More and more companies are investing in technology designed to help them predict the future, especially in terms of employee behaviour: productivity, interactions and even emotional states. These algorithms, which promise increased profitability, represent a soaring billion-dollar market. However, they are no more reliable than a crystal ball.
The first problem is that algorithms are based on “induced reasoning”, i.e. they draw conclusions from a sample assuming that they apply at all levels and do not vary over time. This can lead to unfounded decisions or discriminate against certain groups of individuals. The second problem is that algorithmic analysis generates “self-fulfilling prophecies” because, based on these predictions to act, managers create conditions that produce these predictions.
Furthermore, there is no real point in calculating the probability of future events. Probability is based on the possibility of complete certitude, which is impossible to predict… Consequently, “algorithms do not predict but extrapolate”. Once their basic code is developed, predictive algorithms must be “trained” to refine their predictive power. This training is carried out by supplying them with previous organisational data, from which the algorithm extracts trends and applies them to the future. For each application cycle, the algorithm is continuously adjusted to correct “prediction errors”.
Although a predictive algorithm can guess what will happen on the basis of what has already happened, it is nevertheless unable to anticipate change. The key to effective decision-making does not, therefore, lie in algorithmic calculations but in intuition.
The Conversation, Uri Gal (12/02/2018)