Questions raised about research by OpenAg Initiative at M.I.T. Media Lab | Hacker News Reader

Questions raised about research by OpenAg Initiative at M.I.T. Media Lab

nytimes.com

85 pointsAnon8456 comments

MIT is probably the only land grant campus that has no agriculture department.

That much is understandable, since ag-schools need to do actual literal field work on test plots, and that kind of acreage is hard to come by anywhere near Cambridge. But it also means there was nobody on campus to sanity check this OpenAg nonsense.

The land grant act specified the aims for the funded schools would be "agricultural and mechanical." I believe Massachusetts was the only state to divide the funds between separate agricultural (UMass) and mechanical (MIT) universities.

Too bad MIT MediaLab is under scrutiny now. I really hope their PubPub[1] project will not die. It is the only open source[2] modern scientific publishing platform, like Authorea or Overleaf.

[1] https://www.pubpub.org/

[2] https://github.com/pubpub

There's a tweet from some prominent agriculture researcher pointing out that the fabled Food Computer is in fact a rehash of agricultural techniques used circa 2000. Forgot her name but the remark struck me.

There should be a periodic reminder for tech-minded people to reread Ecclesiastes 1:9 every once in a while.

For those of us less well-versed in the Bible:

“The thing that hath been, it is that which shall be; and that which is done is that which shall be done: and there is no new thing under the sun.”

"All Of This Has Happened Before And All This Will Happen Again"

- related by the Lords of Kobol ?

Not to be confused with:

   PROPHECY DIVISION.

   PERFORM LoopBody VARYING Day

       DISPLAY "Same shit"

       MOVE Cow TO CowsHome(Ncows)

       ADD 1 TO NCows

     END-PERFORM.

  UNTIL AllCowsCameHome USING CowsHome

-- Lords of Cobol

The word is about, there's something evolving,
whatever may come, the world keeps revolving...
They say the next big thing is here,
that the revolution's near,
But to me it seems quite clear

thats it's all just a little bit of history repeating.

- The Propellerheads, History Repeating

Sure, and self-driving cars are just horses all over again right? And ancient Egyptians were already using tablets.

Self-driving cars aren't real. And ancient Egyptians were using tablets, I don't know if you chose this word on purpose.

When Ancient Egyptian tablets had been carved, the Babylons/Hitites couldn't remotely wipe them.

You're thinking of Dr. Sarah Taber, an ag tech and food safety consultant with lots of past exposure to startups as well. The relevant twitter thread: https://twitter.com/SarahTaber_bww/status/117189565787294105...

This is... not the clever revelation that she thinks it is. She worked with "room sized versions of thse back in 2001"? Well yes of course. I looked through all the documentation, etc back when this story broke and the whole original purpose of the Personal Food Computer project was to take those big room-sized setups and shrink them down to small, standardized, well-documented boxes that could be racked up and used to carry out a bunch of experimental runs in parallel quickly and cheaply. The whole damn point of the thing was that it was like those only smaller and easier to replicate. (And a few people did replicate them and grow plants in them.) I presume that was where the name came from; it was meant to be the PC to the metaphorical mainframes of the existing setups.

Part of Taber's point was that it's harder to scale up than down. This is solved technology, and their failure to make miniaturized versions work at all is a further indictment of their project as a harvester of investor funding rather than an agricultural game-changer.

Sounds pretty much like the Theranos pitch?

Her point is that this tech was outdated in 2001. Her point is that the people funding and supporting this research were mislead across multiple ways:

* misled about state of the existing tech

* outrageous and unsubstantiated claims of its performance

* literally faking results in front of investors and stakeholders

Further info Taber links to this account[1] from Babak Babakinejad, an MIT Media lab research scientist who describes an environment of research based on clear fabrication, and one in which the lies extended to the director colluding with MIT Health and Safety to actively mislead Environmental Protection Drinking Water Program officials about safety checks.

> What horrified me the most was that Mr Harper, as director and operator of the permit, actively worked with senior staff at MIT Health & Safety to mislead the Environmental Protection Drinking Water Program officials about his team’s activities.

The behaviour described in these threads demonstrates repeated, rampant, widespread violations of R&D integrity, without which MIT loses its capability to perform research and development at all.

It is not dramatic to say that if fabrications are allowed to to be called research, and a conspiracy of lies can dismantle safety checks, the whole lab is in dire jeopardy.

[1] https://twitter.com/flypolitics/status/1175099839236886528

Some small blame has to be on the people funding this, doesn’t sound like an elaborate fraud but a seeming utter lack of due diligence.

