MIT Media Lab Scientist Used Syrian Refugees to Tout Food Computers That Didn't Work

"So we just put these last week in a Syrian refugee camp in Amman, Jordan," Caleb Harper of MIT's Media Lab told an audience at the Georgia Technology Summit in late March 2017.

He was referring to machines developed by his Open Agriculture Initiative (OpenAg) at the Media Lab, where he is a principal research scientist. The machines had been delivered to a United Nations World Food Programme project that aimed to give refugees in the Azraq camp-located in the Jordanian desert 90 kilometers from the Syrian border-the means to grow their own food, right inside the camp.

The vehicle for this agricultural miracle is called a personal food computer (PFC): an enclosed chamber the size of a dorm-room refrigerator loaded with LEDs, sensors, pumps, fans, control electronics, and a hydroponic tray for growing plants. PFCs are programmed to control light, humidity, and other parameters within the chamber to create the perfect conditions for growing a variety of plants. It's a simple yet potentially revolutionary idea: a portable box that can grow practically any kind of plant just by downloading a recipe and planting some seeds.

The refugees fleeing war in Syria, leaving their homes, loved ones and possessions behind, had no idea where or when they would leave this temporary desert encampment or how they would make do while they were there. But what the refugees really needed, Harper contended, was "to be connected to other growers to share knowledge." He added: "So super proud that that's happening."

On its face, the project sounds like one of the most ambitious and altruistic uses of high-tech agriculture you could imagine. In his talk in Georgia and presentations elsewhere as recently as this year, Harper enthusiastically conveyed a vision for the PFC that mimics how regular digital computing is scaled: PFCs would find a home in classrooms and home kitchens; food-computer "servers" would be housed in shipping containers to supply, say, a restaurant; and data center-scale vertical farms would feed entire cities.

As the name of the OpenAg initiative suggests, the food computer's hardware and software are entirely open source-that is, the equipment specs and code are available free to anyone with the desire to experiment with indoor agriculture. "Nerd farmers," the hashtaggable moniker given to members of the OpenAg maker community, build their own machines and then test their "recipes"-consisting of an array of controlled environmental parameters such as nutrient mix, temperature, carbon dioxide and pH levels, and light color and intensity. The recipe's purpose is to arrive at a specific expression of a given plant's phenome, which is an organism's physical and biochemical traits expressed in response to the interaction of its genes and environment. Nerd farmers share their experiences via the OpenAg community forum and wiki, and can even upload their recipes to a Github repository, allowing others to replicate that exact plant phenome in their own machines.

Launched in 2015, OpenAg differed from other indoor farming efforts in both its ambition and its scope. While the operators of urban indoor farms are careful to locate them in areas that have access to water, electricity, and cheap real estate, often using proprietary software and equipment, open-source food computers could be built by anyone and would be deployable virtually anywhere. Data from food computers all over the world would be fed to machine learning algorithms to optimize recipes and help people grow, say, the most flavorful basil (the subject of this peer-reviewed PLoS-One paper authored by Harper et al.) or replicate an Aleppo pepper grown in Syria in a food computer in the Jordanian desert.

It's a nice idea-if your food computer works.

But the situation on the ground never matched the fantastic claims that Harper made about the WFP project in public appearances during the spring of 2017 and in briefings for corporate patrons of the Media Lab in the spring and fall of 2017. Harper and a colleague also cited the personal food computer's successful deployment in the Azraq camp in emails to potential partners and patrons for the Open Agriculture Initiative and for Fenome Inc., a spin-off company that Harper founded in 2016.

Even as Harper took the stage in Georgia, it was clear to those working with the food computer at the World Food Programme (WFP) and at Fenome that the project wasn't progressing as the team had hoped. Indeed, in September 2017, the WFP project officially ended without any of the machines having completed a single grow cycle, according to the official in charge of the project. The WFP's personal food computers weren't even deployed at the Azraq camp, home to some 35,000 Syrian refugees, but rather at a facility run by Jordan's National Center for Agricultural Research and Extension, in Mafraq, an hour's drive from Azraq.

Harper did not respond to detailed questions about the WFP project sent to him by IEEE Spectrum for this article.

Photo: WFP Innovation Accelerator May 2017 photograph of two food computers at the World Food Programme office in Amman, Jordan.

Recently, the OpenAg initiative has come under scrutiny following the departure in September of Media Lab director Joichi Ito. He was a champion of the project, which started during his tenure and seemed to exemplify his "deploy or die" approach. (In a 2014 TED talk, Ito announced he was changing the lab's motto from "demo or die" to "deploy or die," focusing researchers' efforts on real-world implementations of the technologies they were developing.) MIT is now investigating OpenAg, following allegations that staff were told to demonstrate food computers with plants that were not actually grown in them. Business Insider and the Chronicle of Higher Education first reported these allegations.

