My colleague Harold Jarche pointed me to a post by Dave Snowden about deliberative practice, which I found interesting for a facet not part of the key article (which makes worthwhile points). Among a list of important requirements for meaningful activity that is part of effective learning (i.e. it’s not just 10K hours of practice that makes an expert, but what sort of practice has an effect), Dave cites that “at least half of … experiments should fail”. Think about that for a minute.
What that’s saying is that at least half of the money you invest in new things could be conceived of as being wasted. You might be considered a very ineffective manager if 50% of your investments don’t yield returns! Now, first of all, I’m sure you recognize that failed experiments aren’t a complete waste, as long as you learning something (“when you lose, don’t lose the lesson” as the saying goes). Still, 50% might still seem like a high failure rate. But is low risk really good?
I remember hearing a talk by a Canadian AI researcher (who’s name escapes me after all these years) who had studied the optimal ratio of success to failures in helping a system learn. Now this was particular to the learning algorithm he’d chosen, but his result was roughly that you learned fastest if you failed two-thirds of the time, or around 67% failure. Now that’d be pretty disheartening, but if you could take emotion out of the equation, e.g. made it safe to fail, would learning faster be a big enough argument to support bigger failure?
It depends on a lot: on how well you discern the lessons from failure, how well you tolerate failure, how much social scrutiny and how tolerant that public viewpoint is, but it’s interesting to contemplate what might be an optimal context for failure, and given that, what would be the fastest way to learn, and capitalize on that learning. You want your experiments to be designed in the first place to yield maximum information, but if they do, what would a valuable success rate be?
I do believe that they who adapt fastest will be the survivors. That adaptation may be subconscious, but I think conscious reflection is a valuable component. Certainly for sharing the learning, so no one else has to make the same mistakes. So are you learning just as fast as you can?
Generative points about learning requiring a tolerance, if not a yen, for failure. But something about ” … they who adapt fastest will be the survivors” seems too narrow. Successful adaptation is sometimes defined as survival of the fitter, where fitter can have many dimensions. But survival of the faster is a one-dimensional measure. This seems counter to the earlier statement about effective learning depending on total time *and* the quality of practice. And what about those who learn slowly and whose learning is deep? Some urgency is needed, but If I get anxious about being fast I might not learn very much at all.
Bill, good point. I suppose failing fastest within the constraints of systematic experimentation and effective reflection, perhaps. Still, there are competitive pressures to keep up with, so it’s a balancing act. I do want deep learning, but also continual. Just looking to optimize. And frankly, I wouldn’t mind us slowing the pace so all are brought along, but that’s a different topic…:).
Clark, yes, unhappily, we cannot ignore competition and the pace of life. And that is a different topic. However, I ran across two recent articles related to this conversation. First, a Salon article by Alice Karekezi on why kids need solitude to learn: http://www.salon.com/2011/12/28/why_kids_need_solitude/singleton/. Diana Senechal, interviewed in this article, makes a case for how students need time to wrestle with material in order to learn it. The awkward experiences of not knowing the answer are conducive to deep learning. “Not knowing” is not the same as failure, but I see them as related in terms of learning. And second is an article on how the Finnish education system is explicitly not about competition, but cooperation. This is a stretch regarding your post, but your mention of “competition” created an association for me. That article is here: http://m.theatlantic.com/national/archive/2011/12/what-americans-keep-ignoring-about-finlands-school-success/250564/ .
It would be good if you can find a citation to the talk you heard about a 67% failure rate as optimal. This number has already been stated on Twitter citing your blog post as a source. It would terrific to have something more than a hearsay reference.
A couple of thoughts occurred to me whilst reading both Bill’s and Clark’s comments. For me there is some difference between planning for failure, which I write into my teaching, learning from my own failure (which I suppose I also plan for) and public failure.
If I want students to learn then I need to create situations in which they can be “tested”, that is a time when they can learn to experience knowing that they do not know. This chance to “fail” teaches them not just the subject knowledge they need but also how to tolerate failure and how to deal with failure.
Every time I try to improve my teaching it is because of some sort of perceived failure. This is non-competative failure. In this 50-75% failure rate you mention do we include the failures we have identified for ourselves or only those failures others have identified for us?
Thanks for the two articles Bill.
BTW, I’ve received the name of the researcher I heard speak on this, Brian Gaines, but despite contacting him I haven’t got a pointer to a paper, at least not yet.
If you take the adrenalin–or emotion–out of failing, making it “safe to fail,” you lose the value of the failure.
What is learned from failure is the ability to adapt, redirect, redesign, examine from a different perspective to effect change, progress, perhaps even survival. As I see it, the risk of failure needs to be there to both motivate the individual toward a successful resolution and to realize the benefit of the win.
Just how fast do we learn from failure? Well, that’s a tough question. If the repercussions from a lost sale by your action/inaction caused you to lose your job, you would probably learn quickly. If a failed action caused few or no consequences, you may not learn at all. In my view, there are too many variables to adequately consider it.