Machine Learning is Not AI
It's pretty hilarious that Microsoft's AI chatbot Tay went "full nazi" in less than a day after coming online but not surprising. I wrote before in my piece Creating Sentient Artificial Intelligence that much of what's considered "Machine Learning" right now is more akin your senses or motor skills - the muscle memory part of your body - than to your brain and that it's really not suitable to just throw these algorithms at more general problems and expect them to work.
Right now the algorithms learn solely on repetition and trial-error, but if we think about how we teach in school, the last thing we want is for students to learn by rote memory (parroting the textbook rather than understanding the concepts) - so why do we design our algorithms that way? It's because these algorithms are only a piece of the puzzle. When we as humans do things on repetition, it's to train it into muscle memory so we do not have to think about it. In the same way that we are not conscious of every step we take or every muscle we use to swing our arm, what we've built and dubbed machine learning so far are more like our muscles or senses; they have some level of learning to automate lower-level functions that the higher level intelligence (the mind, decision making piece) does not have to think about. It's just that somewhere along the way (or perhaps we never really stopped to reflect on this), we forgot to look back on how our own intelligence works and see if it makes sense to how we're designing or using the algorithms we have so far. You wouldn't touch a hot stove 100 times to train your hand to avoid it, even though that'd probably work to build up your reflexes - but that's the thing, these are your reflexes, not the thinking conscious part of your brain.
That said, I'm not against machine learning or saying it's not useful, just that it's not artificial intelligence; it's only part of the bigger puzzle. We're less than half way there but flaunting it as it's all the way there. We've come really far in creating a reflexive hand or eye with strong muscle memory, which is a great thing, but intelligence and autonomy is a whole other layer we've still got to add before we can expect anything meaningful on more general problems. We've built the eyes to see the data but not the mind to think about it.
Right now the algorithms learn solely on repetition and trial-error, but if we think about how we teach in school, the last thing we want is for students to learn by rote memory (parroting the textbook rather than understanding the concepts) - so why do we design our algorithms that way? It's because these algorithms are only a piece of the puzzle. When we as humans do things on repetition, it's to train it into muscle memory so we do not have to think about it. In the same way that we are not conscious of every step we take or every muscle we use to swing our arm, what we've built and dubbed machine learning so far are more like our muscles or senses; they have some level of learning to automate lower-level functions that the higher level intelligence (the mind, decision making piece) does not have to think about. It's just that somewhere along the way (or perhaps we never really stopped to reflect on this), we forgot to look back on how our own intelligence works and see if it makes sense to how we're designing or using the algorithms we have so far. You wouldn't touch a hot stove 100 times to train your hand to avoid it, even though that'd probably work to build up your reflexes - but that's the thing, these are your reflexes, not the thinking conscious part of your brain.
That said, I'm not against machine learning or saying it's not useful, just that it's not artificial intelligence; it's only part of the bigger puzzle. We're less than half way there but flaunting it as it's all the way there. We've come really far in creating a reflexive hand or eye with strong muscle memory, which is a great thing, but intelligence and autonomy is a whole other layer we've still got to add before we can expect anything meaningful on more general problems. We've built the eyes to see the data but not the mind to think about it.
410 unique view(s)
