|Alignment Forum Tag|
Machine Learning refers to the general field of study that deals with automated statistical learning and pattern detection by non-biological systems. It can be seen as a sub-domain of artificial intelligence that specifically deals with modeling and prediction through the knowledge extracted from training data. As a multi-disciplinary area, it has borrowed concepts and ideas from other areas like pure mathematics and cognitive science.
Understanding different machine learning algorithms
The most widely used distinction is between unsupervised (e.g. k-means clustering, principal component analysis) vs supervised (e.g. Support Vector Machines, logistic regression) methods. The first approach identifies interesting patterns (e.g. clusters and latent dimensions) in unlabeled training data, whereas the second takes labeled training data and tries to predict the label for unlabeled data points from the same distribution.
Another important distinction relates to the bias/variance tradeoff -- some machine learning methods are are capable of recognizing more complex patterns, but the tradeoff is that these methods can overfit and generalize poorly if there's noise in the training data -- especially if there's not much training data available.
There are also subfields of machine learning devoted to operating on specific kinds of data. For example, Hidden Markov Models and recurrent neural networks operate on time series data. Convolutional neural networks are commonly applied to image data.
The use of machine learning has been widespread since its formal definition in the 50’s. The ability to make predictions based on data has been extensively used in areas such as analysis of financial markets, natural language processing and even brain-computer interfaces. Amazon’s product suggestion system makes use of training data in the form of past customer purchases in order to predict what customers might want to buy in the future.
In addition to its practical usefulness, machine learning has also offered insight into human cognitive organization. It seems likely machine learning will play an important role in the development of artificial general intelligence.
Further Reading & References
Some have proposed that we teach machines a moral code with case-based machine learning. The basic idea is this: Human judges would rate thousands of actions, character traits, desires, laws, or institutions as having varying degrees of moral acceptability. The machine would then find the connections between these cases and learn the principles behind morality, such that it could apply those principles to determine the morality of new cases not encountered during its training. This kind of machine learning has already been used to design machines that can, for example, detect underwater mines after feeding the machine hundreds of cases of mines and not-mines.
There are several reasons machine learning does not present an easy solution for Friendly AI. The first is that, of course, humans themselves hold deep disagreements about what is moral and immoral. But even if humans could be made to agree on all the training cases, at least two problems remain.
The first problem is that training on cases from our present reality may not result in a machine that will make correct ethical decisions in a world radically reshaped by superintelligence.
The second problem is that a superintelligence may generalize the wrong principles due to coincidental patterns in the training data. Consider the parable of the machine trained to recognize camouflaged tanks in a forest. Researchers take 100 photos of camouflaged tanks and 100 photos of trees. They then train the machine on 50 photos of each, so that it learns to distinguish camouflaged tanks from trees. As a test, they show the machine the remaining 50 photos of each, and it classifies each one correctly. Success! However, later tests show that the machine classifies additional photos of camouflaged tanks and trees poorly. The problem turns out to be that the researchers’ photos of camouflaged tanks had been taken on cloudy days, while their photos of trees had been taken on sunny days. The machine had learned to distinguish cloudy days from sunny days, not camouflaged tanks from trees.
Thus, it seems that trustworthy Friendly AI design must involve detailed models of the underlying processes generating human moral judgments, not only surface similarities of cases.
Unanswered non-canonical questions