I was going to write some awful poetry here, but I'll just be blunt: the velocity and volume of material behind machine learning research leaves me on my knees. Every generation has said that general AI is only a few years away—but not every generation has had classification accuracy comparable to a well-trained human expert. Obviously that's not the last step in assembling your very own heuristically-programmed algorithmic computer, or even a plausible tactical drone, but it's a very big deal.
You don't really find this level of applicable theory in many other fields. Physics builds intricate experiments to test the cornerstones of elaborate cathedrals; advances in biology research are mostly a matter of developing better tools to explore things that we can already unravel step by step; psychology is either slavishly descriptive or succumbs to the same issues as physics, or does both at the same time. Pure statistics is alien territory to anyone but the narrowest professionals who demand the precision that only a Gaussian bump can provide. Even the rest of algorithm design is stuck exploring worst-case analysis and approximation ratios—rarely now does a new method appear which presents a dramatic performance improvement for a given problem which widely concerns other domains. My (current) pet problem, sequence alignment, was solved forty years ago—but no one wants to wait for O(L^N) operations!
ML is the keystone field where theoreticians are genuinely and profoundly important, which naturally lends the most successful of its researchers an extremely intimidating air. Not the severely-tongued pomp of medicine, nor the youthful curiosity of logic; despite being close to everything under the sun which could possibly have a subjective aspect within it, ML's attitude remains singularly without pretension. Either you are an Olympian, and you can intuit the concepts, or you are not, and you cannot.
Perhaps it is the humbleness of statistics that stripped out all of the wit from its ancestral artificial intelligence, perhaps the students simply calmed down as they grew older and took on protégés of their own... or perhaps the aching distance from the metal beneath is to be blamed. In the eighties, Eric S. Raymond could compile a dictionary of jargon encompassing everything from colours of patch wire to secret organizations founded in the image of Alonzo Church. Today, I know more about the Windows API than any of the mentors I will ever have, and I'm only half their age.
But that is how things go. Even Newton knew things in fields his students would never much care for. We are all individual classifiers in evolution's AdaBoost implementation, and the total sum of human experience is our dataset. No wonder we left genetic programming behind.
Small comfort to someone seeking permanent assurances about her self-worth.
Maybe tomorrow my life will snap into clarity.
You don't really find this level of applicable theory in many other fields. Physics builds intricate experiments to test the cornerstones of elaborate cathedrals; advances in biology research are mostly a matter of developing better tools to explore things that we can already unravel step by step; psychology is either slavishly descriptive or succumbs to the same issues as physics, or does both at the same time. Pure statistics is alien territory to anyone but the narrowest professionals who demand the precision that only a Gaussian bump can provide. Even the rest of algorithm design is stuck exploring worst-case analysis and approximation ratios—rarely now does a new method appear which presents a dramatic performance improvement for a given problem which widely concerns other domains. My (current) pet problem, sequence alignment, was solved forty years ago—but no one wants to wait for O(L^N) operations!
ML is the keystone field where theoreticians are genuinely and profoundly important, which naturally lends the most successful of its researchers an extremely intimidating air. Not the severely-tongued pomp of medicine, nor the youthful curiosity of logic; despite being close to everything under the sun which could possibly have a subjective aspect within it, ML's attitude remains singularly without pretension. Either you are an Olympian, and you can intuit the concepts, or you are not, and you cannot.
Perhaps it is the humbleness of statistics that stripped out all of the wit from its ancestral artificial intelligence, perhaps the students simply calmed down as they grew older and took on protégés of their own... or perhaps the aching distance from the metal beneath is to be blamed. In the eighties, Eric S. Raymond could compile a dictionary of jargon encompassing everything from colours of patch wire to secret organizations founded in the image of Alonzo Church. Today, I know more about the Windows API than any of the mentors I will ever have, and I'm only half their age.
But that is how things go. Even Newton knew things in fields his students would never much care for. We are all individual classifiers in evolution's AdaBoost implementation, and the total sum of human experience is our dataset. No wonder we left genetic programming behind.
Small comfort to someone seeking permanent assurances about her self-worth.
Maybe tomorrow my life will snap into clarity.
