Why social media should give users control over their algorithms
When building complex products avoid creating black boxes - instead give your users control
People trust and use algorithms more when they have (some) control over them.
Machines often make better decisions than humans. Yes, the danger of algorithm bias is real. But decades of research show that in many domains algorithms are simply more rational and consistent than humans at decision-making. In medical diagnosis, machine prediction models improve judgment accuracy, on average, by 10%. Similarly, machines out-forecast humans in areas such as sales, company performance, and finance.
Still, when using technology users are often averse to algorithms despite their benefits. Surveys show a majority of people who forecast for a living - such as sales people or doctors - do not use algorithms to help them make decisions.
As product managers or technologists, this is a problem. We spend months working on ways to help people make decisions with data - only for them to pay little attention to this work when they are making decisions.
In product, be very careful when taking autonomy away from users.
Based on the work of two psychologists (Deci & Ryan) self-determination theory argues people have three basic need to achieve happiness: competence, autonomy, and relatedness.
Autonomy refers to our need to control the course of our own lives. Having an algorithm rip out this control to be replaced by a mysterious black box lowers our autonomy to zero. No wonder people may be averse to using algorithms to decide their lives - even when outcomes are often better.
Fortunately, there is a way to help people make better decisions with data - give them control over how they use it.
Academics from the University of Chicago took their knowledge of psychology and applied it to how we may be able to help people be more comfortable with algorithmic decision-making.
They ran studies where users were able to adjust and play with the algorithm themselves. Even when the control was limited, this change prompted participants to be more likely to choose an algorithm and perform better on tasks as a result.
It’s a simple approach. Give your users some control over your algorithm and watch trust and engagement increase.
E-commerce brands already give their customers significant control over their algorithm. Function of Beauty and Prose offer in-depth surveys to their customers that result in a custom recommendation for shampoo. Not only do you finish the survey with a sense of investment in the product - the customer has a real sense of the data that is feeding into its recommendation.
Searches for ‘custom shampoo’ are increasing over time - showing a growing consumer interest in having control of their own personalization experience.
All data-driven products should consider how to give their users more control over their algorithms.
Given its success, we should expect to see these controls move out of e-commerce towards any platform that relies on data-driven insights or actions.
Imagine if Instagram allowed users to easily customise their own feeds. Not every person would bother - but just having the ability would increase trust and engagement and - maybe - true value over time as well.1
If you are building a product that asks people to take action based off data - to be successful you will need to give your users more control.
These options may not be that far away. Recent announcements by Jack Dorsey (former CEO of Twitter) and Elon Musk (current CEO of Twitter) suggest movement - on the technical side - to open up algorithms to developers.