Before setting up an algorithmic investment fund I realized two things: one, it will take time to attract investors, and two, my role as a fund manager will have to be expanded to teaching. Nowadays algorithms are applied in various areas – human resource management, marketing, sales, education, customer service, finance, and many other. Everyone, who applies algorithms to make decisions or solutions raises a question – can I trust algorithm completely? My answer is that it depends on your understanding of what algorithms are capable of and where their ability ends.
The key is to understand and correctly identify questions and tasks that can be answered or solved by the algorithms as well as to comprehend what algorithms cannot do. Often algorithms are mistakenly regarded as human beings, users expect them to be exactly as human beings who do certain tasks. It is forgotten that algorithms do precisely what it was designed to do and finds solution only to exactly defined task. It does not interpret. It is not a human, who is influenced by many both internal and external factors.
When applying algorithms, a considerable threat is human desire to rely on their own judgment rather than algorithm at critical moment. Most often this happens when a person decides that the algorithm is not perfect. Of course it’s true. They are not flawless, however they perfectly and precisely do what they were designed to do. So when the task is specific and carefully specified they should be trusted without any interference to decision or solution.
What should be done? My suggestion is to have an interest in algorithms, to learn to know and trust them. Only it seems that trust is a difficult task.
In scientific study “Overcoming Algorithm Aversion: People Will Use Imperfect Algorithms If They Can (Even Slightly) Modify Them”, which was also presented at the Harvard Business Review magazine, the researchers conducted a study which examined three cases: when people ignore algorithms opportunities and rely solely on their own solutions, when people marginally adjust algorithms results based on their own knowledge, and then people only rely on algorithms. The results were the best, when there was no human intervention, slightly lower in case of marginal intervention and worst when algorithms were not used at all. However, people felt more satisfied and more secure when they had some influence on results. The study concluded that it is better to give people the opportunity to make some little contribution to final result or solution rather than use only algorithm, as it does not impair results drastically, but people feel better.