Shared publicly  - 
 
Just got around to reading +Thomas Goetz 's interesting piece on feedback loops in Wired. http://www.wired.com/magazine/2011/06/ff_feedbackloop/

Thomas focuses on psychological loops, but as he mentions, loops are everywhere in nature. They do seem like a pretty universal solution to life in a complicated world. Reading the piece, I was reminded of the hideously complicated feedback loops that E. coli (and all other cells) use to stay stable in an unstable world, which I learned about while writing Microcosm. One of the scientists who did a lot of E. coli work, John Doyle at Caltech, started out doing control theory for airplanes, got into biology in a big way, and then started applying the same feedback loops to controlling the Internet. (I wrote about him here in Discover a few years ago: http://discovermagazine.com/2007/nov/this-man-wants-to-control-the-internet )
21
14
Carl Zimmer's profile photoPeter Monnerjahn's profile photoJane Shevtsov's profile photoLisa Hamilton's profile photo
28 comments
 
Interesting! Did you read Hofstadter's "I Am A Strange Loop"? If consciousness itself is a type of feedback loop, then are these devices effective because they become a kind of augmented consciousness?
 
I haven't read Hofstadter's book, but there's no question that the brain depends on feedback on many scales, from connections between individual neurons to larger circuits between brain regions. I'm not sure what it means for consciousness itself to be a feedback loop, but consciousness certainly depends on feedback loops.

One thing to add is that--as any engineer will tell you--feedback loops are not simple. They can be positive and negative, and when there are several nodes in a loop, the pattern can get wild. Evolution has fine-tuned feedback loops so that they can respond to a wide range of inputs with useful outputs. For example, here's the feedback loop E. coli has evolved to decide when it's time to stop building a tail: http://bit.ly/qepQah
 
Funny seeing people rediscover cybernetics 70 years later....
 
+Brad Johnson Brad--Actually, cybernetics and gene regulation emerged at about the same time in the mid-1900s, and there was at least a little useful cross-talk that I know of. But for a long time gene regulation studies were really crude: REPRESSOR ANGRY! REPRESSOR TURN OFF LAC GENE! Nowadays, diagrams of gene regulation start giving microprocessors a run for their money, and so control theory etc. starts to look really useful again.
 
The first cybernetics papers in the 1940s were on the nervous system; von Neumann was moving to study genetics when he died right as the field was developing (1957). If he or Norbert Weiner (died 1964) had lived an extra decade I suspect the field of gene regulation (operon was only coined in 1960) would have developed differently. Turing was another polymath who was moving into studying genetics when he died (1954).

This may place too much weight on the Great Man version of history, but they really did have an outsize influence on other fields.

An interesting question is what legacy these men left, why (it seems) they failed to leave much of an impression on biology / genetics.
 
When I was writing Microcosm, I thought that intersection of cybernetics and molecular biology in the 1950s would be the coolest part of my research. I was amazed to come up so dry. I really don't know why Wiener and von Neumann didn't make more of a mark on people like Monod and Jacob. In hindsight, it seems so obvious.
 
Where did the molecular biologists come from? Did they come from the chemists, or the field biologists, the evolutionary biologists, medical field? I'm guessing it's mainly a matter of the cultural divide. Were there any major figures in molecular biology at MIT / Princeton?
 
Actually, plenty of them came from physics (think, Max Delbruck).
 
The E. coli loop is elegant and fascinating; thanks for sharing. Feedback loops truly are amazing; I wonder what the very first evolutionarily successful feedback loop was?

(My friend says, "the Big Bang, ha!", but that's another issue.)

On the Hofstadter point: in this second book, he attempts to explain the existence of consciousness, of "self", as a property of that special form of feedback loop, the Godel strange loop. This book and its theories (along with his first, "Godel, Escher, Bach") are far enough beyond my mathematical capabilities that I absorbed only the rougher outlines, but it suffices to say that Hofstadter believes that some awfully interesting properties can emerge from special feedback loops, one of them being consciousness itself.

To my knowledge, no one has attempted to go any further with his theories, i.e. back them with neuroscience. Worthwhile reading, nonetheless!
 
Maybe the explicit record wasn't there in the 1950s, but when I read Le Hasard et la Nécessité, as an EE and math student in the early 70s, I felt that the authors knew how to think in control and digital logic terms and how to write for people like me.
 
+Onuralp Söylemez +Laura Walton That is indeed a landmark paper--wrote about it in Microcosm. I agree that it may not be the first loop, just a good one rediscovered again and again.
 
+Fernando Pereira I think those biologists were thinking in digital logic terms and just didn't know it. Or at least didn't write about it! But someone tells me that James Gleick covers this ground in The Information--I'll have to check it out and see if he turned up something I missed.
 
Looks like the Macy Conferences tried to make this happen...again, maybe a little too early. Von Neumann's cellular automata certainly are in the right vein... 1952 Macy Conference: The relation of cybernetics at the microlevel to biochemical and cellular processes

[big gap]

In 2008 the 4th edition of the Willi-Hans Steeb's nonlinear workbook added genetic algorithms and gene expression to its list of cellular automata, neural networks, and fractals...
Nonlinear Workbook: Chaos, Fractals, Cellular Automata, Neural Networks, Genetic Algorithms, Gene Expression
 
Part of the "missed opportunity" in the 50s is an illusion of hindsight. For all their genius, Turing, Wiener, and Von Neumann, and even more their followers, thought in terms of general principles, not in terms of the "hacks on top of hacks" of evolved biological control systems. We need to remember that they knew nothing about the genetic code let alone transcription regulation, post-transcriptional control, or epigenetics. Nor did they know how computing would go: not cellular automata or diffusion processes, but very complex networks of specialized processes (subroutines). In the end, there are better analogies between regulatory networks and a large software code base than with general frameworks like cellular automata.
 
