MIT Research News' Journal

The elephant in the server room

Suppose you would like to know mortality rates for women during childbirth, by country, around the world. Where would you look? One option is the WomanStats Project, the website of an academic research effort investigating the links between the security and activities of nation-states, and the security of the women who live in them.

The project, founded in 2001, meets a need by patching together data from around the world. Many countries are indifferent to collecting statistics about women’s lives. But even where countries try harder to gather data, there are clear challenges to arriving at useful numbers — whether it comes to women’s physical security, property rights, and government participation, among many other issues.

For instance: In some countries, violations of women’s rights may be reported more regularly than in other places. That means a more responsive legal system may create the appearance of greater problems, when it provides relatively more support for women. The WomanStats Project notes many such complications.

Thus the WomanStats Project offers some answers — for example, Australia, Canada, and much of Western Europe have low childbirth mortality rates — while also showing what the challenges are to taking numbers at face value. This, according to MIT professor Catherine D’Ignazio, makes the site unusual, and valuable.

“The data never speak for themselves,” says D’Ignazio, referring to the general problem of finding reliable numbers about women’s lives. “There are always humans and institutions speaking for the data, and different people have their own agendas. The data are never innocent.”

Now D’Ignazio, an assistant professor in MIT’s Department of Urban Studies and Planning, has taken a deeper look at this issue in a new book, co-authored with Lauren Klein, an associate professor of English and quantitative theory and methods at Emory University. In the book, “Data Feminism,” published this month by the MIT Press, the authors use the lens of intersectional feminism to scrutinize how data science reflects the social structures it emerges from.

“Intersectional feminism examines unequal power,” write D’Ignazio and Klein, in the book’s introduction. “And in our contemporary world, data is power too. Because the power of data is wielded unjustly, it must be challenged and changed.”

The 4 percent problem

To see a clear case of power relations generating biased data, D’Ignazio and Klein note, consider research led by MIT’s own Joy Buolamwini, who as a graduate student in a class studying facial-recognition programs, observed that the software in question could not “see” her face. Buolamwini found that for the facial-recognition system in question, the software was based on a set of faces which were 78 percent male and 84 percent white; only 4 percent were female and dark-skinned, like herself.

Subsequent media coverage of Buolamwini’s work, D’Ignazio and Klein write, contained “a hint of shock.” But the results were probably less surprising to those who are not white males, they think.

“If the past is racist, oppressive, sexist, and biased, and that’s your training data, that is what you are tuning for,” D’Ignazio says.

Or consider another example, from tech giant Amazon, which tested an automated system that used AI to sort through promising CVs sent in by job applicants. One problem: Because a high percentage of company employees were men, the algorithm favored men’s names, other things being equal.

“They thought this would help [the] process, but of course what it does is train the AI [system] to be biased toward women, because they themselves have not hired that many women,” D’Ignazio observes.

To Amazon’s credit, it did recognize the problem. Moreover, D’Ignazio notes, this kind of issue is a problem that can be addressed. “Some of the technologies can be reformed with a more participatory process, or better training data. … If we agree that’s a good goal, one path forward is to adjust your training set and include more people of color, more women.”

“Who’s on the team? Who had the idea? Who’s benefiting?”

Still, the question of who participates in data science is, as the authors write, “the elephant in the server room.” As of 2011, only 26 percent of all undergraduates receiving computer science degrees in the U.S. were women. That is not only a low figure, but actually a decline from past levels: In 1985, 37 percent of computer science graduates were women, the highest mark on record.

As a result of the lack of diversity in the field, D’Ignazio and Klein believe, many data projects are radically limited in their ability to see all facets of the complex social situations they purport to measure.

“We want to try to tune people in to these kinds of power relationships and why they matter deeply,” D’Ignazio says. “Who’s on the team? Who had the idea? Who’s benefiting from the project? Who’s potentially harmed by the project?”

In all, D’Ignazio and Klein outline seven principles of data feminism, from examining and challenging power, to rethinking binary systems and hierarchies, and embracing pluralism. (Those statistics about gender and computer science graduates are limited, they note, by only using the “male” and “female” categories, thus excluding people who identify in different terms.)

People interested in data feminism, the authors state, should also “value multiple forms of knowledge,” including firsthand knowledge that may lead us to question seemingly official data. Also, they should always consider the context in which data are generated, and “make labor visible” when it comes to data science. This last principle, the researchers note, speaks to the problem that even when women and other excluded people contribute to data projects, they often receive less credit for their work.

For all the book’s critique of existing systems, programs, and practices, D’Ignazio and Klein are also careful to include examples of positive, successful efforts, such as the WomanStats project, which has grown and thrived over two decades.

“For people who are data people but are new to feminism, we want to provide them with a very accessible introduction, and give them concepts and tools they can use in their practice,” D’Ignazio says. “We’re not imagining that people already have feminism in their toolkit. On the other hand, we are trying to speak to folks who are very tuned in to feminism or social justice principles, and highlight for them the ways data science is both problematic, but can be marshalled in the service of justice.”

