Project Fellowship Archives - Knight Science Journalism @MIT

by Hilke Schellmann / June 29th 2022 / Fellowship Life, Guest Posts

Five Ways to Become a Savvier AI Reporter

Editor’s Note: This is part of a series of dispatches from the Knight Science Journalism Program’s 2021-22 Fellows.

Since I took a fateful Lyft ride in late 2017, during which the driver told me that he had been interviewed by a robot, I have been covering artificial intelligence in the world of work. I am still fascinated by this new technology, and as a Knight Science Journalism Project Fellow this year, I’ve been thinking about its implications for journalists and for society more broadly.

AI has been marketed with the promise that it will revolutionize healthcare, hiring, financial decisions, criminal justice, and many other fields. But the reality is much more complicated: Researchers and journalists have discovered that some AI tools do not work as promised and may have caused more harm than good.

I have been asked many times by other reporters how journalists can hold artificial intelligence accountable; many are feeling overwhelmed by AI, which often seems so complex that humans barely understand it. I have wrestled with this question for years and have come up with some ideas:

Check the data: Companies might write in their marketing materials that their algorithm, for example, is 90 percent accurate. That seems like a really high accuracy rate, but how did they arrive at that number? How did they validate the algorithm? I have often seen companies build an algorithm with one large dataset. Developers might use 80 percent of the data to train the algorithm and save the other 20 percent as a “holdout sample,” which they use to test how well the newly-built algorithm works. If a company validates their algorithm with their own holdout sample, the accuracy should be high, since it’s all built on the same dataset. If a company did that, I would ask for further validation: Perhaps they have run studies using “data from the wild” or they had a third party test the algorithm?
Check the dataset: It’s also important to understand how the dataset the company used to build the algorithm was assembled. Where did the data come from? What is in the dataset? What demographics are represented? If a dataset, for example, is built by only interviewing college students, will the data be representative of older people? (Spoiler alert: probably not.)
Create edge cases: Test the technology yourself when that is possible. I once tested an AI tool that was marketed as being able to predict someone’s English competency during job interviews. When I had talked to other vendors, they had shared with me that if someone has a speech impairment, or if the tool runs into problems detecting a person’s voice, the software would recognize the problem and display a message. That’s what I was expecting. Here is what I did: I first did the interview answering all the questions in English and scored an 8.5 out of 9. My English level was rated “competent.” I then tested an “edge case,” answering all of the questions in German, my native language. But I still scored a 6 out of 9 on my English ability, and my English level was rated “competent,” even though I did not say one word in English. It led to a fascinating conversation with the developers of the tool.
Benchmark the algorithm against traditional tests: Many artificial intelligence based hiring tools are marketed as a digital version of traditional assessments, such as personality tests. In this story for the Wall Street Journal, my colleagues and I compared the results of traditional personality tests with those of AI tools that attempt to predict a person’s personality based on information pulled from their social media accounts. Half of the AI results were completely different from the traditional assessment methods.
Bring your own data set and run it against the algorithm: We did that when I reported a story for the Wall Street Journal on a company in Israel that claims their tool can find pedophiles and terrorists based on photos alone. To test the tool, we brought our own dataset of photos of convicted terrorists. We learned that the tool doesn’t work on women and in our batch, the tool identified a photo of a 9/11 hijacker as a terrorist, but not a photo of a white convicted terrorist.

Over the coming year, with support from the Pulitzer Center, I am working with other dedicated colleagues on building a more comprehensive guide explaining different methods to hold artificial intelligence accountable.

Hilke Schellmann is an Emmy-award-winning journalism professor at New York University and a freelance reporter holding artificial intelligence accountable. Her work has been published in the Wall Street Journal, The New York Times, The Guardian, MIT Technology Review, NPR amongst others. She is currently writing a book on artificial intelligence and the future of work for Hachette.

by Rene Ebersole / June 24th 2022 / Fellowship Life, Guest Posts

A Journey to Death Row

Editor’s Note: This is part of a series of dispatches from the Knight Science Journalism Program’s 2021-22 Fellows.

