Opinion - Artificial Intelligence Needs Citizen Science
Are you interested in doing science alongside academic and industry researchers?
No, you don’t need to have a background in science. All you need to participate is to enjoy playing games.
Chances are, you would be interested. In fact, roughly 60% of Americans expressed interest in science and technology in 2015. For reference, 44% of Americans expressed interest in sports, so more Americans want to know about new and developing technology than want to know who won the last Superbowl (No, that’s probably not true, but you get the point).
Citizen science allows anyone to become involved in scientific research. In fact, it’s been around for a while, dating back to the 17th century, and has gained steam over the past several years as scientists learn to leverage the Internet and apps to reach wider audiences. Typically, citizen science allows the average person to help researchers collect and analyze data that they may not have access to otherwise. As exciting as data analysis likely sounds (Hint: It doesn’t), recent iterations of citizen science have focused on adapting data analysis techniques to online games. People have discovered protein structures by playing FoldIt, have identified new genes using Phylo, and have sped up Alzheimer’s research on Stall Catchers.
The applications of citizen science are endless, especially in areas of research that rely on datasets that accurately represent groups of people, such as artificial intelligence.
Oh, wait.
Artificial intelligence doesn’t use citizen science?
Well, no, not exactly.
Currently, there are hundreds of different types of datasets available to AI researchers who are looking to train new algorithms. The data is often collected by research groups or industry companies and made public in an effort to support the wider AI community. However, recent headlines have detailed the failures and biases of AI that has been trained on those datasets: from not recognizing images of black women (or identifying black people as gorillas) to being more likely to classify black people or low-income geographic regions as at high risk for crime.
Some of these failures rely on access to data that the average person could not or should not provide, such as recidivism rates or HIPAA-protected health data. Access to that kind of data would require restructuring of data privacy regulations on a federal level, which should be advocated for both to provide better data to researchers and to protect the privacy of the data that is currently in use, but will not happen overnight.
However, there is no shortage of selfies on the phones and social media profiles of Americans, regardless of demographics.
To be sure, those participating in citizen science should be able to do so with the knowledge that their data will be protected and can only be used with their approval. Blockchain has been proposed as a potential solution to this, as has updated and more clearly defined privacy policies for citizen science research that handles personal data, such as the new GDPR regulations.
Researchers need to engage citizens in the scientific process, both to improve technologies that will eventually be incorporated into their daily lives, and to maintain public support of science and technology during this time of increased public doubt in the scientific process. Create apps that allow people to see how their pictures have improved the accuracy of a facial recognition algorithm. Design games that identify biases in day-to-day decisions that researchers can account for in their research. Educate citizens on how they can become involved in science, and why they should, so that they can go back and tell their policymakers to support the research that might save and improve their lives.
After all, AI is the future, and the future is now. Don’t you want it to treat you well?
Discussion, links, and tweets
I'm a graduate student at Harvard-MIT HST. Follow me on Twitter; to see what I'm up to today, and tweet at me if you want to chat!
Tweet Follow @jordanbharrod