A Summer Camp With a Long Plan: Keeping Bias Out of Artificial Intelligence
AI4ALL seeks to end bias in artificial intelligence
Anaya Bussey didn't know much about "artificial intelligence" when she arrived at a camp at Princeton University earlier this summer other than that "it was definitely blowing up." But after just three weeks here she and other students—all incoming high school juniors—teamed up to use the technology to help diagnose melanoma by looking at skin lesions.
Bussey, 15, who is from the Bronx borough in New York City, has been interested in computer science since she was in elementary school. But there have been times when she's been one of only a handful of girls—or black students—in a computer class or program.
That wasn't the case at the Princeton summer camp, run by AI4ALL, a two-year-old nonprofit that seeks to increase diversity and inclusion in AI education, research, and policy.
AI4ALL believes that getting students like Bussey in on the ground floor of AI is about more than just opening the doors to a hot and potentially lucrative corner of computer science to underrepresented groups.
It's also about making sure the technology works better.
AI has indeed "blown up" over the last decade. The technology trains a machine to do tasks that come close to some of the things a human brain can do. That means, by using data, AI can learn to recognize faces and voices, understand natural language, and even make recommendations. It's used in everything from Snapchat filters to self-driving cars. And demand for AI jobs has more than doubled over the past three years, according to research from the job-placement website Indeed.com.
The problem? There's a major potential for bias with the use of AI.
For instance, studies have shown that the AI-driven risk-assessment algorithms used to figure out criminal sentences tend to make harsher predictions about black defendants than white defendants. Amazon's AI-powered hiring system appears to screen out female applicants, which led the company to ditch it. Facial recognition technology has been a lot less accurate in identifying people of color, and women, than white men.
"It's easy to portray the danger [of AI] as killer robots," said Tess Posner, AI4ALL's chief executive officer. But the real danger behind the new technology "might be stopping someone from getting a job because of a biased algorithm that we don't even know is a biased algorithm."
Programs like the Princeton summer camp can help counteract that because, "at the end of the day it's like who's in the room, who's asking the questions, who gets to shape this from the beginning," Posner said. That means making sure that a diverse workforce is helping to develop the technology as it begins to have a bigger and bigger impact on society.
The message isn't lost on the Princeton students.
"I definitely think that having different people in the room will help, because if you're just taking one segment of the population, they are going to be thinking about things relatively the same way," said Daphne Liu, who goes to West High School in Salt Lake City. "Different people will all think different and have different ideas and the collaboration of everyone working together will definitely be much better than if it's just one type of person."
But right now, there just aren't a lot of people of color—or women—helping to develop the technology.
The applicant pool for jobs in AI is 71 percent male, according to a report released last year by the AI Index, which tracks data on the technology. And in 2018, women made up just under a quarter of computer and information scientists in the United States. Black men and women comprised just 5.4 percent, and Hispanic men and women only 5.1 percent.
Solving Big Problems
The camps, which last one to three weeks, kick off with some technical work. Students are exposed to Python, a coding language that helps them program in AI applications. They may do some work in math concepts related to AI, like statistics and probability. They'll also learn about some of the broader data issues that inform AI systems, including how data are collected and cleaned, and how to mitigate bias.
Then, the students break up into smaller teams for projects, typically centered around how AI can be used to solve big societal problems. Past projects have focused on cyberbullying, increasing mobility for aging populations, and sending aid after a natural disaster.
At Princeton, students divided into four groups. One looked at DNA sequencing to help figure out an individual's ancestry, information that can be used to help figure out how suspectible they are to certain diseases or find the best course of treatment.
Another group used AI's image-recognition powers to sort through pictures of skin lesions, determining which were cancerous. A third also relied on image recognition, but used it this time to figure out which sections of the Amazon were hit by deforestation. And the fourth used natural language processing to figure out which news stories didn't match their headlines, a possible sign of "fake news."
On the last day of camp, students presented on how they used advanced machine-learning algorithms—and a lot of trial-and-error—to help tackle these problems.
And they talked about the broader ethical and policy implications of the topics they covered. The fake news group said that the term is tough to define—for some people, any article they don't agree with is fake news, which would mean there are no true news stories. Both the DNA sequencing and the melanoma groups cited Henrietta Lax, a woman whose cancer cells were used in research without permission as a cautionary ethical tale.