Florida’s death row, less than an hour southwest of Jacksonville, is located in one of the oldest and most notorious maximum-security prison complexes in America. Thirty-three years ago, hundreds of spectators gathered here outside the razor-wire fencing in jubilant anticipation of serial killer Ted Bundy’s electrocution, chanting “Burn Bundy Burn” and clanging frying pans.

One day last March, I drove rural Route 121, rimmed in loblolly pines, to meet two current Florida death row inmates slated to join ninety-nine people executed since 1976, when the state reinstated capital punishment. Carrying only my prison-approved notebook, pen, and voice recorder, I stepped in line behind the correctional officers, in bullet-proof vests and combat boots, on their way to work at the Union Correctional Institution. Established in 1913, Union Correctional is a gray building with blue trim and a large watch tower at the entrance. It currently houses roughly 1,700 male inmates, many convicted of heinous crimes.

Sergeant Derin Blue, a guard at the prison for the last 30 years, escorted me through several iron gates and open-air corridors to a room with cinder-block walls, linoleum flooring, and a lacquered conference table encircled with leather office chairs and a dingy old straight-back chair for a prisoner to sit. Blue pointed me toward one of the comfortable office chairs and assured me that he’d be there if I needed anything. “I’ll make it a pleasant experience,” he said. “I’m that guy—they call me the Deputy of Death Row.”

I came to interview Gerald Delane Murray and James Duckett. Murray, 53, was sentenced to die for raping, stabbing, and strangling to death a 59-year-old Jacksonville grandmother named Alice Vest in her mobile home in 1990. Duckett, 65, was sent to death row for the 1987 murder of 11-year-old Teresa McAbee, found floating in a lake in Mascotte, a small city near Orlando where Duckett worked as a police officer.

Besides their iron-barred lodgings and shackles, inmates Murray and Duckett have something in common: Their convictions were based at least in part on an old-school forensic technique that has since been debunked by DNA analysis. The method, known as microscopic hair analysis — which entails visually matching hairs plucked from suspects’ heads and bodies to strands found at crime scenes — is among a long list of forensic disciplines that have been discredited in recent decades. Others include lead bullet analysis, arson investigation, blood spatter, shaken baby syndrome, and bite mark analysis. Murray and Duckett, who maintain their innocence, say hair analysis was used to wrongfully convict them.

Of 28 examiners in the FBI Laboratory’s microscopic hair comparison unit, 26 overstated forensic matches in ways that favored prosecutors in more than 95 percent of the 268 trials reviewed.

Whether they’re innocent or guilty is in the realm of the courts, but the misapplication of forensic science has indeed led to wrongful convictions — and even wrongful executions. The nonprofit Innocence Project, which helps exonerate prisoners through DNA testing, reports that flawed forensic science contributed to 43 percent of the 375 cases in which U.S. prisoners were exonerated by DNA evidence. Countrywide, false or misleading forensic testimony has played a role in 24 percent of the more than 3,000 wrongful convictions in the U.S. that have been overturned since 1989, according to the National Registry of Exonerations.

In August 2020, a Florida court freed former death-row inmate Robert DuBoise, who was convicted of raping and murdering a 19-year-old woman named Barbara Grams based on discredited bitemark analysis supposedly matching DuBoise to an injury on the victim’s body. The jury unanimously recommended that DuBoise be sentenced to life, but the judge overrode their recommendation, sentencing him to execution in the state’s electric chair. Thirty-seven years later, after DNA from an untested rape kit excluded DuBoise and implicated two other men, he was released.

