Why machine-information metaphors are bad for science education, part II: the search for new metaphors

metaphor vs simileWhile discussing some sections of a paper I wrote with Maarten Boudry, we have seen a number of reasons why using machine-information metaphors is bad for science education. As I pointed out before, the full paper also devotes quite a bit of space to arguing that those metaphors haven’t been particularly good in actual scientific research. One of the fascinating things to watch after I posted the first part of this commentary was the number of people who vehemently defended the “biological organisms are machines” take, both here on the blog and on my Twitter feed. It’s like here we are, in the second decade of the 21st century, and there are still a lot of Cartesians around, who have apparently never heard of David Hume. Oh well.

In the conclusion of this two-part series I am going to focus on the last section of my paper with Maarten, where we discuss the search for alternative metaphors, and in the end (spoiler alert!) suggest that the best thing to do at this point is just to describe things as they are, staying as clear as possible of metaphorical language. And when one really cannot avoid it, then use multiple metaphors and be very clear on the limits of their use. Let’s take a look.

In their classic work on metaphors, Lakoff and Johnson argue that the basic function of metaphorical concepts is to structure a new kind of experience in terms of a more familiar and delineated experience. In science as well as in everyday language, metaphors highlight particular aspects of whatever it is we are trying to grasp, but they will inevitably distort others. For example, the image of the “tree of life,” with new species branching off as budding twigs and extinct species as dead branches, is an instructive approximation of the relations of evolutionary descent. However, it can also foster misconceptions about “progress” in evolution, or lead to a simplistic conception of speciation events, or to a downplay of horizontal gene transfer and reticulate (i.e., by interspecies hybridization) speciation events. To give one more example, in physical chemistry the model of the atom as a miniature solar system, with electrons orbiting the nucleus as planets, though still having wide public appeal, is fundamentally inaccurate.

Of course, no metaphor will do its job perfectly, but it is crucial to realize, as Lakoff and Johnson have shown, that the widespread deployment of a particular metaphor can have a feedback effect on the way we perceive things, not just how we present them to others. In the examples discussed in my paper with Maarten, the lure of machine-information metaphors in the history of biology has invited scientists to think of genomes as “blueprints” for organisms, written in the four-letter alphabet of DNA and readable in a manner analogous to a computer code. But as we argue, the machine-information conception of living systems has led both the public and the scientific community astray.

In response to this problem, some scientists and science educators have proposed several alternative and improved metaphors to characterize the relationship between genotype and phenotype. Biologist Patrick Bateson, for instance, was probably the first to compare the DNA sequence of living organisms with a recipe for a cake. The idea of a genetic recipe has several advantages over the blueprint metaphor, the most important being that it takes into account pleiotropy (one gene affecting more than one trait) and epistasis (gene–gene interactions). As a consequence, the simple picture of a one-to-one (or close to) correspondence between particular genes and phenotypic traits is abandoned, which becomes clear when one considers that there is no way to locate particular ingredients in individual crumbs of a cake. Accordingly, there is no possibility of reverse-engineering the end product to the set of procedures (the “recipe”) that made the final product possible. This has important consequences not just for science education, but for research agendas, as the idea of ‘‘reverse engineering’’ is commonly invoked everywhere from genomic studies to the understanding of the brain.

Of course, if carried too far, the recipe metaphor can in turn be quite misleading. To get the desired result, a cook has to lump together different ingredients in the correct proportions and follow a set of instructions for handling the dough and preparing the oven. But actual developmental encoding in living organisms is an enormously more complex and very different sort of procedure, which is also highly dependent on epigenetic factors and unpredictable vagaries of the external environment. The expression of specific genes in the course of development resembles nothing like the way a cook handles the ingredients of a recipe. Living organisms are also highly differentiated in a number of functional parts or components (cell types, tissues, etc.), in contrast with the homogenous cake that comes out of the oven. Moreover, the genome is not written in anything like a ‘‘language,’’ as in the case of a recipe, and it certainly does not contain a description of the desired end product in any meaningful sense of the word ‘‘description.’’