The chance to use AI to deal with big challenges is part of what makes this such an attractive potential career to Kai Lockwood, a 16-year-old from Durham, N.H.
"I've always really wanted to do something that will help people," Lockwood said. "I've definitely learned that AI is less [about] people sitting in Silicon Valley in office buildings, disconnected from the rest of the world. There's people out there in AI who are definitely trying to be connected to the problems that they are solving." Lockwood wants to explore merging AI with another big interest: biomedical research.
Each camp also has a particular theme. At Princeton, it's the intersection of AI and policy. The campers took an overnight trip to Washington where they met with people working in AI at the Federal Trade Commission and the National League of Cities.
Another goal of the projects: to give every kind of student an entrée into AI, even if they aren't wannabe coding wizards. More-advanced students may actually work on coding the algorithms used in their projects. In other cases, they'll tell the instructors—in this case, Princeton graduate and undergraduate students—what the outcome should look like and the instructors will code it for them. Still others may delve into ethical implications behind their projects. "The projects are an opportunity for us to connect with the campers and try to really reach every one of them and [help them] see a path for themselves through AI," said Olga Russakovsky, an assistant professor of computer science at Princeton and one of three co-founders of AI4ALL. "They can explore data rather than be a coder."
It isn't all seriousness. Some of the students in the rainforest group took a break to sing along to a Hannah Montana YouTube video before their presentations began.
That camaraderie, though, is also part of the point.
"I've done coding a lot, but there's not a lot of girls in my computer science classes," said Liu, the Utah camper. "When I came here, I felt a lot more comfortable because there are just more people that I feel like I can bond with."
AI4ALL's camp program, which this summer operated on 11 college campuses, including tech magnets like Stanford University and Carnegie Mellon, accepted just 10 percent of the thousands of applications it received in 2018. It places a priority on recruiting students who are low-income, students of color, and women. Last summer, 78 percent of participants were women, and 22 percent were men. About 84 percent are students of color. Another 11 percent are white. Roughly five percent are multiracial or declined to specify their race.
The Princeton program specifically focuses on recruiting women and underrepresented minorities. This year, the Princeton program includes 32 students. The program looked for strong math and science students with demonstrated leadership ability. Two-thirds of the students are girls, and one is a gender non-binary student. About half identify as black, Hispanic, or Latino. And two-thirds are receiving financial support to cover part or all of the $6,000 tuition.
The selection process puts a premium on students who show demonstrated leadership potential, but not necessarily on those who have aced Advanced Placement Computer Science classes. AI4ALL is looking for kids who might not have access to those types of courses back at their home schools, Posner said.
Equally important is AI4All's burgeoning alumni network, which will stand at roughly 540 at the end of the summer. The program continues to stay in touch with its former campers, offering resources like webinars on college access and connections to research, internships, and other learning opportunities at research universities and big-name tech companies like Google, Microsoft, and Amazon, or those in less technical areas, like General Motors and Prudential.
Sixty-one percent of alumni continue working in AI after they leave the program, by starting an AI club at their school, for example, or tackling a research project. One alumna went back to her farming community in Salinas, Calif., and used AI to help solve problems with water quality. Seventy-seven percent of alumni say they want to go into AI in their careers.
Maya Kahló De Los Santos, now headed into her senior year at Bergen County Academy in Hackensack, N.J., a public magnet school, participated in the Princeton program last year.
Before joining the program, De Los Santos didn't know much about AI. "My background was a little bit of coding, a lot of engineering," she said. "AI to me was like, the Terminator. That's the grasp I had of AI."
Now she's hooked on the technology. She started an AI club back at her school. She applied for—and got—a $1,000 grant from AI4ALL to help fund a summer technology camp in her district for middle school girls. And she spoke at a STEM education conference at Georgia Tech in Atlanta.
One of the best parts of the camp? Getting to meet women and people of color working in AI.
"I think with all STEM fields, they are heavily dominated by males," De Los Santos said. "What AI4ALL has done is shown me people who look like me. … It's a different feeling when you get to see somebody who looks like you in a career you'd like to pursue."
Vol. 39, Issue 02, Pages 6-7Published in Print: August 28, 2019, as This Is Not Your Average Summer Computer Camp