Brain’s Hidden Learning Limits

this video was brought to you by Squarespace the human brain is a learning Powerhouse from mastering foreign languages to performing intricate physical movements there seems to be almost no limit what we can learn yet anyone who has struggled to master a new skill knows that our brains are not infinitely flexible some abilities come naturally While others feel impossible to grasp no matter how much we practice but does it just come down to not trying hard enough or could there be something deeper fundamental constraints wired into our neural architecture that make certain patterns of brain activity

and thus the resulting behaviors physically impossible to achieve in this video we will explore a groundbreaking study published in nature neuros science this January that revealed something remarkable there are indeed Hardware limitations built into our neural circuitry that determine what we can and cannot learn no matter how hard we try at any given moment each nerve cell in your brain is either silent or firing an electrical impulse together creating intricate patterns of activity across millions of neurons these patterns unfold in time like a carefully orchestrated Symphony driving our behavior when you reach for a cup

of coffee for example your neurons follow a specific sequence of planning the movement trajectory and executing it controlling your muscles with precise timing the Dynamics of these patterns of neural activity flows through your brain like water through a landscape and just like Rivers tend to follow established channels neural activity might have preferred Pathways shaped by our brain's physical wiring this raises a fundamental question how flexible are these patterns when learning a new skill can our brain generate any sequence of neural activity it needs to or are some sequences much harder than others or even outright

impossible because of how our neurons are connected and they're underlying by a physics but how can we test this idea scientifically when you perform any action millions of neurons are firing in complex patterns all across your brain even if we could hypothetically record the activity of all of them we would still run into a problem that lies at the heart of modern Neuroscience you see in any scientific field to understand the rules of a system you need to systematically test different conditions formulate hypotheses and validate them for instance if you are studying a new chemical

reaction you might need to separately vary the temperature and concentrations of individual compounds to see how they affect the reaction rate but when it comes to neural activity we Face a unique challenge to understand what patterns are hard to learn we want to systematically test different neural sequences maybe start with simple patterns then try increasingly complex ones reverse them shuffle them speed them up or slow them down the point is that we want to carefully control which neural sequences our subject is trying to learn to test specific hypothesis about what makes some patterns harder than

others but there is a fundamental problem while we can't do some kind of systematic testing at the behavioral level like asking the subject to draw shapes of increasing complexity or learn to press the buttons in a specific sequence we have no way of knowing what neural pattern each Behavior will result in for example suppose we wanted to know whether the motor cortex The Movement Center of the brain can generate newal trajectories in reverse order kind of like rewinding a tape we might think that asking someone to do the reaching movement in Reverse would work but

there is a catch When You Reach forward your triceps push the arm out but during the reverse movement it's the biceps that are doing the work these are different muscle groups with different neural trajectories not simply the same pattern being played in Reverse in other words without Direct Control and direct access to brain activity we can't study and test hypothesis about what makes certain neural patterns difficult so it seems like Neuroscience might be hit in a limit but turns out there is a very clever solution instead of trying to design behaviors that might or might

not result in certain brain activity patterns we're interested in what if we make specific neural sequences themselves the very goal of learning here is the key idea what if we could show people their own neural activity in real time now this isn't as far-fetched as it sounds think about controlling your heart rate or breathing normally these processes happen automatically with without conscious control but when you hear or see your heart rate on a monitor your brain gains a remarkable ability to control it voluntarily this is an example of biof feedback a phenomenon When You observe

some kind of signal from your body in real time your brain learns how to influence it and the same principle could work for the neural activity itself imagine you could see the activity of a 100 neurons from your brain on the screen in front of you like a 100 tiny lights flickering on and off now suppose you had some kind of Target pattern a specific sequence of Lights you need it to match with your own brain even though you have never consciously controlled individual neurons before with enough practice and feedback you might learn to generate

patterns closer and closer to the Target the key Insight here is that our brains are constantly learning through trial and error and with this kind of very direct feedback it can quickly identify thoughts and mental strategies that work developing an intuitive feel for how to control the lights on the screen however watching a 100 flickering lights is an overwhelming amount of information to keep track of and since we can only start individual neurons in monkeys rather than humans as we'll see shortly asking them to monitor all these lights is virtually impossible what if we could

simplify it what if we could reduce the complexity of those 100 lights into something more directly interpretable and intuitive for example a cursor moving on the screen this is exactly how brain computer interfaces or pcis work the computer takes in real time activations of hundreds of neurons and transforms them through mathematical mapping into a pair of two numbers X and Y that drive the cursor position when you think of something your neural activity changes and the cursor moves to a new location through trial and error and the direct feedback your brain adapts and learns how

to control its own neural activity to align with this mapping in order to achieve the desired Movement by carefully designing this mapping we can create tasks that require learning to generate specific neural sequences but how exactly would such a task look like and what can we probe with it let's dive deeper into the paper speaking of fundamental constraints unlike our neural circuits what if you could build something without limitations this is where our today's sponsor Squarespace comes in Squarespace is an all-in-one platform that transforms the complexity of website Creation in management into something anyone can

