In ‘learning trap’ experiment, adults leap to conclusions while children explore
wsj.com
wsj.com
For me, this sentence at the end breaks the whole argument down - so the adult strategy was indeed more effective at the the thing being optimized for!
The question then is - could they set this up such that kids actually perform better than adults?
Skimming the study, it seems like the kids' ability to learn the correct rule by "exploring" was cancelled out in some of their tests by their ability to completely forget it when new [irrelevant] information was introduced in the form making the objects a different colour
Also kids aren't guaranteed to catch up adult scores if they're more likely to lose motivation. We're talking about a study that noted it excluded a participant who "refused to continue after seeing the sad face" :D
Those kids would of course grow up to become adults with all the knowledge gained from the exploration but without any of the costs of exploring further, that is how the kids strategy becomes the winning one.
The people who try to figure out the exact rules of everything they do tend to be less effective than people who figure out a strategy that works and simply execute.
There is catch - adding money into equation can trigger "avoid risk" behaviours.
If the game was 10x longer, adults would behave differently.
Large codebases can do many amazing things quickly, but they tend to have a lot of technical debt making other things very hard to do. Kids are small codebases, they lack all those abstractions for solving problems quickly but they also lack all the technical debt, allowing them to quickly change and adapt their thinking.
Why are adults better at playing games? Because good board games are designed to align with abstractions adults are likely to have and don't hit the parts with lots of technical debts. It is very easy to design games that goes against the abstractions adults have learned, but those games will be horrible to play so you wont see them.
Edit: Also board games aren't meant to be mastered, you just need people to learn the simple basics. If you add the requirement to master the board game, then we can look at chess and see that people who started learning as a kid becomes way stronger than people who started learning as adults. Adults can learn to play chess much quicker than children, but an adult wont master the game.
And I am not even talking about "mastered" game. Most kids are not playing chess. It is not just difference in experience of board games.
Small kids don't figure out completely new situation faster then adults. They just dont. The quickness of understanding comes in much much later.
Kids older then toddlers care great deal about winning. It is not true that they don't understand why they should care. If you play boardgames with kids, they understand concept of board games in general. Babies don't, toddlers don't, but it is also impossible to play with them.
Framing difference between kids and adults as just lack of experience and pre-existing knowledge is, imo, more of wishful thinking or theory then anything actually observable.
So to me it is obvious that at least some of the difference is that adults have a lot of knowledge to apply while kids don't. In scenarios where the adult lacks a lot of relevant knowledge the kid performs way better.
Small kids, surprisingly, completely sux at understanding computers too. Not software engineer level. They sux at basic user stuff. Small kids don't learn UI fast, it takes them a lot of time. Once they get it, they think how fast they were getting it. But in reality, they sucked for a long time.
Small kids don't navigate UI fast, unless someone in their life patiently shown them and made them feel that way.
---------
And also, adulthood starts somewhere between 18-21. Those people in fact learn computers, UI and anything new quite fast. You have to go to retirement age to get that "super difficult to learn" thing.
I agree with this idea in general, but I think it’s prudent sometimes to ask if one has the best strategy or just a good enough strategy.
Folks who rarely or never reflect on this often end up at a (relatively low) local maximum.
In this experiment, it would probably be reflected in the results if they had a larger sample and made a combo with the black dots block award more stars. Many adults probably would not discover this, while many/most kids probably would.
Lucky for you, this set is empty.
> ... tend to be less effective than...
At what? If you're an AI, then you might be overfitting on your training data.
Snide jabs aside. I usually find that "good enough, just execute" has its own failure modes, different than those of "keep digging, understand deeply". Better yet, there are no "failure modes" or "successes" at all unless you have a goal to evaluate against.
> At what?
Accumulating government issued currencies? I do have an impression that technical literacy rarely pays off. It is not like human history is full of trillionaire theoretical physicists and practical machinists.
I mean, not just in terms of making money, but inventing new technologies, saving lives or any tangible measure of success. At some point, you need to play the game.
Edit: Also all of Microsoft, Apple, Facebook, Google and Amazon were started by people in their 20's. That is a pretty big coincidence, especially given that younger people have fewer connections and less money than older ones and therefore should in theory have a harder time starting a company.
