It’s a funny twist of brain gymnastics that the philosophical notion that learning can not exist without forgetting popped into my head. I don’t even claim this is a solid notion, but it is going to be an important focus of social interaction as our HCI gets sophisticated. It’s well-acknowledged that computers (and indeed paper and pen) are useful in augmenting our limited working memory and can augment the mechanics of human memory. However, would “socially adept computers” improve their “social performance” by forgetting? Of course only after learning. And forgetting in the “hidden and distant memories” sense, not terminally. And should that forgetfulness be a simulated facade or a chronic and incurable condition as actually experienced by organic minds? When Mr V “forgets” to balance while doing his dance routine, it’s superficial and a sign of bad health (actually, the balance algorithm behind “walk” and “stand” is simply not used during “dance” and forgetting in this way is metaphorically equivalent to a cognitive defect or ill health).
I do, therefore I’ll write more on these new questions when there’s something new to report.
The brain gymnastics I was performing was describing the computer vision design and research process we’ve been going through this week. In short, turning a seemingly “near impossible visual emotion detection research and software acquisition problem” into a “simple, incremental, component-structured, low risk, actionable and manageable approach” to hopefully bring us to a few incrementally improving implementations over the next couple of months. My forget-event, was remembering why (with the eigenfaces – quick and dirty – approach), each pixel starts life as a dimension, even though in other CV (computer vision) thinking “key points” (far fewer in number that the number of pixels in an image) are each described (e.g. in a vector) and each of those descriptions are themselves (almost always) multidimensional. So surely pre-PCA eigenface images should start off with a [number of pixels] X [number of dimensions per pixel descriptor] number of dimensions. Actually, each pixel is one dimensional with eigenfaces (just the greyscale luminance value), but then how on earth can these “keypoint” descriptors be used to search/match keypoints in other images? (And in writing this down I am remembering more and more, so I need to stop or else I won’t be experiencing this wonderful phenomenon anymore!)
But the simple truth is, I forgot. Therefore it didn’t make sense. Therefore the eigenvalue approach was flawed (more so than my impression of my own cognitive performance being flawed), therefore I am practicing human cognition that was flawed (and thus normal), therefore I would pass the Turin Test easily. Cool. I forget, therefore I am of the class of best performing and most versatile cognitive beings in (our known) existence. So to be this amazing, I need to learn so I can forget. These musing streams and eddies could go on for hours, but I end them here.
In any case, I’m happy to rob philosophy of this prize and offer it to science: the science behind learning cycles which would classify this forget-event as a catalyst or stimulant for reflective review. Thus I learn the correct formuation: I forget, therefore I learn. Stranger and stranger.
And as for the original intent for writing and brain gymnastics – the computer vision design thinking description. Coming soon to a demo near you!