Carl -- you mention complex diagrams of feedback loops in gene regulation. Can you recommend a single (or a ocuple) esp good stories/explainers on those loops?
 
As a statistician developing tools for analysing gene expression data and other "omics" data I have not been able to contribute to the use of control theory and other "deep knowledge" based modelling in bioinformatics. I have been struggling to keep up with the pace of innovations in data collection and annotation. The "boring" part of the modelling (such as dealing with measurement errors, and mapping new semantic model to old ones in order to use the two simultaneously) is complicated enough so you barely have time to implement something equivalent to a t-test before the kind of data you are modelling become outdated. Also the size of the data sets and the quality has until recently (and, I believe, for many projects still) not made complex modelling feasible - you want to test one null hypothesis for each of 30000 genes with a sample size of 10 measurements per gene, obviously this creates loads of false positives. So more complex modelling with multi-dimensional effect sizes is not feasible. I realize this is a bad excuse for not eliciting the biologist's knowledge in a way that would lead to simple (few unknown parameters) yet advanced (modelling non-linear effects in a biologically plausible way). I guess that biologists and statisticians who are able to speak each other's language are far between.
 
There was a lot of discussion of feedback loops in the early work in circadian biology - Pittendrigh, Goodwin and Winfree all discussed it in their early work (1960s, some even before) (see "Origin of timing mechanisms." subheading in http://scienceblogs.com/clock/2006/07/clocktutorial_3b_whence_clocks.php for a brief recap). The 1960 Cold Spring Harbor meeting on biological rhythms was full of models incorporating feedback loops, and the discovery of first clock genes in the 1980s immediately spawned mathematical models of their regulation in terms of feedback loops. Discoveries of second clock genes in same organisms in the 1990s (e.g., TIM as a partner to PER in Drosophila) spawned a whole industry of math modeling as feedback loops. The infatuation with molecular feedback loops actually hindered studies of feedback loops that also incorporate other cellular components (stuff in the cytosol, membranes, etc. stuff that was discussed a lot in the 1960s - see http://www.scientificamerican.com/blog/post.cfm?id=circadian-clock-without-dna--histor-2011-02-11 ) and of course the study at higher levels of organization in tissues, organ systems and organisms was for decades all about feedback loops: how SCN, pineal, retina, ovary etc. drive each other's circadian cycles in feedback loops (some positive some negative loops).
 
Retailers do it with bonus points. Same mechanics...
 
Since you said that this kind of conversation makes G+ really nice – why exactly doesn’t this kind of thing work on Facebook, do you think?
 
+Carl Zimmer +Onuralp Söylemez So the candidate loops put forward in the paper-- the three-node feed-forward loop and, to a lesser extent, the four-node bi-fan loop-- seem to represent, at least, the most efficient generally-useful looping mechanism that systems have evolved thus far?
 
Ecologists like the Odum brothers picked up on cybernetics and mathematical modeling very early on. (Look up "systems ecology".) My Ph.D. advisor, who's currently 78 or so, co-authored a paper with Eugene Odum in the 1980s called "The Cybernetic Nature of Ecosystems" and has been studying networks for most of his career. For a good history of this stuff, try the book An Entangled Bank by Joel Hagen.

Limited computational power was a huge obstacle. My advisor has stories about submitting models on punch cards to be run, only to find out the next day that you had left out a comma.
 
Two comments:
1. Crick left molecular biology for Neuroscience and spent much of his later life thinking and writing about consciousness. Feedback was a big player. But I don't know how he related this to molecular biology.
2. The Macy's symposia, which ran annually from the late 1940s for about a decade, brought a broad array of scientists together, and were largely about the relation of cybernetics to science. A remarkable array of minds, including Von Neumann and Weiner. But I don't know where to find records of what was said and presented.
 
Fairly early yet still accessible was Minsky & Papert's book on Perceptrons. I read the 1987 version but I believe it originally published in the 60's. It was certainly one of the early attempts to posit higher level brain activity as arising from a neural network using mechanistic feedforward circuits. Some of the "strong AI" proponents (a la Simon & Newell) jumped on this concept to extend to consciousness as a property that can arise out of these networks. I don't happen to agree with that (student of John Searle myself) but it's led to interesting work that's advanced both biology and engineering.
 
Lisa, the chronology is a bit different. Newell, Shaw, and Simon published their General Problem Solver in 1957, before Minsky and Papert's work, which was intended to show the limitations of learners based on linear threshold circuits.
 
That makes sense Fernando, since the GPS had such strict limitations on the types of problems it could handle, making it not at all like real humans solving problems. But it was an audacious and ultimately useful leap from industrial revolution constructs to neuroscience and cognition, wasn't it? And I agree that there are better analogies to use now that more of the intricacies (and delicacies) of gene regulation are becoming known.
Add a comment...