There's a small industry already making plant growth chambers.[1] There are little desktop ones, refrigerator sized ones, and bigger ones. You can get LED lighting, temperature, humidity, airflow, water, and C02 control, networked computer interfaces, and data collection software. Basic tool in ag R&D. They look a lot like the "food computer", but with better environmental controls and insulation.

Not usually used for production; once the desired conditions have been figured out, the production system can be simpler.

[1] http://www.conviron.com/

It, ah, looks like Conviron's smallest unit is still a little larger than a typical full-height fridge - and that one's really new, having only been announced in April this year. (Reverse osmosis system not included.) Doesn't look as well insulated as your typical fridge though. All the other ones are even bigger. They're not really comparable - or doing the same thing, given that the PFC was hydroponics-based.

Thanks for sharing this thread.

(Fwiw: I got curious about openag’s efforts/papers last week and made a comment here: https://news.ycombinator.com/item?id=20937105 )

https://twitter.com/SarahTaber_bww/status/117189565787294105...

In one set of articles OpenAg is a therenos grade scam with everything faked, all plants purchased instead of grown, etc.

In another set of articles, they're in trouble because the nitrate levels in their waste water are too high: https://www.wbur.org/edify/2019/09/20/open-agriculture-initi...

So which is it? A total sham or a polluter? These positions sound incompatible.

These don't appear to be incompatible or mutually exclusive claims. Like Theranos, it seems there was actual research, development, and small scale production work going on at OpenAg. Also like Theranos, it seems that work was nowhere near producing the results that were claimed by leadership, leading to falsifying results for the sake of publicity and funding. This means it's totally possible that the unsuccessful research they were doing was outputting wastewater containing excessive nitrogen levels while also misleading the public, press, and investors with faked demos.

Why not both?

It sounds like OpenAg engineers were trying various solutions for the device, but by the end of the effort, demo culture prevailed and the end-results were faked. At some point their staff decided to dump their concentrated nitrogen solution as waste water.

It's alleged that they made misleading claims, but it's not like the whole operation was a complete fabrication. There were plants and machines and whatnot, as can be seen in several articles on tis topic.

So it's plausible that they illegally dumped wastewater as part of whatever it is they're doing that isn't working as advertised.

Fair enough, though at some point it starts becoming a bit hard to distinguish between what OpenAg was doing and what a significant fraction of all demos do.

It's typical for some amount of 'rigging' to exist in demos, particularly because Murphy's law has an ugly way of rearing its head. At a very minimum, almost everyone goes over the demo many times and fix or avoid any issue that causes a failure or questions that can't be easily and decisively answered.

I really did get the impression from the first articles I read on OpenAG that the claim was that they were doing absolutely nothing and that it was 100% faked. The article on the nitrates discharge was surprising to me in light of that.

I'm perhaps a bit jaded also by seeing how much computer science / signal processing academic publication is essentially fake. "The desktop version doesn't really exist but we have a basement version doing stuff" seems like a fairly mild level of deception compared to many other things I've encountered.

OpenAg had several projects - a Personal Food Computer, which is what I believe has the allegations of not being performant as claimed and another project called the Personal Food Server, which is where I imagine such a high discharge of waste water would come from. In any case, organizations can be failing in many ways simultaneously.

"Happy families are all alike; every unhappy family is unhappy in its own way." - Leo Tolstoy

edit: I wrote 'image' instead of 'imagine'

One of the defining qualities of the media lab was smoke-and-mirror demos. I thought everyone understood this and was OK with it. There's a place in the world for concept art, to inspire people with what might be possible.

This is like "Questions Raised that TED Talks Oversimplify Complex Issues".

Rigged demos are the Media Lab's main output, and have been since the start. Shut it all down.

Do you have other examples we can look at?

The original goal of the Media Lab was to explore the impact of having access to far more computing power than you could possibly have access to today (for whatever is the current value of today). Not "more than most people have", but more than you could have that point in time. That's an important and valid subject to research (eg they did lots of work on understanding the impact of speaker-independent voice recognition long before we could really even do speaker-dependent voice recognition reliably) but almost by definition the only way you can study it is via rigged demos.

I would follow up by saying that from 2004-2010, while I was a researcher there, the emphasis was on mixing ideas from places as far flung as possible. There was no need to fake a demo, as the ideas were not impossible, just uncanny.