Perhaps the only unqualified success of OpenAg was Harper's ability to sell his idea. His first big public unveiling of the food computer came in a 2015 TED talk that has been viewed more than 1.8 million times. Audiences and the press alike swooned. Glowing reports about the food computer, including one in Spectrum, quickly followed, and continued right up until the most recent revelations. And Harper helped raise the capital to start up his OpenAg spin-off, Fenome.

Last month, The New York Times reported that four former researchers connected to OpenAg have complained about some of the claims Harper makes in his talks, including that the average age of an apple in a U.S. grocery store is 11 months sometimes and 14 months other times, statements refuted by a U.S. Department of Agriculture official in an email reviewed by the Times.

It's one thing to get an incidental fact wrong. It's quite another to repeatedly state that refugees were benefiting directly from food computers and enjoying a taste of home, when they were doing no such thing.

The WFP project started off with the best intentions. According to Nina Schroeder, Head of Scale Up Enablement at the WFP Innovation Accelerator and the World Food Programme official in charge of the Jordanian hydroponics project in 2017, the long-term goal of the project was indeed to deploy food computers at refugee camps. "First we wanted to come up with a concept that we could bring to a larger scale that actually makes sense to deploy. For the early research phase, it wouldn't have made sense to deploy it inside the refugee camp."

As Schroeder described it, the pilot program would let researchers evaluate the technology and determine if it was appropriate to install PFCs at the camp. If everything went well with the pilot, then the Azraq camp would receive the food computers.

The project launched at the end of January 2017, when a team from Fenome went to Jordan to assemble and install the food computers. At the time the company was based in Salt Lake City, with a staff of 17 there plus two employees in Boston and one in Seattle. Four food computers were placed at the WFP's office in the Jordanian capital of Amman and six at the National Center for Agricultural Research and Extension (NCARE) facility at Al-Khalydeha Salinity Research Station in Mafraq, about a one-hour drive from the Azraq refugee camp.

The plan was that after installation, the project would be monitored remotely from Utah via the Internet and by three NCARE staff on site in Jordan. The NCARE experiments focused on testing the technology using local water and changing the light spectrum of the food computers' LEDs and the nutrient mix of the hydroponic solution, according to a person close to the project who spoke with Spectrum anonymously for fear of retaliation. Plants tested included cucumbers, basil, and baby lettuces. The source confirmed that Fenome's team communicated regularly by phone with their Jordanian counterparts.

Photo: Joi Ito/Flickr (CC by 2.0) November 2014 photograph of Caleb Harper (left) and Media Lab director of operations and strategy and Fenome board member Barak Berkowitz, taken by Joi Ito.

According to corporate filings and internal Fenome documents obtained by Spectrum, Harper had helped raise $4 million for his startup, where he was executive chairman and a director. Barak Berkowitz, director of operations and strategy at the Media Lab and Mitchell Baker, chairwoman of Mozilla, also served on the board. Other directors included representatives of two venture capital firms that funded Fenome: Lucas Mann from Campbell Soup's VC arm Acre Venture Partners and Ignacio Martinez of Flagship Pioneering, founded by Noubar Afeyan, who is also a member of the MIT Corporation, as the school's board of trustees is known.

Martinez declined to comment on the WFP project or Fenome. Berkowitz, Baker, and Mann did not respond to questions posed by Spectrum.

But despite all this money and brainpower, things soon went awry in Jordan. Schroeder, in a phone interview, told Spectrum that the conditions at the NCARE site were harsh, with a very dry desert climate and high indoor temperatures. The power frequently failed, which shut down the building's air conditioning and the food computers' LEDs. When the air conditioning conked out, it sometimes reached 45 C (113 F) inside the lab.

Worse, the Wi-Fi was unreliable. A Wi-Fi connection was necessary to remotely monitor some of the parameters inside the grow chambers, which were equipped with cameras and sensors that measured temperature, humidity, and pH levels. Whenever a food computer went down, it had to be connected to Wi-Fi so that the remote team could reboot it. The software was still quite buggy, so not all features could be controlled locally at the NCARE facility. The Fenome team returned a month after the initial deployment to modify the boxes and some functions, and to allow the machines to be rebooted locally, according to our source close to the project.

But while Fenome might have solved some problems, others cropped up, according to Schroeder. Algae grew inside the containers, possibly because of low-quality water and light shining through the food computers' clear acrylic access doors. The doors also deformed due to the heat, creating gaps that let ambient air into the grow chamber and contaminated what is supposed to be a controlled environment.

In all, the Fenome support team visited NCARE four times to set up the food computers, train local teams, and adjust the personal food computers. The last visit was in May 2017.

In late April, just a few weeks after OpenAg Inc. officially changed its name to Fenome, Inc., 15 of its 17 employees in Utah were dismissed. In the fall of 2017, the company left Utah and relocated to offices at its VC partner Flagship Pioneering, in Cambridge, Mass.

