AlphaGo can beat a human because it makes use of its ability to train itself millions and millions of games, a kind of big-data approach. A human can never play so many games in his life time. It might be the best strategy to train a computer to play Go or table tennis in this way. Is it such big data the only way (if at all) for a human to learn things and become intelligent? Let’s make things simple: is big data also a good way to recognize / identify / distinguish a cat?
No, absolutely not! In fact, a human hardly becomes intelligent through big-data training. Instead, a human unconsciously builds something in his brain. This something is inaccurate, raw and fuzzy but work pretty well. No idea how this something is built but you must have seen how a 3-year old baby recognize a cat quite easily.
You show the baby one or two real cats or even pictures of cats and the baby immediately spots a real cat accurately like an “expert”. You have just used a very small set of data to train the baby, how come the baby becomes a much better cat “recognizer” than a NN (neural network) computer which has been trained with ten of thousands of cat pictures? Lot of things are unknown but one thing for sure: it was not big data! It was small data and something else!
Babys are clever computers. If we do AI, let’s first think and later do as a 3-year old!
Leave A Comment