Once someone like e.g. Saul Griffith came up with the idea of Kinematic Replication [1], others could easily replicate it (or at least anyone with a laser cutter at that time). Same goes for my projects. [1.25] For goodness sake, the source code and build instructions are published so that people can replicate it. These ideas are not like Caleb Harper's claims of high productivity versions of existing processes. They're fresh formulations of concepts one might not ordinarily have had the notion and resources to combine. Just look at Hiroshi Ishii's work! [1.5] That stuff is not hard to do once you've seen it done once- but try to be the originator of some of that stuff. That's where the magic is!

Take a look at my pal Brian Whitman's work [2] which was the basis of Echo Nest and eventually Spotify's recommendation engines. These algorithms are published at conferences and in journals so anyone can go in and replicate them. Brian's work isn't remarkable because it's so unbelievable it can hardly be true - it's remarkable because it dares to combine machine learning with signal processing and social media. These were all new ideas at the time and and it was up in the air whether they could work. Now they're the bread and butter of many peoples' commercial and personal activities. Anyone saying these demos are rigged is revealing their own shallowing understanding of what takes place there.

I don't know much about how the culture of the lab changed after I left and Ito became the directory almost immediately after... I do know while I was there, it was actually a bit scrappy. The original "shop" in 2004 was...a 10x20 room with a single laser cutter, single waterjet cutter, a drill press, a Bridgeport-like mill and a CNC lathe. There were tools hanging over a tattered wooden workbench. When the second building was constructed around 2008(?), shop size expanded about 10x. It went from a four floor building to 6 story building with maybe 8x the space. Ito was a kind of cult figure. I felt out of the loop bc ppl talked about him like an angel but couldn't explain to me what he'd done.

[1] https://alumni.media.mit.edu/~saul/PhD/index.html

[1.25] https://web.archive.org/web/20100627183050/http://web.media....

[1.5] https://www.media.mit.edu/people/ishii/projects/

[2] https://alumni.media.mit.edu/~bwhitman/

You just perfectly described the problem with many universities nowadays. The way to advance your career in academia is to build a shiny new building and kick off an initiative or institute or something real you can put on your resume.

We have a lot of shiny new half empty buildings in all the colleges, and a lot of very well paid "administrators" trying to get their name on some project. Tuition and fundraising demands are skyrocketing decade after decade and there is very little to show in real benefits for students or researchers.

Compare the over-the-top media hype around the announcement of OLPC with the overpriced, underpowered netbooks they ultimately shipped.

1. This has nothing to do with whether demos where rigged or not.

2. OLPC is an education program, not a hardware program. netbooks don't have the educational software, the user-editable OS, etc. that netbooks had.

2.5 The netbooks came out after the OLPC's XO-1. They were a response to it. [0]

3. If you want to talk about hardware power, the Asus EEE had the 630 MHz Intel Celeron processor [1] where the XO-1 had the Geode LX700 [2] running at 667 MHz [3] https://en.wikipedia.org/wiki/OLPC_XO

4. The original Asus EEE netbook retailed for around $300, costlier than XO-1's $200. [4][2]

[0][1] https://en.wikipedia.org/wiki/Asus_Eee_PC

[2] https://en.wikipedia.org/wiki/OLPC_XO

[3] https://en.wikipedia.org/wiki/Geode_(processor)

[4] https://gigaom.com/2008/09/01/choosing-a-netbook-a-guide/

So maybe there was hype and maybe it was deserved because it was cheaper, faster, and earlier than the netbook and because it included an educational curriculum?

And your claim of rigged demos beyond this Caleb Harper dingbat remains standing on nothing but a piece of dirt.

You're correct that the OLPC came before and spawned the whole netbook thing.

Also, the OLPC's hybrid monochrome/color display that Mary Lou Jepsen designed was truly innovative, power saving, easily manufacturable, green electronics, and it was even quite efficient and crisply legible and under direct sunlight (requiring no backlight for the high resolution 200 dpi reflective grayscale LCD pixels, which could stay on while the CPU was asleep).

https://en.wikipedia.org/wiki/Mary_Lou_Jepsen

She's written honestly about the Epstein situation, too:

https://medium.com/@maryloujepsen

John Perry Barlow typing away on his OLPC:

https://imgur.com/gallery/ogE9y

Here's an excerpt from a technical summary I wrote up in November 2008 about the OLPC display, and the problems and successes of the project:

https://donhopkins.com/home/documents/OLPC-Notes.txt

[...]