Florida is far from the only state where juries have been misled by unproven forensics. Claud Jones maintained his innocence all the way to the Texas death chamber, where on December 7, 2000, prison officials strapped him to a gurney and inserted two intravenous needles in his arm, delivering a fatal cocktail of drugs that stopped his heart. The only piece of physical evidence linking Jones to the 1989 murder of East Texas liquor store owner Allen Hilzendager was a single piece of hair collected from the store counter. DNA testing to confirm the hair belonged to Jones didn’t exist at the time of his 1990 trial. But the technique was widely available when Jones was set to be executed. In the weeks prior to his death, he requested a stay of execution from two Texas courts to allow for the hair to be tested. Both courts denied Jones’ pleas.

His last hope was Texas governor George W. Bush, then on the verge of becoming president pending the Supreme Court’s ruling in Florida’s 2000 election between Bush and Al Gore. The governor had recently granted a 30-day stay of execution for Ricky McGinn, convicted of raping and killing a 12-year-old girl in 1993, to allow DNA testing on key evidence in the case. (When the tests supported the jury’s verdict, McGinn was executed.) In a June 2000 interview with CNN, Bush explained his decision to allow time for the testing. “To the extent that DNA can prove innocence or guilt,” he said, “I think we need to use DNA.”

However, Bush was never briefed about Jones’ request. A general counsel memo, later acquired through the Freedom of Information Act, revealed he was only notified about a chemist testifying that the hair was a match to Jones; based on the facts, the General Counsel did not recommend a reprieve. Hours later, Jones, 60, became the 152nd execution during Bush’s time in the Texas governor’s mansion.

Whether they’re innocent or guilty is in the realm of the courts, but the misapplication of forensic science has indeed led to wrongful convictions — and even wrongful executions.

Fast forward to November 2010, when a DNA test revealed the hair didn’t belong to Jones after all; it was from Hilzendager, the murder victim. While the finding wasn’t proof of Jones’ innocence, it revealed a disturbing truth: Texas convicted and executed a man based on false evidence. “It is unbelievable that the lawyers in the General Counsel’s office failed to inform the governor that Jones was seeking DNA testing on evidence that was so pivotal to the case,” former Texas Governor and Attorney General Mark White said in a statement. “If the state is going to continue to use the death penalty, it must figure out a way to build safeguards in the system so that lapses like this don’t happen again.”

In April 2015, the U.S. Justice Department and FBI formally acknowledged that microscopic hair analysis, used for more than two decades prior to 2000, was seriously flawed. Of 28 examiners in the FBI Laboratory’s microscopic hair comparison unit, 26 overstated forensic matches in ways that favored prosecutors in more than 95 percent of the 268 trials reviewed. Counted among the cases were 32 death row sentences, including 14 defendants who had already been executed or who died of other causes while waiting for their turn in the death chamber. With federal officials’ acknowledgment that nearly an entire unit of FBI forensic examiners overstated testimony about hair matches for decades, legal analysts said the big question would be how courts and prosecutors respond to criminal convictions that may have relied on such evidence.

At Union Correctional, Gerald Murray and James Duckett, in orange shirts, matching pants, and handcuffs, took turns shuffling into the conference room to tell me about how the discredited forensic science of microscopic hair analysis contributed to their convictions. “The FBI says our agent lied, gave misleading testimony, made up evidence, and admitted that there was no expertise in hair comparison,” Duckett said. “You would think, especially in death penalty cases, they’d throw out the hair and start over with a new trial.”

MIT Knight Science Journalism Project Fellow Rene Ebersole, an award-winning investigative reporter, is developing a book about the biggest scandal in American forensics—how debunked hair testimony has cheated hundreds, if not thousands, of defendants from receiving fair trials.

by Rebecca Boyle / June 22nd 2022 / Fellowship Life, Guest Posts

Relearning the Art of Covering the Conference

Editor’s Note: This is part of a series of dispatches from the Knight Science Journalism Program’s 2021-22 Fellows.

I walked into the Colorado Springs resort building in April with a mask, a lukewarm cup of coffee, and low expectations. It had been two and a half years since I’d attended an in-person conference, and in that time, I’d changed almost as much as the world had. I’d moved across the country, written a book, and had a baby. So I wasn’t sure what to expect when I entered the 37^th Space Symposium.

