So there's a couple of things going on here. There basic biology, which is that we've evolved to have a really long childhood. We're 24 before myelination has mostly taken over our brain, greatly reducing it's plasticity. You can think of myelination as the brain spending energy to cement synaptic pathways that have so far proven to be useful. Now, new synaptic connections need to compete with existing connections, which I suspect actually takes more energy to form, and thus feels more effortful when learning.
Aside from biology, I suspect we have a hard time learning new things as we get older for similar reasons that other computer learning gets harder with age. At some point each new piece of data represents a smaller percentage of the total data gathered. Which is to say, that if you're trying to generalize something each new data point can only move the average/median by a smaller and smaller amount. As such you need to spend more and more time feeding in variants of this new data to have any hope of it having an effect on your model. This can be seen most painfully, if you take something like a neural net and try to train it to do something other than what it was initially trained to do. You're not going to find a chess playing NN, that will be able to then learn to play go any time soon. And if it did learn to play go to any decent degree, it would probably no longer be able to play chess nearly as well.
But, aside from all that, that doesn't mean you can't learn as you get older. Brains are far more plastic, even into old age than we once believed. "You can't teach an old dog new tricks" has been empirically shown to be false, time and time again, especially as we get more and more FMRI data from older people.
I will say, that what others have talked about in this thread, is that there are obstacles other than biology that get in our way for learning. Sleep is a big one. Sleep is key for long term memory formation. If you're not getting enough full REM cycles every day, learning is going to suffer greatly. Also things like adult obligations can really get in the way. There seems to be some growing evidence that people's achievements severely dip in their 30's and 40's, which coincides with raising kids, or caring for aging parents. But mostly, in my experience, seems to be sort of a peak for social obligations that can't easily be blown off. There are times when I'd like to stay home on a sunday night, and make time and space to do something new and difficult, but instead have to go to the inlaw's for dinner to maintain that relationship. And as cliche as that sounds, I like my inlaws, so I make that a priority. Eventually, people die, or move away, and maybe you can work fewer hours, and learning can get easier again, largely because you have more large uninterrupted chunks of time. I hear many anecdotes that musical instruments, of all things, are something that retired people get really good at, relatively quickly, because of the time they have to practice.!
Comments
So there's a couple of things going on here. There basic biology, which is that we've evolved to have a really long childhood. We're 24 before myelination has mostly taken over our brain, greatly reducing it's plasticity. You can think of myelination as the brain spending energy to cement synaptic pathways that have so far proven to be useful. Now, new synaptic connections need to compete with existing connections, which I suspect actually takes more energy to form, and thus feels more effortful when learning.
Aside from biology, I suspect we have a hard time learning new things as we get older for similar reasons that other computer learning gets harder with age. At some point each new piece of data represents a smaller percentage of the total data gathered. Which is to say, that if you're trying to generalize something each new data point can only move the average/median by a smaller and smaller amount. As such you need to spend more and more time feeding in variants of this new data to have any hope of it having an effect on your model. This can be seen most painfully, if you take something like a neural net and try to train it to do something other than what it was initially trained to do. You're not going to find a chess playing NN, that will be able to then learn to play go any time soon. And if it did learn to play go to any decent degree, it would probably no longer be able to play chess nearly as well.
But, aside from all that, that doesn't mean you can't learn as you get older. Brains are far more plastic, even into old age than we once believed. "You can't teach an old dog new tricks" has been empirically shown to be false, time and time again, especially as we get more and more FMRI data from older people.
I will say, that what others have talked about in this thread, is that there are obstacles other than biology that get in our way for learning. Sleep is a big one. Sleep is key for long term memory formation. If you're not getting enough full REM cycles every day, learning is going to suffer greatly. Also things like adult obligations can really get in the way. There seems to be some growing evidence that people's achievements severely dip in their 30's and 40's, which coincides with raising kids, or caring for aging parents. But mostly, in my experience, seems to be sort of a peak for social obligations that can't easily be blown off. There are times when I'd like to stay home on a sunday night, and make time and space to do something new and difficult, but instead have to go to the inlaw's for dinner to maintain that relationship. And as cliche as that sounds, I like my inlaws, so I make that a priority. Eventually, people die, or move away, and maybe you can work fewer hours, and learning can get easier again, largely because you have more large uninterrupted chunks of time. I hear many anecdotes that musical instruments, of all things, are something that retired people get really good at, relatively quickly, because of the time they have to practice.!