Einstein already knew that measurements and observations are never neutral. They are always connected to what has been measured and observed. This is part of the theory of relativity. In data science, one occasionally pretends that this basic physical law does not apply. This is apparent in the hype surrounding artificial intelligence (AI).
AI is not an aim in itself, it always exists in resonance with reality. The crucial question is how we shape this dependence: meaningful, useful, fair, or exclusionary, destructive. The human being programs the learning algorithm, the human being is in control. None of the AI should be to blame. It can’t help itself if it was programmed stupidly. I recommend the book “Attack of Algorithms” by Cathy O’Neill. The mathematician speaks of “Weapons of Math Destruction”. Her explanations abound with examples of unreflected belief in technology and the delusion of feasibility. And where does that come from? It is a widespread sin of omission of numerous AI solutions that they only draw the superficial benefit from the data and that learning from feedback with reality is forgotten. Sooner or later, that’s bound to go wrong.
The self-learning processing of big data often reinforces dominant patterns.
This leads to exclusion, misincentives and social injustice – a toxic cocktail. Examples include chatbots who have become racist, sexist monsters because they have value-free reinforced the attitudes prevailing online. Also exciting are those cases in which American tuition fees have risen by 500 percent over time, unnecessarily impairing the quality of education in the USA.
If it is possible to strengthen individual attitudes and increase the costs of a system, it must also be possible to reduce costs and increase inclusion. I think so. So the machine only becomes really intelligent when it automatically supports us in the direction of fairness, inclusion and quality. Artificial intelligence only deserves its name if it can critically observe its own effect and control it itself along basic values. Everything else should please be called stupid intelligence.
Column by Esther-Mirjam de Boer, published in the Handelszeitung on 9 May 2019.