"When they closed down the Utah office, that made it very difficult to continue the experiments that were going on," says Schroeder. The WFP officially ended the Jordanian project in September 2017. Not a single grow cycle was successfully completed, Schroeder says.

"The food computer we tested there wasn't ready for our purpose, and it was still in the development stage," Schroeder says. Her team is now deploying lower tech, locally adapted hydroponic systems to food-insecure communities in Algeria, Chad, Jordan, Kenya, Namibia, Peru, and Sudan.

The concept of the food computer "is so attractive that you have the possibility to grow locally," she says. "But you need to have the right kind of environment. That food computer version was too early."

While it may have been the most high profile, the World Food Programme wasn't the only Fenome partner left high and dry.

Photo: Charisa Moore Charisa Moore (right) and her students experimented with three food computers.

In the fall of 2016, Charisa Moore, a biology teacher at Bainbridge Island High School in Washington state, watched a recording of Harper's TED talk. The food computer sounded like just what Moore had been looking for to beef up her curriculum with content centered on ecology. Moore called Harper.

She says Harper told her he could talk about what OpenAg had done in putting food computers into Boston-area schools but warned her that they didn't work in a lot of the schools where they were deployed.

"Well, I can make it work!" Moore told Harper. Harper invited Moore, another teacher, and a star student to MIT for a week to learn how to build, program, and troubleshoot the food computer and experiment with plant recipes. Fenome would provide Moore's school with the hardware, help her and her students build the units on-site, and support them free of charge.

When she got home in late April 2017, Moore and her team decided to give a TED-esque talk themselves to about 400 people in the Bainbridge community about the project they were about to embark on with the help of Fenome.

"We did basically Caleb's presentation using his Fenome team. And then that week we built the computers."

Students and teachers started running experiments with the food computers. The food computer cameras and sensors sent data to Fenome in Utah, and the Utah team communicated with Moore on what they saw happening in the machines.

"And then it kind of got really weird," Moore recalls. "We started not hearing very much. We used [the food computers] through the summer. Starting the next [school] year, we started to hear Fenome was going to go out of business. So that team was not able to then really troubleshoot any of our stuff."

Without tech support from Fenome, software maintenance proved difficult. Moore struggled to push updated software that had been published on Github to the machines. "It's very complicated," she told Spectrum. "This is way beyond my expertise. I can only barely code in Python."

Hardware bugs were even more difficult to fix. "The equipment is really not sustainable," Moore says. "It corrodes. You have a cooling unit on it, the Freon comes out, it freezes-it just becomes messy. So to clean it, you have to go and order the stuff and replace those items. And good luck finding them."

Her team did come up with a solution for one glaring, design-for-demo's-sake specification: the food computer's clear acrylic door, which let in ambient light and contributed to the algae problems in Jordan.

"The thing about the food computer that sort of didn't make a lot of sense to me was that it's open.... It's not controlled," says Moore. At her students' urging, she went to Home Depot and bought some silver wrapping and clad the chassis with it to shut out unwanted light.

Photos: Charisa Moore The students modified the food computer by wrapping the door. They added a flap on the small rectangle to keep all light out.

Moore says that she and her students continued to experiment with their food computers, uploading plant recipes to the OpenAg open source forum. They also set up an experiment to see which equipment grew microgreens more effectively: a food computer or a basic UV light bank shining down on plants potted in soil. Moore's team found that the conventional indoor setup grew microgreens at four inches per week-twice the rate of the food computer.

Moore concluded that the food computers "are pretty much not usable, because they just are not user friendly. They're too hard to troubleshoot. Any Joe could not just walk up and figure out how to do it. You couldn't market that to put in your pantry at home unless you knew how to do all that stuff."

Moore found herself with three food computers on her hands. She gave the "most unusable" one to a student, who took it home and converted the box "into a kind of a simplistic one with [manual controls] instead of electronic ones." He used it to earn a Boy Scout merit badge.

Even as Fenome and its partners were struggling, Harper continued to enchant audiences with his tales of nerd farmers around the world. Harper, who holds a master's degree in architecture from MIT and is a member of the World Economic Forum and a National Geographic Explorer, managed to parlay the exposure from his TED talk into a lucrative side gig as a speaker. He earns $20,000 to $30,000 per talk, according to his agent's website. He also used a version of the TED talk at a pitch meeting with investors in the summer of 2016 to help get his company its first series of funding according to an email obtained by Spectrum.

In May 2017, Harper repurposed his Georgia Technology Summit talk for a Red Hat conference. He again spoke about how a refugee camp in Jordan was using the food computers. He also described what the refugees grew and the significance they attached to the machines: "We didn't tell them what to grow. They decided to grow things that they missed from home. Things that they can't get any more."

