From Algorithm Aversion to Appreciation: Optimizing Algorithm Recommendation Disclosure with Dynamic Field Experiments and Deep Reinforcement Learning

摘要

Recently,many companies have rushed to disclose their deployment of algorithms such as artificial intelligence and machine learning tools(Aenlle 2018;Moshe 2018;Osterloh 2017;Welsh 2019).For example,Amazon,Burger King,Lexus,Toyota,UPS,and others disclose AT/ML algorithms in their campaigns and product recommendations(Griner 2018;Google 2019;Mogan 2018).From the aspect of the supply side,such algorithm disclosure may garner a myriad of business benefits because algorithms are advantageous over human judgments in computing big data and making individualized predictions(Egan 2019;Hosanagar et al.2013).However,on the demand side,customers may have pushback and resistance against algorithms.Against this backdrop,our research investigates three questions:(1)Does the disclosure of algorithm-based,versus human editor-based,recommendations decrease or increase customer purchases?(2)Can firms leverage free promotions to mitigate algorithm aversion? And(3)how to craft an optimal policy of algorithm disclosure so as to lower each individual's algorithm aversion and maximize the total sales revenues? These questions are critically important for managers to understand the algorithm disclosure effect,mitigation strategy,and the optimal policy for both the demand and supply sides.