Brynjolfsson & McAfee (2017) Flashcards
machine learning (ML)
when systems learn how to perform tasks on their own. within the past years it has become far more effective and widely available. it represents a fundamentally different approach to creating software; the machine learns from examples, rather than being explicitly programmed for a particular outcome.
Polanyi’s Paradox
much of the knowledge we have is tacit, meaning that we cannot fully explain it. in other words, we know more than we can tell. this is overcome by machine learning.
supervised learning systems
the machine is given lots of examples of the correct answer to a particular problem. this process almost always involves mapping from a set of inputs, X, to a set of outputs, Y. they often make use of a training set of data with labeled examples. the system can then predict answers with a high rate of accuracy. the most successes in AI and ML have been in this category.
deep learning
an approach that uses neural networks. it has a significant advantage over earlier generations of ML algorithms: they can make use of much larger data sets. the old systems could improve as the number of examples in training data grew, but up to a point. it does require more processing power, so it is often run on supercomputers or specialised computer architectures.
unsupervised learning systems
seek to learn on their own. it is difficult to develop a successful system that works this way. if we do, it will have a lot of possiblities, eg. discovering patterns in complex problems in fresh ways.
reinforcement learning
the programmer specifies the current state of the system and the goal, lists allowable actions, and describes the elements of the environment that constrain the outcomes for each of those actions. using, the allowable actions, the system has to figure out how to get as close to the goal as possible. they work well when humans can specify the goal but not how to get there.