The ballroom full of people in military dress uniforms was nothing too unusual—I’d gotten used to that since moving to Colorado Springs—but it was still jarring to listen to a speech in three dimensional space, with other people, and then sit down across from a four-star general willing to take my questions.

I was there to see Gen. John W. Raymond, chief of space operations for the U.S. Space Force, the military’s newest branch. I live near three of its six bases, and my parents live next to a fourth; many of my friends and neighbors serve in the branch, where they are known as Space Force Guardians. So to me, the Space Force is very real, though to many of my fellow science writers it is best known as the butt of left-wing political jokes and the subject of a comedy show starring Steve Carrell.

At the conference, most of the other journalists who asked questions were focused on the war in Ukraine, and how the U.S. military was using space for “situational awareness,” a term from the military-industrial thesaurus that basically means information gathering. I was more interested in what Raymond had to say about satellites’ impact on the rest of us.

To military brass like Raymond, satellite constellations are both dangerous and wonderful.

For the past few months, I have been reporting and researching about artificial light — work that I am doing as a Knight Science Journalism Project Fellow. Part of the story includes an increasingly large swarm of satellites in low-Earth orbit. Some of these take pictures of Earth, which the military and private companies can use. Others will eventually transmit internet connectivity around the globe. SpaceX is launching and operating most of these satellites so far, but companies like Amazon and others will join them soon, which means the space environment close to Earth is going to be filled with stuff. Much of this stuff, crucially, can affect the natural night sky that astronomers study, and that we all inherited from our forebears. But the satellites will also provide modern communications networks for people who otherwise couldn’t access them, including Indigenous groups, people in the global south, and even members of my own family.

To military brass like Raymond, satellite constellations are both dangerous and wonderful. Dangerous because they can collide, causing an avalanche of space debris that could threaten other assets, like military satellites. Wonderful because some of these swarms, like Earth-imaging constellations, are inherently safer than single satellites. Taking out American reconnaissance birds (more Pentagon thesaurus) will be much harder for Russia and China when the birds’ views are distributed across multiple small spacecraft, instead of one or two big ones. Raymond used words like “resiliency” to describe what these satellites can do for the military. I came away from his discussion with a sense that I had picked up on something new and important, which I now need to share with others. That sense of discovery has animated my entire fellowship and my previous research on this subject, which dates back more than a decade, but it felt tangible again in that hotel ballroom.

I expected to learn about satellites and the military that week. I didn’t expect this symposium to make me feel like I had finally re-entered the world. The pandemic, the demands of parenting young children, and the all-consuming work of writing a book had kept me cloistered since early 2020. Even my KSJ fellowship is remote. I have been alone with my family and my ideas the vast majority of the time since I moved to this Space Force-Air Force-Army city. The Space Symposium snapped me out of my isolation reverie. There’s so much to learn from simply chatting with people, rather than setting up pre-scheduled Zoom interviews. There are so many questions I haven’t even thought to ask. It just took some bland hotel coffee and a gracious four-star general to make me remember that.

Rebecca Boyle is an award-winning science journalist and author based in Colorado Springs, Colorado. She is a frequent contributor to Scientific American, Quanta, and The New York Times, and is a contributing writer at The Atlantic.

by Julia Rosen / June 17th 2022 / Fellowship Life, Guest Posts

When Reporting Changes How You See the World

Editor’s Note: This is part of a series of dispatches from the Knight Science Journalism Program’s 2021-22 Project Fellows.

On a recent Sunday morning, my family went hiking at a favorite spot in the Columbia River Gorge. It lies east of the Cascade Mountains, where the dense, wet forests of Douglas Fir give way to open meadows and stands of white oak. The area is known for its wildflower displays, and when we visited, the blooms did not disappoint. Arrowleaf balsamroot — also known as Oregon sunflower — grew in riotous yellow bouquets, and stalks of purple lupine stood almost as tall as my toddler, who stopped often to smell the flowers.

