Humans Get Cozy with AI
The increasing presence of machines playing human roles in our lives isn’t anything new. We’ve all abhorred the automation of functions that were once performed by humans. We’ve all rolled our eyes at automated call response systems on customer service phone lines. The takeover of telemarketing does not seem the best implementation of such a promising technology.
Distrust in machines and their ability to competently perform human functions seeds doubt in many of us about the potential of Artificial Intelligence. If I can't trust a machine to take my order or field my customer service request, then how can I trust it to accurately predict the outcomes of my business processes?
The truth is that humans create risk-- at least in certain tasks. Humans tire of data entry tasks and make mistakes. Human biases seep into decisions. Humans are notoriously flawed at forecasting outcomes. This isn't a knock on humans– we are much better than machines in many areas– but we should be honest about where our collective strengths and weaknesses lie.
Recent studies have revealed however that the degree to which human beings trust machine algorithms has improved remarkably. You may know that algorithms already make a sizable number of decisions in our daily lives. From suggesting movies to watch, or books to read, to what clothes would look good on us are all aided by machine algorithms these days. Despite its usefulness, many companies have tried to camouflage the AI component.
The soothing voice on the automated call routing system, a friendly name for the ‘Chatbot’ avatar, or terming machine algorithm selections as “personalized or handpicked” are all ways of humanizing machine algorithms that are otherwise entirely automated.
But in a recent study titled “Do People Trust Algorithms More Than Companies Realize?”, researchers Jennifer M. Logg, Julia A. Minson, and Don A. Moore have discovered that "that people do not dislike algorithms as much as prior scholarship might have us believe.” The research found that people exhibit what the researchers termed as “algorithm appreciation”, where people trust the advice given by a machine algorithm more than the advice provided by other humans.
The study randomly picked online participants and grouped them into 3 categories of prediction.
Group 1 was asked to predict the likelihood of certain geopolitical events occurring, Group 2 were asked to make predictions about what song would end up on the Billboard Top 100 chart. And Group 3 were asked to play matchmaker by showing them the dating profile of a woman and then showing the photo of her potential date and predicting how much she would enjoy their date together.
All the predictions that participants were asked to make were in numerical terms, where the participants were asked to predict the likelihood that an event would or would not occur predicted in 0 to 100%. After the participants had scored their predictions, however, they were provided with a piece of advice in terms of percentage and told that the advising party generating that number was either a human or a machine algorithm.
The participants were then incentivized to revise their predictions with a monetary award. In other words, they were told either that a person predicted 78% likelihood of an event taking place and an algorithm predicted 52%. The participant would then guess which of the two were closer to the actual results. Those that guess which was closer would receive a monetary prize.
Researchers made a numerical estimate of how much a participant would change their own estimate after receiving this advice. If a participant completely ignored the percentage that another human or machine predicted, a 0% would be assigned and would signal that the advice had no impact on the participant’s estimate. A 100% would signal the participant changed their entire decision based on the advice, or at least changed their estimate.
At the end of this, researchers were surprised. What they found was that people relied more on the advice if they thought it came from a machine than from a human being. What was even more encouraging was that the impact of advice based on whether it came from a machine algorithm was superior equally in all three scenarios.
That begs the question: are the efforts to humanize machines really necessary? Next time you encounter an obvious machine, think twice about dismissing its ability to perform.