    + Display

      + Designed by Mary Lou Jepsen.
      + 1200x900, 200 dpi, 6x4 inch (152.4x101.6 mm), 6 bits (262k colors).
      + Pixel size 0.127 mm, about 1 arc minute. 
      + You can always see 200 dpi grayscale, even in direct sunlight. 
        (reflective layer)
      + You can get color from the LED backlight. 
      + Colors wash out as the sunlight gets brighter.
      + The backlight uses power (though not nearly as much as cold cathode
        fluorescent backlight).
      + You can turn the backlight down or off to conserve power. 
      + Turning off the backlight tells screen to give slightly higher 
        resolution, to make reading comfortable. 

      + Display Design Goals: 
        + Maximize the number hours of ebook reading.
        + Minimize the power consumption.
        + Maximize the resolution and readability.
        + Mesh with how human perception works.
        + Display sharp high quality text, that's easy on the eyes 
          during the day (reflective) and night (backlight).

      + Critical design innovations:
        + Remove the subtractive color filters that absorb 85% of the light.
        + Replace them with plastic diffraction gratings and lenses, 
          stamped like DVDs. 
        + Much brighter display for a given amount of backlight.
        + Can be manufactured with existing technologies and processes. 
        + Uses efficient environmentally friendly LEDs,
          instead of fragile, expensive, high voltage, cold cathode 
          fluorescent lamp backlights.

      + Combination of two separate screens: sharing an LCD glass. 
      + One normal backlit screen.
      + Another normal reflective screen. 
      + LCD is 1200x900 square grid, with 64 gray levels (6 bits).
      + Off pixels transparent, on pixels opaque. 

      + Backlit screen shines through a color filter on the 1200x900 grid. 
      + Filter gives each pixel just one color: red, green, blue. 
      + Individual grayscale pixels behave like sub-pixels of a normal 
        backlit display. 

      + Reflective screen has reflector behind LCD grid. 
      + Room light passes through grayscale LCD and bounces off of back
        reflector. 
      + 1200x900 pixels depending on ambient outside light to display. 

      + The light the user sees comes from both sources (reflected outside 
        light plus filtered backlight).
      + Color filters use fresnel prisms to pass most light, instead of 
        color filters that absorb most light, wasting less energy. 

      + The amount of color and perceived resolution depends on backlight
        brightness and outside light level. 
      + The ambient light level of the room changes the perceived 
        resolution of the display.
      + In direct sunlight you see the reflective screen (exactly 1200x900, 
        200 dpi),
      + In a dark room you see the backlit screen (approx. 800x600 
        perceived, 133 dpi), 
      + In between you see both (approx. 1024x768 perceived).
      + The "official story" is "1200x900 mono resolution, 
        693x520 color resolution". 

      + Screen layers:
        + LED backlight.
        + 1200x900 grid of color filters (fresnel prisms).
        + Semi-reflective layer.
        + 1200x900 LCD.
        + Each pixel has a single color behind it.
        + Colors are arranged in a diagonal pattern.
        + Each pixel has:
          + Fixed hue (r, g or b).
          + 6 bits (64 gray levels) of luminance.
          + Chrominance depends on relative strength of room light 
            to backlight.

      + DCON screen driver chip
        + The screen can stay on while the processor is turned off. 
        + Automatically interpolates ("swizzles") lower resolution color 
          pixels, so the unusual screen format is invisible to software. 
        + Looks just like a regular 1200x900 color framebuffer to software. 
        + DCON has different modes:
          + Monochrome.
          + Color swizzled antialiased.
          + Color swizzled not antialiased.
          + Video pass through.

      + Power saving techniques
        + Uses low power LEDs instead of high voltage cold cathode 
          fluorescent backlights. 
        + Turns off processor while leaving the screen and wireless 
          network on. 
        + Uses fresnel prisms to split most white light into primary 
          colors, instead of color filters to throw away most light. 
        + Turn down refresh rate.
        + Dynamically turn off sections of the screen.

    + Green Electronics

      + Order of magnitude less power consumption. 
      + Designed to use as many environmentally friendly components 
        as possible. 
      + Fully compliant with EU's Restriction of Hazardous Substances 
        Directive (RoHS). 
      + One of eight laptops to receive EPEAT's Gold rating for 
        environmental performance.

[...]

      + Successes
        + Two good things have happened to the project in the broader sense:

        1) Spawned this whole netbook thing. 
           + Nobody was doing that before.
           + Now 15 different major companies are doing the same thing. 
           + Finally people are exploring the low cost low power end 
             of the laptop spectrum. 
           + OLPC is still unique in that it was ruggedly designed to be
             dropped and survive. 
           + Other companies have not figured that out yet, and are still
             making cheap low quality laptops.