We had hiked here before, and it looked much the same as I remembered. But I saw the landscape differently this time.

As a Knight Science Journalism Project Fellow, I’ve been researching the world’s grassy ecosystems and the mounting threats they face. I’ve read countless textbooks and scientific papers about grass. In April and early May, I spent two weeks touring grasslands with researchers and conservationists, absorbing as much of their knowledge as I could. In the process, I gained what felt like a new superpower: the ability to understand grassy landscapes just by looking at them.

As we set off from the trailhead, the path wound through widely spaced oak trees that I once might have described as a woodland. Now, I understood that it was actually a savanna. The difference lay in the understory, which was full of sun-loving grasses and wildflowers, not shade-tolerant ferns and shrubs. In other words, this wasn’t an open forest; it was a tree-filled grassland. Historically, the landscape probably burned often, likely with help from native peoples. That wouldn’t have been a problem for large oaks with thick bark. But decades of fire exclusion had allowed scraggly young trees to start crowding the trail.

Next, we crossed a small stream and climbed a blossom-choked hill that offered a wide view of the surrounding countryside. A meadowlark sang as we ate snacks and surveyed the scenery. Immediately, I noticed that the wildflowers that grew in such abundance on the hilltop were nowhere to be found in the grasslands below. And I knew this was likely because those areas, which were flatter and more accessible, had probably been plowed or grazed — or both.

Indeed, after I got home and did some Googling, I confirmed that arrowleaf balsamroot — an edible plant used by tribes around the West — is like many other perennial grassland flowers; it has deep roots that can penetrate up to nine feet below the ground, which makes it extremely tolerant to disturbances like fire and moderate grazing. But it cannot handle heavy grazing and rarely recolonizes fields after they’ve been tilled.

Arrowleaf balsamroot has a deep root that allows it to survive fire and moderate grazing. But it can be damaged by overgrazing or plowing. (Credit: Julia Rosen)

We descended the other side of the hill through more oak savanna, and emerged onto one of the flowerless fields we had seen from above. Gone were the slender, ankle-high grasses of the wildflower meadows, replaced by a monoculture of thigh-high grass, thick and sturdy and crowned by wheat-like heads of seed. I wondered if this was a hayfield or an abandoned pasture.

A few months earlier, I might not have thought much of the change in vegetation, if I’d noticed it at all. For most of my life, grass was just grass. That Sunday, however, I understood that the boundary reflected some profound difference in the histories of these two neighboring plots of land, and what was lost when one turned into the other.

The field ended as abruptly as it began, and we crossed back into the short grass and wildflowers as we made our way back to the car. By now, I had come to suspect that we were in what one of my sources would call an “old-growth grassland” — a grassy ecosystem that has existed long enough to develop the same sort of complexity and diversity that we cherish in old-growth forests. I was exhilarated by the realization because it made this place feel even more special (old-growth grasslands are rapidly disappearing) and because I felt like my reporting had made visible a new layer of reality that I hadn’t been able to see before.

Learning to decode the natural world is a big part of the reason I studied geology as a student; I’ve always been fascinated by the stories etched into our surroundings, available to anyone who learns how to read them. Continuing this education has been one of the great joys of journalism. It’s also one of the reasons I write.

I want to help readers see landscapes not as the sets upon which we act out our lives, but as repositories of the natural and human histories that brought us to our present moment. That’s particularly true when it comes to grasslands, which are easily neglected and widely misunderstood. If we can learn to appreciate them, however, they have much to offer — and plenty to teach us.

Julia Rosen is an independent journalist covering science and the environment from Portland, Oregon. Her work has appeared in The Atlantic, The New York Times, Scientific American, and many other publications.

by Emily Willingham / June 15th 2022 / Fellowship Life, Guest Posts

The Kids Are Not Alright

Editor’s Note: This is part of a series of dispatches from the Knight Science Journalism Program’s 2021-22 Project Fellows.

