SODOM & GOMORRAH: As technology becomes more complex, the temptation to apply it to the wrong fields intensifies.
After the Russian Revolution of 1917, the newly formed Soviet Union began efforts at full scale industrialization. Much of the Soviet Union was still a feudal and agrarian society, and the new socialist paradise had to create an industrialized class of middle class worker to carry out its social policy. As a result, the Soviet Union decided that the impartial domain of science should determine policy. Modern scholars sometimes refer to this as “technocratic socialism“. Technocracy as an idea is not new and people have suggested subordinating politics to science for years.
In this week’s Economist, the weekly publication advocates using machine learning to help determine policy. Machine learning is a branch of computer science that instructs a computer to “learn” appropriate (or inappropriate) outcomes from inputted data. The field developed from pattern recognition and artificial intelligence and today powers things like search engines, spam filters, and computer vision programs. Now there’s a push to use machine learning algorithms to do things like fight crime, reduce poverty, and make other public policy suggestions.
This push is misguided. Already, early models of machine learning programs that would be used to fight crime are wrongly labeling black people as future criminals at rates much higher than mislabeling whites. The data being inputted to such programs is only as objective as the person selecting it, and is therefore subject to the same set of biases. Unlike other biases, however, the ones found in machine learning programs become enshrined in a false legitimacy because they’re seen as more “scientific”. The Economist suggests that paying attention to these biases, writing better programs, and improving the technology will provide “vast” benefits.
Yet we know from other applications of technology that this isn’t always the case. Technology does not intrinsically lead to fair outcomes or promote the best of human nature; technology only enhances human nature and our capabilities. Nuclear fission can be used to obliterate cities as easily as it can be used to provide cleaner energy. Machine learning, like any technology, doesn’t solve for bias or bad policy; it only enhances the policies we already have.