        2) The free software community is widely in favor of the project. 
           The deal with Microsoft got bad press, and most people in the
           free software community mistakenly think the free software part
           is over, and OLPC is now shipping laptops with Windows. 
           But actually they are not.

I'll repeat and extend what I've said in several other Media Lab posts. The lab has over its lifetime been a place where the creative arts, computer science, and popular culture have come together. Have there been demos of ideas that didn't, or possibly even couldn't, work? Yes. Has the placed helped shape the connected world we live in? Yes. Has real stuff come out of it? Yes.

Semour Papert and Mitchel Resnick are luminaries who helped define the landscape of computation and play for kids. E Ink came out of the lab. Todd Machover has been influential in shaping technology in music (Guitar Hero was developed by some of his students). Professors Muriel Cooper and John Maeda have had similar impact in technology and graphics arts and design.

Steve Benton's Spatial Imaging Group and its student diaspora had a huge influence on the holography and 3D display world. We developed full color reflection holography, the first holographic video system, and the first image predistortion algorithms for light fields. Our 1 meter sq. hologram was at the time the world's largest computer generated 3D display. Unfortunately, Steve died pre-Google, and his contributions are not fully recognized.

I was at the lab a long time ago (from its opening in 1985 until 1997). What made it unique at the time was breaking down the barriers between computation, technology, creative arts, and popular culture in a way that seems so natural now. On a campus where hundreds of other PhD students seemed to be working on amazing research on very detailed topics, we were building pieces of the future and demoing them daily to industry and lay people alike. That wasn't considered hard science at a place like MIT, but the ultimate impact on society was huge.

One other area of huge but unsung impact was speech technology. Chris Schmandt's Speech Research Group group envisioned and developed audio assistants that paved the way for Siri and Alexa. His students and research have also helped shape an entire field.

As for Nicolas Negroponte, he started the lab understanding that corporations were going to drive the computation and Internet revolution, while the rest of MIT was primarily government/defense funded. I have no doubt that that insight accelerated bringing technologies to market that we now consider ubiquitous.

The current funding controversy was long after my time, as is the current work of the lab. There's just plenty of innovation in the history of the lab.

I was there during the '00s. I'll admit my hostile attitude is partly sour grapes on my part, since I deservedly flunked out (otherwise known as getting a master's). But nearly all the bad press the lab has been getting lately rings true to me. Thinking back to how little of what I saw there turned into any kind of practical advances or useful products in the years since...what was the point? And of course I'm horrified at the source of so much of the funding.

But then I've also been souring on the value of academia and research in general. "Publish or perish" leads to a lot of garbage getting published, as increasingly large amounts of money are thrown at increasingly small results. There's just not enough scientific truth to go around. The Media Lab, being particularly prominent and press-friendly, has this worse than many other places, but it's a problem everywhere.

>There's just not enough scientific truth to go around.

There's plenty of truth, just not enough resources. It's learned behavior, the scrounging and sensationalization, that has developed in response to larger forces driving societal investment away from academia.

I bet many people at MediaLab knew what was going on in OpenAg. Research world is small and these are the kind of things people cannot help but internally talk about. We now live in the world of instant popularity and power. Once you have good affiliation and can make big claims, you can count on getting on late night shows, TED, 60 minutes, NYT etc. Corrupt people fully leverage this media circus to become a force to get new funding, make org wide decisions, sit on boards and committees. I bet there are few other "Theranos" at MediaLab and everyone knows who they are. The question is not that whole lab was corrupt but the fact that lab culture allowed and even perhaps embraced these corrupt groups to exist and grow.

> Semour Papert and Mitchel Resnick are luminaries who helped define the landscape of computation and play for kids.

Have Papert or Resnick done anything in the field of compuation and play for kids that stands up to critical empirical evaluation?

"My point, however, is more mundane: we have created a demo literally from smoke and mirrors, and the Corporate World bought it." ... "Dance for me, Corporate America! I'm SHIT-HOT!" -Hunter S. Negroponte

https://news.ycombinator.com/item?id=20905450

Anyone wanna plagiarize the article so I can read it?

Both of my mobile browsers wont let me, and they also detect private mode

Rough copy/paste

https://pastebin.com/iCtipMin

Apologies for any ads that got in.

Reply on news.ycombinator.com