During the past nine months I have been working, as a Knight Science Journalism Project Fellow, on a book about adolescents and the maturation of the adolescent brain. A key element of the book is the voices of adolescents themselves. I’ve interviewed dozens of them, at times asking them personal questions about bullying, harassment, and their relationships. It has forced me to think long and hard about a question that doesn’t often come up in science reporting: How should we, as journalists, talk to young people?

In my interviews for this book, I’ve had to consider factors that my usual reporting doesn’t require. Among these factors is age. Adolescence spans ages 12 to the mid-twenties. I chose to home in on the middle part of this stage, which made one of my tasks a little easier: I knew that the questions would be sensitive, and I wanted to be able to explain the interview ground rules in an understandable way. Interviewing adolescents in their late teens and early twenties made that clarity more likely. It also meant that they had extensive experience with social media platforms, and peer pressure interactions, which are key themes of my book.

I also undertook preparation for sensitive interviewing practices. I knew I would need to clearly communicate to these young people the risks of talking with me, and we would need to have a plan in case they became upset. That preparation proved essential, because their stories often got way more intense than I’d expected. What young people harbor in their hearts and minds — and often do not share with adults — can be harrowing and sad.

The plan that I implemented had several elements. Because my interviewees were still developing their capacities to grasp long-term consequences, I determined beforehand not to identify any of them by features other than gender, age, and general geographical location. My reasoning was that these young people, faced with the permanency of the internet, have a lifelong inability to hide their past from the world. I didn’t want to share public information about them that could have unanticipated negative consequences down the line. It’s hard enough to predict these outcomes as a middle-aged professional with two decades of experience wrangling the commentariat. I wanted to avoid imposing that risk on the teens and young adults taking the time to speak with me and trusting me with their stories.

Another element of my plan was to make quite clear that they could “nope” out of any question, or even the entire interview, at any time, no questions asked and no judgment. I made sure to communicate at the beginning of each conversation that this choice was available at all times.

I also planned to avoid two pitfalls during our conversations. One was easy: Stay neutral about the interviewee’s life choices and plans. I don’t have a problem maintaining neutrality or being supportive about how other people live their lives, as long as nothing criminal is involved, so holding that line was easy. The tougher pitfall to avoid was my instinct — as the parent of three adolescents and a long-time educator — to want to help. It’s hard to listen to stories like these and not jump into problem-solving or advice-giving mode, especially with a three-plus decade age difference between my interviewees and me. But I determined beforehand to stick with a sincere “I am sorry” and “that must have been really hard for you” and other acknowledgments that didn’t extend into unasked-for-advice territory.

What young people harbor in their hearts and minds — and often do not share with adults — can be harrowing and sad.

Over the course of the interviews I subsequently conducted, I learned a couple of things I hadn’t expected.

First was the arc that the adolescents’ stories commonly reflected. One theme of my book is how the adolescent brain develops — especially the prefrontal cortex, the foundation of being an adult. As we shapeshift from child to adult, building out this region of the brain means building a toolkit of filters, planning, and problem-solving skills to support us as independent adults. I was fascinated to hear interviewee after interviewee relate how they’d shifted from uncontrolled social media exposure in early adolescence to a more measured use of the platforms as they aged, with their choices reflecting the unseen arc of a maturing prefrontal cortex.

Second was the excruciating experiences some of the people I talked to have endured. Regardless of background, demographics, country of birth, gender, or sex, almost all of them had unbelievable stories to tell about incidents with peers and adults. Most striking was that many of their experiences would have been almost impossible 20 years ago. Teens described unconsented sexting, propositions from older men (always men), sex abuse, and peer-to-peer bullying on a level and scale that were unimaginable before social media amplified and magnified these exposures.

Although I have more reporting left to do, the proposal for my book is near completion and ready for my agent’s critical eye. The book will include perspectives and insights from neuroscientists and other researchers who study adolescence, but much of it will take its contours from the stories my adolescent interviewees told me — stories shared out of trust that I would not violate our agreements or put them at risk for more exposure than they’ve had already. It’s fitting that their voices guide a book about tools for adolescents who are wayfinding their path to adulthood — and who sometimes just need us adults to get out of their way.

