Note: Two sections of the book go through 1) the early development of the brain just before and after birth and 2) learning languages. I didn’t take extensive notes for those two sections, so I omitted them here.
What is learning
Learning exists to enable an organism to adapt itself to an unpredictable environment in which it’s going to live. To learn is to form an internal model of the external world.
The four pillars of learning in humans:
- Attention
- Active engagement
- Error feedback
- Consolidation
Pillar 1: Attention
Attention is the filtering of information to reduce sensory overload. Selection and amplification of relevant information is a requirement. It can also lead us down the wrong path when the system makes errors.
Filtering is achieved by highlighting one area and suppressing everything else, even to the point of blindness (the famous selection attention experiment). The problem is that we don’t notice what we’re missing, so it can have fatal consequences like driving while talking on a phone and not seeing a person walking.
Executive function is critical to controlling attention. An example with babies (<1y): when an object is consistently hidden in place A and then in place B in the last instance, babies look for it in place A. Piaget made these experiments and observations, but his conclusions about babies lacking object permanence were incorrect; babies understand and see where the object is (place B) but can’t resist the habitual response they learned by observing place A many times because of weak executive function.
A similar thing happens with kids and adults if you have the same number of items in a set, but you spread one set or if the objects are bigger—visual size competes with abstract concepts like numbers.
Pillar 2: Active engagement
We must engage with the learning material or environment to learn, reflect and reframe, and look for abstractions. Passively absorbing can’t happen or happens exceptionally slowly.
There were theories to let kids explore fully on their own (full discovery learning), but it didn’t work out at all – none of the experiments showed that it’s better than directed and guided teaching. On the contrary, directed teaching was vastly superior for kids in terms of understanding and abstracting concepts.
There is no scientific basis for learning styles (visual, auditory, kinaesthetic). However, some learning modes are better suited for specific lectures or areas, but all learners benefit from that better mode in the same way.
Curiosity. We get a dopamine hit when we learn new information. The evolutionary background is probably that curious animals go out and learn more about their environment (eg. around their den) because it might increase their chances of survival.
Bell curve of curiosity—we’re attracted to adjacent things that can be learned quickly and aren’t too complex and abstract.
Epistemic curiosity is the irresistible desire to know.
Pillar 3: Error feedback
Learning only happens if there is a gap between what the brain predicts and perceives. Organisms learn when events violate their expectations. No surprise, no learning.
Aside: This reminds me of advice by professor Tadashi Tokieda to always guess before finding out a result because “if you guess wrong, you are […] shocked, and that engages your thinking. You can learn what happened and then it makes you a little smarter next time.”
Error feedback is not (shouldn’t be) punishment.
Pillar 4: Consolidation
From executive centers into long term memory for efficiency. Sleep is critical for consolidation. Sleeping prevents forgetting.
Why our brains learn better than current machines
- Humans can abstract and question our beliefs. Machines are not there yet; they’re still superficial and only recognize patterns.
- Machines are data hungry and humans are data efficient.
- Social learning.
- One trial learning and integrating it into existing knowledge.
What humans consistently do and what our brains are great at is searching for high-level rules abstracted from a specific situation and subsequently tested on new situations.
Somewhat technical definitions of learning:
- Setting values on parameters.
- Minimizing errors. Randomization to avoid the local minimum.
- Maximizing the reward function. AI playing games and adversarial networks.
- Restricting search space. Convolutional neural networks. In nature, learning from zero doesn’t exist. It’s much more effective to start from somewhere and have assumptions, prior experience, or constraints. Babies come with knowledge about entities that move and can’t pass through each other. That’s true wherever a person is born and will live, so it’s written in our genes. (Not all scientists agree that we come with much pre-existing knowledge.)
Babies’ invisible knowledge
- The object concept is in-born.
- Numbers, statistics, logic.
- Knowledge of animals and people (they can move and have the motivation to move around and overcome obstacles).
- Personalities of others (dislike for people who intentionally do harm rather than by accident).
- Face perception.
- Spoken language is very easy to acquire (hearing starts in utero, so there is a preference for native language).
Nurture’s share
Hormones:
- acetylcholine (focus): amplifying the signal that comes from outside, a spotlight
- dopamine (reward and motivation)
When there is a traumatic event, if we interfere with chemicals to prevent the correct forming of synapses from capturing that event, the mouse won’t remember it (won’t learn). That’s why we think there’s a causal effect between these chemicals and learning.
Memory is distributed across the brain and reconstructed every time we remember.
Recycle your brain
Recycling = reusing capabilities and structures of the brain for new uses. Not full plasticity in which the brain would fully transform itself into something else. It’s the same regions, just reused.
Approximate quantities in primates, humans use the same regions for more complex uses like mathematics.
Sighted and blind mathematicians are using the same part of the brain when doing mathematics. In addition, blind mathematicians use a part of their occipital lobe (for vision) that has adapted to process more abstract tasks like mental calculations.
We can quickly compare numbers that are far apart and make more mistakes when they’re closer. We also struggle with very large numbers, and our tolerances grow as numbers do. That’s why we are OK with rounding up or down several thousand when buying an apartment, but we look at pennies when buying coffee. It’s all based on our initial brain wiring and biases come from that.
Reading uses visual and speaking circuits of the brain. If people start learning to read early, the reading circuits strengthen a part in the brain’s left hemisphere that is dedicated to recognizing faces and objects; those latter functions continue developing in another part in the right hemisphere.