Master while starting a website from scratch can be overwhelming to say the least squarespace's new design intelligence feature changes everything this powerful AI toolkit creates a custom website based on your business and branding complete with tailored content and images giving you the perfect Foundation to build upon every element can be customized using their intuitive drag and drop interface from text content and styling to Dynamic animations that bring your story to life Squarespace is your all-in-one destination for everything online from building beautiful websites and launching newsletters to setting up online stores and processing payments everything integrates

LLY into one unified system experience it yourself with a free trial at squarespace.com and when you're ready to launch head to squarespace.com AUM to save 10% off your first purchase of a website or domain imagine you're teaching someone to play a video game but with a bizarre twist instead of using a joystick they have to control the game using their thoughts alone this is pretty much what researchers did with monkeys in their groundbreaking experiment monkeys sat in front of a screen showing two Targets and a cursor that they needed to move between the targets to

obtain rewards tiny electrodes in their morto cortex were recording the activity of around 90 neurons and transforming it into the position of the cursor but what exactly is this transformation how do you turn the complex Symphony of 9 neurons into something as simple as moving the cursor left or right to answer this we need to understand how to describe the activity of neural populations at any given moment each neuron's activity can be described by a single number how many electrical impulses it fires per second with 90 neurons we get 90 numbers which together Define a

point in 90 dimensional space this is similar to how we can describe the location of an object in our familiar 3D World using just three numbers X Y and Z coordinates but for the neural case we will have 90 coordinates as the neural activity changes over time it traces out a trajectory in this High dimensional space now to turn these trajectories unfolding in the neural activity space into two-dimensional cursor movement researchers used what's called a linear projection think about how a three-dimensional object casts different two-dimensional shadow depending on the angle of the light source similarly

while we can't visualize the full 9-dimensional space we can mathematically project it onto different two-dimensional viewing angles with each projection given us a different perspective on the neural activity the experimental began by finding an intuitive mapping during a calibration session they had the monkeys watch a cursor movement on its own and recorded their neural activity this revealed a specific projection so-called movement in tension view where the neural trajectories looked remarkably similar to the cursor's paths on the screen when the researchers turned on the brain computer interface with this mapping the monkeys quickly learned to control

the cursor moving it smoothly between the targets at first glance the overlapping cursor trajectories for leftward and right rightward movements might suggest that neural activity is indeed quite flexible after all if we see the same path being traced in opposite directions maybe we could just play the same neural tape in any direction however remember that we're only seeing one particular two-dimensional view of the full 9-dimensional activity which doesn't convey the full picture for example imagine looking at a DNA double helix from above it appears as a simple Circle but when viewed from the side you

see its complex spiral structure similarly looking at the neural activity from a different angle might reveal hidden structure indeed the researchers found another projection the separation maximizing view where the leftward and rightward trajectories appeared completely distinct curving in different directions this revealed something profound these movements were not mirror images of each other at the neural level at all instead of the same neural path being traversed in reverse the brain was using an entirely different set of patterns for left board versus right Bo movements now what would happen if monkeys could see this new separation maximizing

view instead when the group switched the interface to use this mapping the animals initially began to see their cursor moving in curved paths rather than straight lines you might expect them to use this visual feedback to modify what commands they send to the interface in order to straighten out the paths after all when reaching for an object we naturally move in straight lines and automatically correct any deviations surprisingly though the monkeys kept moving the cursor in curved paths showing no signs of trying to straighten them this suggests that these neural traj rectories might be outside

their conscious control constrained by the underlying brain circuitry but an alternative explanation might be that perhaps the monkeys simply were not motivated enough to change their activity after all they were still getting rewards despite taking longer curved paths to test this researchers tried a modification of the task with a specific constraint the cursor had to stay within a narrow Corridor between the targets the only way to achieve the goal was to make the cursor move in a way that looks like a Time reversed version of one of the previous trajectories for instance we know that

when moving from Red Target to blue one monkeyy neurons fired in a specific sequence now to move from Blue back to Red while staying in the corridor the brain needed to generate these same patterns but in reverse order even with this strong incentive the promise of reward and the clear visual indication of the corridor must monkeys could not reverse the natural flow of their neural activity they consistently failed at the task suggesting something profound there are fundamental constraints on how neural activity can flow through the brain constraints that even strong strong motivation and practice cannot

overcome what we have seen today reveals something really profound about how our brains work just like a river has channels that water naturally follows physical connections between our neurons create per flows of neural activity while we can learn new skills and adapt our behavior in remarkable ways there are still some fundamental constraints on neural dynamics that we cannot overwrite because they are built into the very architecture of our neural circuits so when we struggle to master new skills it may be not because of lack of motivation or practice but because they require playing out specific

sequences of neural patterns that go against our brain's intrinsic dynamic on the flip side skills that feel natural might be those that work with those intrinsic Dynamics rather than against them but one thing is clear our brains are not infinitely flexible computers that can be programmed to do anything they are biological systems with their own sets of rules and preferences and understanding those constraints might be the key to understanding why we learn and think the way we do if you enjoyed the video share it with your friends press the like button and consider subscribing to

the channel if you haven't already stay tuned for more interes in neuroscience and machine learning topics coming up