Emily Willingham is a science journalist and regular contributor to Scientific American. She is the author of “Phallacy: Life Lessons from the Animal Penis” (Avery, 2020) and “The Tailored Brain: From Ketamine, to Keto, to Companionship, A User’s Guide to Feeling Better and Thinking Smarter” (2021, Basic Books). She lives in the San Francisco Bay Area.

by Jessica Camille Aguirre / June 10th 2022 / Fellowship Life, Guest Posts

Tiny Worlds

Editor’s Note: This is part of a series of dispatches from the Knight Science Journalism Program’s 2021-22 Fellows.

In general, ecosystem science is a field science. Researchers go out into the world to gather samples, and to make observations about the dynamics of complicated systems. In a few, obscure, places around the planet, though, that relationship is flipped. Instead of observing the workings of nature, scientists try to recreate it – they become the architects and masters of their own, miniature worlds.

Perhaps the most famous of these projects was Biosphere 2, a man-made Earth ecosystem enclosed in a 3.2-acre glass and steel building in the middle of the desert. The building was tightly sealed from the planetary elements, since it was intended to be fully self-sufficient and regenerating. It contained six different kinds of ecosystems, including mangroves, a desert, a rainforest and a tiny ocean with a living coral reef. The facility, which now hosts visitors at the original campus outside of Tucson, Arizona, is verdant and thick with life; clutches of trees press against the glass ceiling and aromatic shrubs cover the rocky hillsides.

In contrast, most artificial ecosystems are compact and contained. The bulk of research in the field has been focused in space exploration. In order for astronauts to travel long distances from Earth, where resupply is impossible and any extra weight associated with bringing needed supplies is costly, they need life support systems that can regenerate. These systems have to also be as lightweight as possible, and they have to be subject to control and manipulation by the people who rely on them.

The most extreme of these is a project developed by the European Space Agency that has been operated for 30 years. Composed of a consortium of universities across Europe, the project has a pilot plant at a laboratory outside of Barcelona, where I travelled in April with support from an MIT Knight Science Journalism Project fellowship. I had read extensively about the plant, called Melissa (which stands for microecological life support system alternative), and started by a group of researchers led by Claude Chipaux who were inspired by lake ecology to create a closed system based on anoxic bacteria.

In order for astronauts to travel long distances from Earth, where resupply is impossible and any extra weight associated with bringing needed supplies is costly, they need life support systems that can regenerate.

Francesc Gòdia Casablancas, a chemical engineer at the Autonomous University of Barcelona, oversees the pilot plant. He met me on the university campus and we went to the cafeteria in his department to have a coffee and talk about the program. At three decades, it’s by far the longest operational bioregenerative life support system research project in the world – others, notably in Russia and China, are either defunct or still extremely new. The Melissa plant has taken a unique tact, focusing on incremental steps and decentralized research. “The idea was, what are the main functions in an ecological system on Earth? What do we need for a system to be self-sustainable in terms of organic waste degradation, biotransformation, oxygen production, CO2 capture, edible material production?” Casablancas told me. “What would be the simplest way to reproduce this magic?”

We walked through the long hallways of the university offices, passing the laboratories where researchers ambled by with trays full of beakers. The Melissa plant was down a side corridor, inside a few adjacent rooms. From the outside, it wasn’t much to look at – a tangle of steel vats and tubes, with an abundance of boxes and levers attached, it looked more like an oversized car engine than a lake. But as Casablancas began to explain the interplay between elements of the system, it began to seem like a new, different kind of magic. Like a miniature world, preparing for takeoff.

Jessica Camille Aguirre is a freelance magazine journalist whose writing on climate and extremes has appeared in The New York Times Magazine, Harper’s, Vanity Fair and n+1.