Every day we use information about cause and effect to help make decisions. With advances in machine learning, computers are now able to turn data into sophisticated models of causation.
These models have the potential to help inform the choices that people make in their daily lives. However, it’s unclear how well people can use such models to make real-world decisions.
To me, this is about the difference between data, like raw numbers, and real information, like how the numbers interact with each other (causation, influence, etc.)
About 4,000 people completed online surveys through Amazon’s crowdsourcing website Mechanical Turk.
This is not exactly a random group when subjects self-select like this.
Participants were given hypothetical scenarios and asked to respond to multiple-choice questions.
The researchers were interested in understanding how people combine new information with existing knowledge and beliefs.
Some of the scenarios involved familiar topics like weight management or saving for retirement. Others were completely novel—involving, for example, mind-reading aliens and musical robots.
The surveys showed that additional information can actually lead to worse decision-making.
In a real-world scenario about body weight, a topic on which participants would have experience, “Jane” was trying to avoid gaining weight during her first year of college. Some of the participants were given additional information (e.g., “30 minutes of exercise three times a week is recommended to maintain a healthy weight”) either as text or a diagram. Others received no additional information.
When asked how Jane should avoid weight gain, the people who were given no new information were more likely to make the correct choice. It seemed that familiarity with a topic might influence whether additional information hurt or helped their decision-making.
To test this idea, the team posed a question about diabetes management to a group of people, some of whom had personal experience with the disease. Those with experience did worse on a question about controlling diabetes if given a diagram about maintaining healthy blood sugar than if given no information.
The researchers hypothesized that new information affected the confidence of people with existing knowledge on the topic. This caused them to second-guess themselves and make incorrect decisions.
More surveys confirmed that people became less confident in their knowledge when given causal information about a familiar area.
The opposite was true of a novel topic, where additional information increased confidence.
The authors note that more causal information isn’t bad in itself but needs to be tailored to the individual
When it seems everything else these days is “tailored to the individual”, why are standard restrictions on doses of opioids applied to all patients for all types of pain all the time?
When have people ever let the truth get in the way of what they want to believe?