Explain Algorithms to Affected Individuals

Convincing a consumer to accept an AI-based unfavorable decision about them requires establishing trust in the algorithm by describing why the AI tool made the decision. To address this demand, organizations can provide, along with a decision, an explanation of how the machine logic reached its conclusion.
     Researchers at University of Calgary and Mount Royal University were interested in what characterizes the best of such post-hoc explanations. Their studies established that ones which include concrete, feasible steps the consumer can take to positively influence future outcomes were especially effective in preserving positive attitudes toward the decision and the marketer. The researchers name them sensitivity-based explanations because they describe how much the values of specific variables would have to change in order to alter the decision outcome.
     A scenario used in the studies was for a car insurance applicant who did not receive the best rate. A sensitivity-based explanation was, “If 10% or less of your driving took place at night, you would have qualified for the cheapest tier. If your average miles per month were 700 or less, you would have qualified for the cheapest tier.”
     The positivity of study participants’ ratings of fairness and of intention to do business with the insurance company were higher than for study participants given a case-based explanation: “This decision was based on thousands of similar cases from the past. For example, a similar case to yours is a previous customer: She was 38 years old, with 18 years of driving experience, drove 850 miles per month, occasionally exceeded the speed limit, and 25% of her trips took place at night. Claire was involved in one accident in the following year.”
     The researchers are suggesting that the marketer describe the logic of the particular decision. They are not suggesting that the marketer attempt to explain the entire logic of the AI model. This latter would almost surely overwhelm the customer with details. It also might reveal the marketer’s trade secrets or allow the customer to subsequently game the algorithm.
     I’ll add to those reasons the fact that the entire logic of the AI model may be unknown to the marketer. Aside from the challenge of earning trust from the consumer, the complexity of machine learning—AI systems which learn on their own beyond the data they’re originally fed—readily results in marketers questioning whether to grant trust. Marketers want decision-making transparency, too.

Successfully influence the most prosperous & most loyal consumer age group. For the specific strategies & tactics you need, click here.

Click for more…

Relax Guardedness with Gricean Norms 

Scotch Faux Explanations to Explanation Foes

Image at top of post based on photo by Eugene Triguba from Unsplash