With the development of Artificial Intelligence (AI), more and more companies started adopting AI in their decision-making process. Human’s reliance on algorithm advice increases rapidly. However, an interesting phenomenon called “algorithm aversion”, which presents a negative effect on the adoption of the algorithm, is recently discovered.
There are two main methods in the decision-making process: human method and algorithm method (Dietvorst, 2014). With human method, human decision-makers review relevant information manually and make a forecast. With the algorithm method, human only need to enter historical data into the statistical model and the algorithm will generate the forecasting outcome automatically, which is more efficient than the human method.
If the algorithm consistently outperform human, then why not adopt the algorithm method in every decision-making process (e.g. university admission, merge and acquisition, etc.) However, people still often prefer humans’ forecasts to algorithms’ forecasts (Dietvorst, 2014), which refers to algorithm aversion. Dietvorst (2014) stated that people are more likely to exhibit algorithm aversion when they see the algorithm err and they tend to have a higher tolerance to human forecaster’s mistakes. Especially in uncertain decision domain, human is more likely the decision maker that they believe is more likely to provide a perfect answer, which in turn leads to a riskier decision-making method (such as human judgement) and results in the underused best possible algorithm (Dietvorst, 2015). Because people believe that human can get better after practicing and learning from the mistakes, though algorithms can improve as well (Frick, 2015).
Algorithm aversion should be overcome. The reason is twofold: First, human is subject to the influence of noise (irrelevant factors), which leads to significant negative effect in forecasting accuracy (Harrell, 2016). Second, comparing to algorithm, human is not good at giving input factors appropriate weight consistently (Harrell, 2016).
However, there are ways to tackle algorithm aversion. Dietvorst (2015) found that people are more willing to use algorithms when they can modify them, even they know the algorithms are not perfect (Dietvorst, 2015). Surprisingly, people are insensitive to the extent that they can modify the algorithms, which means people don’t mind modifying algorithms in a constrained manner (Dietvorst, 2015).
Dietvorst, B., Simmons, J. and Massey, C. (2014). Algorithm Aversion: People Erroneously Avoid Algorithms after Seeing Them Err. SSRN Electronic Journal.
Dietvorst, B., Simmons, J. and Massey, C. (2015). Overcoming Algorithm Aversion: People Will Use Algorithms If They Can (Even Slightly) Modify Them. SSRN Electronic Journal.
Harvard Business Review. (2019). Here’s Why People Trust Human Judgment Over Algorithms. [online] Available at: https://hbr.org/2015/02/heres-why-people-trust-human-judgment-over-algorithms [Accessed 2 Oct. 2019].
Harvard Business Review. (2019). Managers Shouldn’t Fear Algorithm-Based Decision Making. [online] Available at: https://hbr.org/2016/09/managers-shouldnt-fear-algorithm-based-decision-making [Accessed 2 Oct. 2019].
Hello Zixiao,
Thanks for the Interesting reading. There are quite some insights/research results I did not know about. It is funny how people are. I personally find two things conflicting or even a little hypocritical:
1. People allow other human to make mistakes but not the algorithm, because they trust in human’s future improvement.
I personally find this irrational, sounds more like human are actively looking for excuses to blame and not to use algorithms. As you mentioned in the article, algorithms improve and evolve too, and rationally speaking, I think Algorithms’ improvements is way more stable and trust-worthy than human’s. In fact human may have a bigger chance making the same mistake.
Here’s some interesting crowd thoughts about why people tend to make the same mistakes: https://www.quora.com/Why-do-people-have-the-tendency-to-make-the-same-mistakes-even-after-learning-those-mistakes
2. People are more willing to use algorithms given the condition that they can modify the algorithm, but at the same time they do not care how much they can modify it.
This sounds like someone saying “I just want to know that I am in control! but what exactly and how much do I control, I really don’t care.”
I think people should be more educated about what algorithm does and how. Hopefully this way human can let go of some unnecessary ego and make use of algorithm better.