"The food computer," he said, "became a cultural object more than just a manufacturing object."

Meanwhile, Harper's startup was laying off staff and planning its relocation to Massachusetts. Just days after the Red Hat appearance, Harper posted about Fenome to the OpenAg Community Forum: "hey guys-the startup (fenome) in its infancy has had a couple gaffes (oops) and obv communications is one of them." He explained that the Fenome team was working to fix bugs and upgrade these "crazy expensive and not fully functional" "1st run prototypes" so that the company could start selling PFC kits. He told the forum that "after some development we all think its [sic] better to be based in Cambridge and is in the process of moving."

Trouble at his startup did not derail Harper's traveling show.

Less than a month after Red Hat, he dusted off his talk and delivered it at the EAT Stockholm Food Forum. He repeated his claim that the food computers at the Azraq refugee camp had created much more than mere plants:

"We've deployed in the world with the World Food Programme in Amman in a Syrian refugee camp. We did not tell them what to grow. Turns out they wanted to grow things from home. It became a cultural object for them. They missed the flavor of the place that they were from and that creates their culture and creates happiness for them."

Harper's story about Azraq evolved further in an interview earlier this year with science journalist Miles O'Brien at a Purdue University event on 26 February. This time, he revealed how St. John's Wort plants had been grown by a "person at the camp that happened to be a Ph.D. on St. John's Wort." Harper claimed that the person started a business selling the medicinal plant to treat a population "rife with depression":

Besides these public claims posted to YouTube, documents obtained by Spectrum reveal that Harper and at least one associate also misrepresented the World Food Programme project in email correspondence with potential funders and partners.

In a February 2017 email chain that included Nest cofounder and iPod coinventor Tony Fadell, Harper and his assistant tried to arrange a meeting with Fadell, now principal at Future Shape LLC, an investment and advisory firm based in Paris. Harper sent links to a couple of blog posts, one from 2016 about his lab and another about the "2017 expansion of our ecosystem with a nonprofit and a venture." He ended his 14 February 2017 email with "Btw we just deployed food computers to a Syrian refugee camp in Jordanon [sic] a contract with the UN. Pretty cool. C".

And about five months after the project at NCARE had ended, OpenAg was in talks with a group at Google about supporting research at OpenAg, according to another email chain obtained by Spectrum. In an email dated 30 January 2018, Google's Jess Holbrook, senior staff UX researcher and UX manager, asked several questions regarding food computers, including "Has anyone picked up the design and adapted it to specific use cases like edu, refugee groups (I know you mentioned Jordan), etc."

Hildreth England, the OpenAg Initiative assistant director at the time and currently codirector of the Media Lab's PlusMinus program, answered the next day, "...yes, the PFC v2.0 was deployed in a Syrian refugee camp with the World Food Program." England declined Spectrum's request to comment, citing "an open inquiry being led by MIT's Office of the VP for Research."

Around the same time as the exchange between England and Holbrook, Dr. Babak Babakinejad, then the lead researcher for OpenAg, was testing a food server being set up in a shipping container in Middleton, Mass., at MIT Bates Research and Engineering Center. Babakinejad told Spectrum that he documented several problems with the equipment, including differences in temperatures in various areas inside the food server, in what is supposed to be a controlled environment, and a lack of control over carbon dioxide levels, humidity, and temperature. He told Spectrum that he had reported these issues to the OpenAg team.

Photo: Babak Babakinejad Food server set up in a shipping container at MIT Bates Research and Engineering Center.

Babakinejad showed Spectrum an email he sent on 16 April 2018 to officials with MIT Environment, Health and Safety to report that OpenAg was discharging nutrient solutions beyond state-permitted limits, a controversy that was examined last month in a joint report by ProPublica and WBUR. Babakinejad also took his concerns about OpenAg and Harper to Media Lab director Ito.

In an email to Ito on 5 May 2018, Babakinejad stated that Harper was making claims in public talks about "implementations of image processing, microbiome dosing, creating different climates and collecting credible data from bots across the world that are not true."

In addition, Babakinejad wrote, "He [Harper] takes credit for deployment of PFC's to schools and internationally including a refugee camp in Amman despite the fact that they have never been validated, tested for functionality and up to now we could never make it work i.e. to grow anything consistently, for an experiment beyond prototyping stage."

Ito responded and asked Babakinejad if he could share these concerns with Harper. That's the last Babakinejad says he heard from Ito on the matter. Within a month, Babakinejad had taken a leave of absence. He officially left the OpenAg project in September 2018. Two months later, Harper was promoted to principal research scientist at the Media Lab, a position that as of this writing, he still holds.

If you have any information about MIT's Media Lab or its Open Agriculture Initiative (OpenAg), you can contact Harry Goldstein at h.goldstein@ieee.org or on Twitter (DMs open, ask for Signal number).