And that's the problem. In the adult world, we too often optimise for short-term gain instead of learning. This comes at the cost of long-term benefits. Focusing on learning costs you in the short term, but can pay immense dividends in the long run.
I see no such an argument in the article. It is modern psychology, it tries to avoid judgements like "good" or "bad", it doesn't try to prove that kids are dumb (or geniuses), it tries to learn how it is. The core principle can be expressed roughly "people acts the best possible way until proven otherwise". It is like an axiom, though it lacks a mathematical rigor. So this article points to an another direction: while it is beneficial for children to sacrifice material rewards in exchange for better learning, for adults it is more beneficial to concentrate on the rewards. This difference causes them to employ different strategies shown in the experiment. Though this last sentence is not proven by the experiment, there may be other causes, but Gopnik did a lot of research pointing to the same direction: babies and kids are optimized for the learning.
From this follows a practically helpful advice: if you cannot understand your kid's behavior, try to find what she/he is trying to learn by doing what is she/he doing. For example, you told her not to open a cupboard, you know she learned the rule "do not open the cupboard", but now she opens it while looking you in the eye? ("terrible twos" may do this) She is trying to learn how would you react to her misbehavior, and what the best way to deal with it. So you'd better get a grip on yourself, turn on your inner teacher and show her exemplary reaction to misbehavior, which she may adopt later herself against other people breaking rules. She needs to explore the role of a rule breaker to learn how to deal with the rule breaking from both sides, so let her play that role by playing the role of a rule enforcer.
> The question then is - could they set this up such that kids actually perform better than adults?
Probably they can, but what the point of such an experiment? To prove that a strategy stressing learning can give more material rewards when compared to a strategy maximizing material rewards? It is not a psychology per se, it is more like a research of different strategies of a decision making, so it would be better if such a research was done by an AI-researcher.
... in a way adults are not, I suppose. Begs the question, when do you go from being a kid to being an adult and can we postpone this transition, and should we?
I recently realized that I had failed to uphold the promise I made to teenage me to never become adult. I consider it a failure because of my view that most adults transition way, way too early and therefor have minds that are too narrow for comfort.
Unfortunately, transitioning back to "kid state of mind" seems impossible.
The difference is I’ve learned how to handle various adult tasks that did not interest me before. Bureaucracy, taxes, investment, insurances etc.
But I also still watch the shit out of TV, love video games and other various things I loved as a kid. I’ve just learned to be passionate about even more things than I was, and love them even more.
There have been some changes in my tastes but they’ve been fairly rare by comparison. Most changes came from learning and maturing.
For what it’s worth though I don’t think most people are like this. All in all, I’ve found this to be the minority case.
I don't think it's impossible, though it certainly requires practice and deliberate cultivation in most adults.
Besides the advantages to learning new things, I think it's also a beautiful way to look at the world: to see everything as if for the first time, your mind filled with wonder, curiosity, eagerness to explore.
Practically, I have found this attitude to also lead to more rewarding social interactions. When someone explains something to me that (I think) I already know, or have some experience with, my mind doesn't jump to "heard that before" and shut off, but I can remain engaged and interested in the conversation, letting my prior knowledge just be there in the background.
In addition, we live in a more rapidly changing world than our evolutionary ancestors did, which makes necessary to continue learning throughout our adult lives. e.g. I'll probably have to learn new JavaScript frameworks until the day I retire
Did you get as far as the title?
"Why Children Learn Better Than Adults"
Given the study didn't choose to do that, it's a dubious claim imo that the kids figured anything out at all. Likely they were just putting a bunch of blocks in the machine ad hoc.
You miss confounding factor, children have abysmal cognitive performance compared to adults.
So you compare 1 heuristic algorithm running on weak CPU and RAM-starved, with another heuristic algorithm running on many times stronger CPU with up to 3x times more RAM (working memory).
I wonder if changing the overt objective to ‘learn the rule’ would result in the adults outperforming the children?
If you’re told to maximize the number of points, then learning the point value distribution of the blocks is a secondary objective, and we could also wave the conclusion of this experiment as “children don’t follow arbitrarily-valued objectives as well as adults.” Or we can understand this result as an expression of an adult’s developed executive function suppressing the desire to learn for the sake of reaching an objective.