Mathematical model could lead to better treatment for diabetes

One promising new strategy to treat diabetes is to give patients insulin that circulates in their bloodstream, staying dormant until activated by rising blood sugar levels. However, no glucose-responsive insulins (GRIs) have been approved for human use, and the only candidate that entered the clinical trial stage was discontinued after it failed to show effectiveness in humans.

MIT researchers have now developed a mathematical model that can predict the behavior of different kinds of GRIs in both humans and in rodents. They believe this model could be used to design GRIs that are more likely to be effective in humans, and to avoid drug designs less likely to succeed in costly clinical trials.

“There are GRIs that will fail in humans but will show success in animals, and our models can predict this,” says Michael Strano, the Carbon P. Dubbs Professor of Chemical Engineering at MIT. “In theory, for the animal system that diabetes researchers typically employ, we can immediately predict how the results will translate to humans.”

Strano is the senior author of the study, which appears today in the journal Diabetes. MIT graduate student Jing Fan Yang is the lead author of the paper. Other MIT authors include postdoc Xun Gong and graduate student Naveed Bakh. Michael Weiss, a professor of biochemistry and molecular biology at Indiana University School of Medicine, and Kelley Carr, Nelson Phillips, Faramarz Ismail-Beigi of Case Western Reserve University are also authors of the paper.

Optimal design

Patients with diabetes typically have to measure their blood sugar throughout the day and inject themselves with insulin when their blood sugar gets too high. As a potential alternative, many diabetes researchers are now working to develop glucose-responsive insulin, which could be injected just once a day and would spring into action whenever blood sugar levels rise.

Scientists have used a variety of strategies to design such drugs. For instance, insulin might be carried by a polymer particle that dissolves when glucose is present, releasing the drug. Or, insulin could be modified with molecules that can bind to glucose and trigger insulin activation. In this paper, the MIT team focused on a GRI that is coated with molecules called PBA, which can bind to glucose and activate the insulin.

The new study builds on a mathematical model that Strano’s lab first developed in 2017. The model is essentially a set of equations that describes how glucose and insulin behave in different compartments of the human body, such as blood vessels, muscle, and fatty tissue. This model can predict how a given GRI will affect blood sugar in different parts of the body, based on chemical features such as how tightly it binds to glucose and how rapidly the insulin is activated.

“For any glucose-responsive insulin, we can turn it into mathematical equations, and then we can insert that into our model and make very clear predictions about how it will perform in humans,” Strano says.

Although this model offered helpful guidance in developing GRIs, the researchers realized that it would be much more useful if it could also work on data from tests in animals. They decided to adapt the model so that it could predict how rodents, whose endocrine and metabolic responses are very different from those of humans, would respond to GRIs.

“A lot of experimental work is done in rodents, but it’s known that there are lots of imperfections with using rodents. Some are now quite wittily referring to this situation as ‘lost in [clinical] translation,’” Yang says.

“This paper is pioneering in that we’ve taken our model of the human endocrine system and we’ve linked it to an animal model,” adds Strano.

To achieve that, the researchers determined the most important differences between humans and rodents in how they process glucose and insulin, which allowed them to adapt the model to interpret data from rodents.

Using these two variants of the model, the researchers were able to predict the GRI features that would be needed for the PBA-modified GRI to work well in humans and rodents. They found that about 13 percent of the possible GRIs would work well in both rodents and humans, while 14 percent were predicted to work in humans but not rodents, and 12 percent would work in rodents but not humans.

“We used our model to test every point in the range of potential candidates,” Gong says. “There exists an optimal design, and we found where that optimal design overlaps between humans and rodents.”

Analyzing failure

This model can also be adapted to predict the behavior of other types of GRIs. To demonstrate that, the researchers created equations that represent the chemical features of a glucose-responsive insulin that Merck tested from 2014 to 2016, which ultimately did not succeed in patients. They now plan to test whether their model would have predicted the drug’s failure.

“That trial was based on a lot of promising animal data, but when it got to humans it failed. The question is whether this failure could have been prevented,” Strano says. “We’ve already turned it into a mathematical representation and now our tool can try to figure out why it failed.”

Strano’s lab is also collaborating with Weiss to design and test new GRIs based on the results from the model. Doing this type of modeling during the drug development stage could help to reduce the number of animal experiments needed to test many possible variants of a proposed GRI.

This kind of model, which the researchers are making available to anyone who wants to use it, could also be applied to other medicines designed to respond to conditions within a patient’s body.

“You can envision new kinds of medicines, one day, that will go in the body and modulate their potency as needed based on the real-time patient response,” Strano says. “If we get GRIs to work, this could be a model for the pharmaceutical industry, where a drug is delivered and its potency is constantly modulated in response to some therapeutic endpoint, such as levels of cholesterol or fibrinogen.”

The research was funded by JDRF.