UC doctoral candidate awarded prestigious doctoral fellowship

“I heard about it from my supervisor, but I think these are the kind of things that you don't really know the significance of until you've been in the field for maybe 10 or more years,” says Ataneka, a doctoral candidate in the University of Cincinnati School of Education’s Educational Studies PhD program (Quantitative and Mixed Methods Research Methodologies concentration). It was only after he was announced as one of only 35 award recipients nationwide that his supervisor, professor Ben Kelcey, explained the importance of the recognition. “When it was confirmed, he told me that everyone in education knows this one is a little bit too competitive!” Ataneka laughs.

To say this level of academic success is a long way from Ataneka’s origins would be an understatement. He was born in the island nation of Kiribati, in Muribenua village on Nikunau island, a coral atoll with a population of close to 2,000 that sits south of the nation’s capital island, Tarawa. Kiribati is recognized as one of the least developed countries in the world by the United Nations, and life on Nikanau is defined by subsistence activities, primarily fishing and collecting coconuts. And while the capital island features some modern amenities, Ataneka wasn’t exposed to many technological advancements as a kid—though what he could access made an indelible impact. 

“We have a TV,” says Ataneka, referring to one single television set used collectively by everyone in his village. “That's our access, our window to see the outside world. You go there, and you pay three coconuts to enter.” It was on this TV that he caught his first glimpse of a life he’d eventually pursue. “I'm pretty sure it was one of the Harry Potter movies. That's where I heard [the term] ‘professor.’” Ataneka couldn’t totally suss out what a “professor” was, thinking the label was perhaps a certification, but he understood the term was descriptive of someone who possessed wisdom.

“I wanted to be a professor.”

This drive to attain wisdom was apparent from an early age and has guided Ataneka throughout his life. He studied hard with a dream of being able to one day be accepted to the best high school in Kiribati, which was in Tawara and where he’d be exposed to a more modern way of living. When he achieved that goal, Ataneka figured he’d go on to university in Fiji, the only pathway to higher education he was aware of—Kiribati has no universities. Then, during his senior year, a teacher told him of a scholarship provided by the Australian government to one student annually to attend college Down Under.

“Typically the top student in the country would go to Australia,” he explains. “The next five people go to New Zealand, and then the rest go to Fiji, funded by the local Kiribati, New Zealand and Australian governments.” Ataneka noticed that the top scholarship to Australia tended to go to students who grew up in the capital, not students from the outer islands who had to navigate cultural differences alongside their rigorous studies, and made it his goal to beat the odds and win the scholarship. “So that's when I kind of started studying extra hard,” he says, adding with a laugh, “I didn't enjoy high school much because I was just studying.”

Scholarship winners were announced on the radio, and his whole village gathered at the maneaba (community hall) to listen as Ataneka was named the winner of the Australian scholarship. “It was a big moment for my village,” he says. “They all know that's a very tough one for our outer-island kids.”

Upon graduation Ataneka headed to the University of Queensland, where he received a bachelor’s degree in economics, then came back to Kiribati. But after a few years working in a government job, he started thinking once again about becoming a professor. He began researching scholarships to attend graduate school in the U.S., eventually finding the U.S. South Pacific Scholarship, which provided funds for four individuals from across the Pacific islands to pursue graduate studies at the University of Hawaii. It cost him nearly a week’s pay to mail in his application, but within a few months he received word he’d been awarded one of the scholarships.

While he was at the University of Hawaii studying public administration, Ataneka learned about how people fund doctoral studies in the U.S., which only one other person from Kiribati had ever done. He knew he wanted to apply his learnings in quantitative economics to the field of education, and he started looking for programs that fit his educational aspirations. He found two programs with faculty doing work he was interested in, emailed both, and when UC was the first to respond, he made the choice on the spot.

Today Ataneka serves as a graduate assistant with Kelcey as he works on his Spencer Fellowship-winning dissertation, “Causal Machine Learning with Reflective Latent Variables: A Latent Deep & Targeted Learning (LDTL) Architecture.”

