Mistakes and Rationality
So, like all of us, I’ve made numerous mistakes in my life. And I agree with the cliche belief—that I’ve learned numerous useful things from these mistakes. Mistakes are often a way for me to “test my social boundaries”. And that’s important, because many many potentially novel behaviors are behaviors that do test on various social boundaries, and it’s important to have some intuition about where these boundaries lie, so that I can be innovative without being offensive (and also so that I can be efficient and waste as little time as possible on unnecessary social formalities).
Furthermore, past mistakes are often a strong impetus for motivation. I’ve tried many strategies in the past that simply didn’t work. And due to all the valuable time I wasted on them, I always am able to motivate myself by reminding myself of these past mistakes that I’m still very ashamed of (mistakes such as staring at math books for hours and hours on end, while not getting anything out of them).
I’ve also had the nasty experience to see many of my old friendships end badly. But I’ve learned in those examples—I’ve learned how to be better to people, to not expect too much out of them, to try to be appreciative to them and to anticipate what they want, to try to care about them (if possible), and also to see through their numerous white lies. Theoretically, this could have been done if I didn’t have friendships that ended badly. But it would have been harder to do without emotional destabilization, since even I am prone to psychological inertia.
Now, is it rational to make mistakes early on? There are a few things to keep in mind:
(1) Some mistakes have the potential of permanently setting us back in life. My parents, for example, often threatened to force me to get a job, which would have had the very strong potential of setting me back for life. Unfortunately, as someone with both ADD and Asperger’s, I simply cannot do a job and learn at the same time, and that happened, I may never be able to go on to graduate school. I am, however, fortunate enough to have parents willing to pay my way through college. If I didn’t have that option (and was in the lower classes), then yes, some of my mistakes could have forced me into a perpetual cycle of repetitive low-wage jobs effectively for life.
And that’s the key thing with mistakes. Mistakes can sometimes permanently doom your future, especially if you’re in a vulnerable position. Mistakes can also result in permanent social damage, which can be devastating if you’re stuck in that group and have few other alternatives.
(2) Many people have a tendency to overreact to mistakes. Sometimes, they start becoming so overprotective against making future mistakes that they simply don’t experiment as much as they used to (and open themselves to much fewer things than before), causing them to miss potentially important stimuli.
Furthermore, if these mistakes are made in a social setting, there is often significant pressure to overreact to mistakes. Because if you don’t become overprotective against future mistakes, people may believe that you’re incapable of change. And that perception is a often a very dangerous perception, especially given all the disasters from history that have come from rulers who were truly incapable of change.
(3) It is, of course, often best to learn from the mistakes of others. But the circumstances behind their mistakes is often not as local or as personal as the circumstances behind my mistakes. And so I’ve had to experience most of these mistakes myself.
What are your thoughts? Is there a finely-defined optimal number of mistakes to make? (for now, I’m actually more interested in the real life implications of this than the AI implications).
This would also be interesting to AI researchers too, since rational agents might also make mistakes while trying to explore “utility space”. Some areas will have high local utility, and some areas will have high global utility. With AI, reputation and overreaction certainly matter less. But time (and other) costs still matter, and “mistakes” are often then cases where you spend an extended amount of time in areas of “low local utility”, or where you get the utility function wrong, assume that the utility function is applicable in a domain where it actually isn’t, or make the wrong decision.
“Unfortunately, as someone with both ADD and Asperger’s, I simply cannot do a job and learn at the same time, and that happened, I may never be able to go on to graduate school.”
I had to pull you up on this one. I have Asperger’s and am working full-time while studying for a Master’s degree. I have a friend with Asperger’s and ADD who has managed to achieve two Master’s degrees and a PhD while employed.
Don’t let diagnonsenses become an excuse for you not to do things—it may well be that you aren’t capable of some things. If so, it’s because you aren’t capable of those things, not because of ADD or AS. But it may well also be that you are capable of those things, and are being put off from doing them by identifying yourself as ‘someone who is prevented from doing these things’.
See also “On Reasonable Efforts”—http://theferrett.livejournal.com/1587858.html
I found this jarring to read. While encouragement to not dwell on weaknesses and avoid self limiting beliefs is all well and good it is not ok to frame it in a way that prematurely invalidates InquilineKea’s expression of their circumstances with that sort of presumption.
It certainly can be because of ADD or AS. For a reasonable definition of ‘you’ AS is part of who ‘you’ are and can definitely be a causal factor that either prevents certain behaviors outright or makes them incredibly unwise.
I don’t try to let them be an excuse for me to not do things if there’s a very good reason for me to do them. There would have been no point for me to work if I could have avoided it, because I am at the point where I do have to learn as much as possible for grad school.
And while you may know people with both, my forms of ADD and Asperger’s are significantly more severe than most people’s.
Oh absolutely—I’m not saying you shouldn’t make your life easier. If you can do grad school without having to work, that’s obviously better. Just that if you start thinking “I can’t do this because X” then you’re less likely to end up doing things at all. Much better to think of it from a positive POV—“I’m lucky enough to be able to study without having to work as well—I would, of course, be able to do both, but I don’t have to and this is better”.
These ideas are studied in AI theory. “Making mistakes” is called exploration and utilizing existing knowledge is exploitation. I am not familiar with the details or the math of this concept. A quick search finds http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.85.8423&rep=rep1&type=pdf .
Yes, this is the right connection to make.
The purest form of the exploration-exploitation balancing act is the k-armed bandit. If the OP is looking for an abstract treatment of this topic, that’s the way to go.
http://en.wikipedia.org/wiki/Multi-armed_bandit
There is a very clean solution to the k-armed bandit problem. With small probability pick a completely random lever. Otherwise, consider the average payoff from each lever during these randomized exploration steps, and pick each lever with a probability which depends exponentially on the average payoff.
No solution does asymptotically better in the worst case.
Its not a very good model for real life :)
In real life, being worst-case unbeatable is not the only criterion for success! But of course I agree that k-armed bandits don’t answer the OP’s real-life question.
If you intend to make a mistake, you didn’t make a mistake. Or more usefully, if you already knew the subject well enough to make the right mistake, you won’t learn that much. Useful information only happens for me when I genuinely don’t know if what I’m doing is right or not.
This is similar to the idea from probability theory where maximal evidence is gained from an experiment when your prior is 50⁄50.
Yea. On the other hand, if you’re tutoring someone else it might be effective to deliberately manipulate your students into making the right mistakes.
Mistakes certainly can be beneficial. I acquired most of my present level of skill at rationality in the course of throwing off religion. I recently looked back at some arguments I wrote up for the existence of God (although I was still terrified of death, whether or not there was a God, interestingly) and I feel absolutely ashamed for my ineptitude. I certainly was intelligent, I was very persuasive, but it was flat out wrong. That mistake (or rather, series of mistakes) is one of my strongest motivators to become stronger. Tsuyoku naratai. At the same time, I have held out from running certain experiments because I knew that if the experiment didn’t work the way I expected, it would be very detrimental to my own situation. I had to find other ways to do it.
At the same time, Eliezer has talked quite a bit about that moment when “the tools break in your hands”. I’ve never had that experience. I have never had a truly surprising, major failure, where I applied everything I knew and it still didn’t work. Eliezer seems to think that it helped him a lot, and if that is true, I kind of want that to happen. Obviously, I can’t make it happen, since that would defeat the purpose. Still, it might be good to keep pushing yourself further and further, until you reach the outer limit of your abilities.