The integration of Artificial Intelligence into companies will create complete new job profiles. One of them is the “AI Compliance Officer”, but also the “AI Coach”.
After AIs already won in Chess and other board games against their human opponents, in 2015 Google let their DeepMind-AI learn different classic Atari games, such as “Ms. Pac Man”, “Space Invaders”, “Video Pinball”, “Q-Bert” and “Montezuma’s Revenge”. The AI started with trial-and-error. The software learned based on “classic conditioning”, as it received a positive amplifier. This similar to “Pavlov’s Dog”. In the beginning the AI made only slow progress, but then advanced and at the end could beat the human high scores in Space Invaders and Video Pinball. With the more complex Ms. Pac Man and Montezuma’s Revenge the program still struggled.
Around two years later Microsoft changed the setup to “cognitive learning”. The AI observed human players and learnt from its mentors. In total the human players created 45 hours of game-play, which was analyzed by the machine. Nevertheless that the AI still had its problems with Montezuma’s Revenge, in average the machine learnt faster from the human players, as it did on its own before. Based on this, Microsoft concluded that with better human teachers an AI is learning faster. Just as it applies for humans.
Inside an organization employees may require temporary coaches, as they have the required technical knowledge, but may lack of soft skills. Such raw talents need support by experienced colleagues to reach the next level of their career. A possible solution is to team them up with a higher manager, who acts as a coach, so that he can learn the required skills, such as emotional intelligence. Cognitive learning is used.
Similar to this scenario, special coaches may teach AI software to make the adequate decisions. This is relevant as decisions not only have to be maximized for the short-term, but to ensure sustainability to maximize the long-term profit. Ethics & Compliance have to be obeyed, even if impunity would not punish violations to law. The AI has to understand that nevertheless there is a cost of corruption, which can manifest itself, for example, in shrinking markets, raising costs and low profit margins. Even fines based on global investigations have to be considered. Business decisions have to be based on law and values. Furthermore, the AI have to fit to internal organization, including to the human employees. Based on region or even group culture, the software has to interact differently with its human colleagues. Human and machine diversity are no single topics, but the human machine group requires such.
Human AI Coaches can teach such ethical decision making to ensure that the algorithm mathematically understands that transparent business ensures long-term success, even if on the short-term this may lead to lower results. As each company has its individual Code of Conduct, mostly based on its founder, decision making has its individual variations, so that the machines could not learn such behavior automatically from similar software used in different companies. Trail-and-error can easily lead to high fines and reputational damage, cognitive learning and “human understanding” is the more effective solution.