"We built a new AI architecture to address two fundamental limitations of AI and machine learning models for scientific research," he explains. "First, the core architecture behind today's popular AI tools—for example ChatGPT, which is built on a deep neural network called a Transformer—is designed for prediction, not causal inference. These models are remarkable at spotting patterns: what tends to go with what. But noticing that two things go together is not the same as knowing that one causes the other, and the why is the heart of science.

"Does this new medicine actually work, or did patients improve for other reasons? Did this education program actually raise children's reading skills, or did the better-off families simply enroll first? Did this policy cause the change we see in a community, or would it have happened anyway?

"Science has other important roles but causality sits at the core. It is the difference between knowing what happened and knowing what to do about it. Answering that requires specific assumptions and study designs that must be satisfied before causal claims can be made—machinery that current AI was simply never built for."

The second limitation, says Ataneka, is that current AI architectures alone (such as Transformers) do not know the difference between what is measured directly and what is measured imperfectly.

"Some things in the world can be measured directly: your height, your weight, your age. If the scale says 70 kilograms, that number basically is the truth," he explains. "But here is the thing most people never stop to consider: Almost everything we actually study in science about people and societies is not like height or weight. Intelligence, depression, anxiety, math ability, reading skill, customer satisfaction, trust in government, quality of life, poverty, motivation—none of these can be placed on a scale or measured with a tape. We call these latent constructs: real things that exist, but that we can only see indirectly, through their shadows. We measure them with tests, surveys, questionnaires and assessments—a set of questions whose answers we combine into a score."

Every such score, he says, contains errors. He points to how a student's test score is not their true ability, but rather their ability plus the effects of a bad night's sleep, a confusing question and/or a lucky guess. Likewise, a depression questionnaire does not capture depression itself but rather captures an imperfect reflection of it. The science of measurement—psychometrics—has spent a hundred years studying exactly how large this error is and how to account for it.

"Today's AI models know none of this. To a Transformer neural net or any standard machine learning model, a number is a number," says Ataneka. "A test score is treated with the same blind confidence as a reading from a scale. The model cannot tell that one number is a direct observation and the other is a noisy shadow of something hidden.

"Why does this matter? Because it is well established in literature that when you ignore measurement error in your data, everything you compute afterward becomes unreliable. Relationships look weaker or stronger than they truly are. Effects get missed or invented. Conclusions drift away from the truth, and you have no warning that it happened. If we feed imperfectly measured data into AI models and treat the output as solid science, we are building on sand.”

Working under Kelcey’s supervision, Ataneka in his dissertation develops new AI-based methods that address these two limitations of AI for causal inference with latent variables. “In other words, we are building a new AI architecture that enables researchers to utilize the predictive power of machine learning models while integrating the theory-driven principles of causal inference and psychometrics.”

If this sounds like potentially world-changing research to you, you’re in the good company of the National Academy of Education, whose Spencer Dissertation Fellowship provided a prestigious high point for Ataneka’s academic journey thus far. He’s grateful for the financial component of the award, of course, but he’s equally excited to travel to Washington, D.C. for two professional development retreats, where he will meet top researchers in the field and collaborate with them.

“This is the ultimate achievement of my life,” he says.

Working with machine learning, Ataneka finds himself right in the heart of an exuberant industry, and his career options have expanded in a world being transformed by AI. “If I go to a conference, many people tell me You should consider industry as well,” he says. “Because I’m doing machine learning work. I'm open to it, but my main aim is academia. I want to stay in that.”

All these years later, in a story that spans continents and technological advancements, Amota Ataneka still wants to be a professor. And when he completes his doctoral program next year, he’ll be on his way to a dream that started with a few coconuts and a shared TV screen on a tiny island in the Pacific Ocean.

Featured image at top: Amota Ataneka, recipient of one of only 35 NAEd/Spencer Dissertation Fellowships, sits outside of Teachers-Dyer Complex on the UC campus. Photo/CECH Marketing