I’m reading the researchers’ interpretation of this experiment as “children have a higher desire to learn” when executive function could explain that while also not diminishing the desire of adults to learn.
I am reminded of a/b testing in websites or emails, for example.
If the goal is to maximize clicks, you may never "learn the rule," as it were, and instead lean on familiar and previously-tested concepts, getting stuck in a local maximum.
On the other hand, if in an explicit mind shift you say "let's try everything to see what happens," you may "lose" far more often, but gain a comprehensive map of the landscape in return, which can help you in future efforts to break into new heights of engagement.
Adults trained with external reinforcements (performance bonuses such as monetary rewards, gold stars, grades, rankings, etc.) will perform better during training compared to ones that receive no performance bonuses.
However, after the training is finished, the ones that received no bonuses will generally be more skilled than the ones that did, because without the allure of the bonus they went for internal motivation instead, and challenged themselves with tough cases which allowed them to get better.
So the next time you're in an environment with a performance bonus component, ask yourself whether you really want to optimise for the short term over the longer term. If you prefer long term learning, maybe change your environment.
I also remember reading a research group managed to train adults to learn like kids again. Can't seem to find it now, but maybe it's not set in stone and we could benefit from both approaches where needed in life.
Noel - what a sweet sweet handle. TIL
Exactly. EG: In 10,000BC you hear a cat hissing in a cave. You figure you shouldn't go into the cave. You wouldn't try to second guess yourself and "explore".
Figuring out "which blocks make the machine light up" is not the same as "earning as many stars as you can", apparently.
If the original research introduced this ambiguity, then I don't think the conclusions of a "learning trap" are supported. Compare adults' and children's performances given a single goal, either "figure out the rule" or "earn more stars", but not both.
The children are likely to have misunderstood the directive to mean that the way to earn the most stars is by figuring out the rule, while adults properly understood the tradeoffs to earn the most stars
Adults usually having more and varied responsibilities makes it better for them to jump to conclusions, thereby leaving time and energy for the other stuff. Children have less responsibility, so they can afford the curiosity that may find them stars the adults missed.
Indeed, one major adult responsibility is keeping children safe, which often means curtailing their excessive curiosity.
Still true in alot of the world
As we experience more of the world over time, the model of the world we build internally gets more and more elaborate. Eventually, we experience enough we can start noticing metapatterns. Something like, "if something is bad, it usually doesn't get better" is an example of a metapattern which seems like it could be in play in this study.
Of course, we lose out on some experiences, because that heuristic is quite prone to false negatives. I think that's basically what the experiment in the article illustrates, this "lossy filter" in our brain's future-planner.
(I am making no claims about the physical structure of the brain.)
“Indeed, [due to an immature prefrontal cortex] adolescents are risk-taking and novelty-seeking individuals and they are more likely to weigh positive experiences more heavily and negative experiences less so than adults. This behavioral bias can lead to engagement in risky activities like reckless driving, unprotected sex, and drug abuse.”
There is a great deal of research on this. Its implications for society are profound. Robert Sapolsky is a prominent public face of the latest science on the development of the human brain and the implications for society. He can be found all over YouTube, and I can personally attest his latest book, “Behave,” is illuminating.
But it's pretty unhelpful to the children if you give up on understanding their problems like that and just label them bad.
What are they going to better after you call them an asshole.
Kids in general can't be expected to act like adults because their brains are still developing. They also lack experience to inform their actions. Some seem like assholes because of these factors.
It's just important however to realize that some kids are assholes. No matter how you humor them or mollify them, they'll still be assholes. After trying to humor them and understand them if they turn out to be assholes you can just deal with them (or don't) like you would any other adult asshole.
Wish they said how many children figured out the rule.
However, I agree this involves some fairly broad principles. I think you could even say this is kinda how Bayesian inference works too. If you use a [strongly informative] prior, on average you get a better estimate, but you're taking the gamble that your prior applies.
The better thing could be the maximum number of stars aren't revealed and then it could be explore vs exploit scenario.