which in turn triggers the next one and so on, leading to a stream of thought, something that likely doesn’t exist in other animals. As Professor Liane Gabora explains, “With this ‘self-triggered recall and rehearsal loop’ we could now activate and re-activate visions and dreams, such that with each successive conception of them they were looked at from a different angle, embedded a little more firmly in the constraints of reality as we know it, and potentially turned into a form in which they could be realized.”

This cognitive ability is the direct result of how our brain stores information associatively, and is the reason why humans are able to come up with novel and creative ideas.

Think about how a computer stores information. To store the word “apple” in computer memory, each word is broken down to its letters and each letter in turn is converted to its binary code and stored. For example, the binary code for “a” is “01100001”, for “p” is “01110000” and so on. That’s not how human brains store information.

Human brains store each concept as a whole, connected to other concepts. So the word “apple” is stored as a concept by itself and is linked to other concepts with different kinds of links. So “apple” might be connected to “fruit” by a thing to category link, to “red” by a thing to property link, or “rash” by a cause and effect link if someone is allergic to apples. These links have different strengths in the brain so the dominant link for one person might be apple to red, but apple to fruit for someone else.

When you consciously think of an idea, your brain automatically activates some of the connecting ideas and brings them into your consciousness, leading to a stream of thought.

This also explains why it is sometimes hard to think of new ideas or solutions. As we think about a problem, some of these links get reinforced and strengthened making it hard to change perspective or think in a different direction. In other words, “These same pathways, however, also become the mental ruts that make it difficult to reorganize the information mentally so as to see it from a different perspective.”

The associative nature of the brain also comes into play when it encounters ideas that are not related. To see how this works, look at the two words below:

                          Bananas                            Vomit

If you are like most people, the moment you read the two words, your brain automatically tried to connect the unrelated words with a causal connection, forming a scenario where eating bananas led to vomiting, leaving you with a somewhat unpleasant feeling. 

You didn’t have to consciously think of this, your brain did the work of finding the best possible connection between the two words.

These two aspects of the associative nature of our brain – activating connected ideas and finding connections between random ideas – are what make it possible for us to think creatively. In fact, most creative thinking techniques rely on these two underlying mechanisms in one form or the other to generate novel ideas.

  1. Traversing Connected Ideas: Techniques like “Slice and Dice” and “Cherry Split” in Thinkertoys or segmentation in TRIZ, work by forcing the brain to traverse different paths in the associative network by explicitly listing out the triggers.
  2. Adding a Random Component: By simply introducing a random element into the mix, the brain automatically tries to find the best way to incorporate the random element into the solution. Techniques like the “Brute Think” and “Hall of Fame” in Thinkertoys are an example of such an approach.

What made us come so far is likely because of our unique cognitive strengths – how we store and process information in our brains, combine different ideas and run mental simulations. These strengths allowed us to solve problems, make inventions and build on each others ideas, and they just might turn out to be key for our future as well.