Condit and colleagues have discussed the recipe metaphor as an alternative to talk of blueprints, pointing out that it was adopted ‘‘with surprising swiftness’’ by science popularizers and the media in the 1990s. However, they also remark that, as a new ‘‘master metaphor’’ to capture the relationship between genotype and phenotype, the image of a recipe for a cake has little to recommend either. For example, evoking recipes can invite people to think of the genome as a step-by-step manual that describes ‘‘how to make a human,’’ in that sense falling into the same trap as the idea of a blueprint.

That being said, if contrasted with the blueprint metaphor, the recipe metaphor conveys the point about lack of one-to-one correspondence between genes and phenotypes very well, and hence it highlights an important fact about development and what biologists call the Genotype => Phenotype map. If the recipe metaphor is used within this restricted context, for example in explicit contrast with the characteristics of a blueprint, it is immediately clear what are the salient points of connection with living systems, and people are less likely to be misled by stretching the metaphor beyond usefulness. If the recipe metaphor is presented as an alternative to the blueprint, however, it is bound to mislead people no less than its rival.

The same point applies to other interesting metaphors that have been proposed in this context, for example Lewis Wolpert’s comparison of early embryonic development with the Japanese art of origami. The analogy highlights the circuitous step-by-step development of the early embryo, but of course in a piece of origami art the structure is imposed top-down from an intelligent agent, whereas the functional differentiation in the embryo is regulated bottom-up by a complex interaction between genes and environment. Moreover, origami simply fold to yield the final product, which in a very real sense is already there from the beginning. This is definitely not the way embryos develop, with their ability to respond to local and external environmental fluctuations.

The general problem that we have been discussing seems to us to be not just that one kind of metaphor or another is woefully inadequate to conceptualize biological organisms and their evolution. It is that it simply does not seem to be possible to come up with a metaphor that is cogent and appropriate beyond a very limited conceptual space. Although some of the alternatives are more accurate than the blueprint metaphor (in some respects), Maarten and I certainly have not found one that we would recommend as a replacement. Should we therefore try to avoid the use of metaphors in biological teaching and research altogether? Or do we simply expect too much from metaphors in science and education?

Analogical and metaphorical thinking is widespread among human beings, although of course different cultures and historical moments inspire people to use different metaphors. After all, a metaphor is an attempt to make sense of novel concepts by pairing them with known ideas to increase our overall understanding. Metaphorical thinking is therefore part of our language, and language is inextricably connected to our thinking, but to put it as Wittgenstein did: ‘‘It is, in most cases, impossible to show an exact point where an analogy starts to mislead us.’’ Yet a great part of doing philosophy consists precisely in clarifying our language in an attempt to advance our thinking. To quote Wittgenstein again: ‘‘Philosophy is a battle against the bewitchment of our intelligence by means of our language.’’ To complicate matters further, there is emerging empirical evidence that the human brain processes metaphors in a specific fashion: research on Alzheimer’s patients, for instance (see ref. in the paper), found that impairment of the brain’s ‘‘executive’’ function, associated with the prefrontal cortex, leads to poor understanding of novel metaphors (while, interestingly, comprehension of familiar metaphors is unaffected). Metaphorical thinking seems to be a biologically entrenched functional mode of our brains, and may therefore be hard to avoid altogether.

Both science and philosophy have made ample use of metaphorical and analogical thinking, sometimes with spectacularly positive results, at other times more questionably so. Nonetheless, it seems that nowhere is metaphorical thinking so entrenched — and so potentially misleading — as in biology. Given the maturity of biology as a science, and considering that it deals with objects whose nature is not as alien to our daily experience as, say, those of quantum physics, Maarten and I do not actually see any good reason for clinging onto outdated metaphors in biological education and research for characterizing living organisms, their genomes and their means of development. Taking into account the fact that the machine information metaphors have been grist to the mill of ID creationism, fostering design intuitions and other misconceptions about living systems, we think it is time to dispense with them altogether. Still, we are also not as naive as to expect that this advice will be followed by scientists and science educators any time soon, precisely because the machine/information metaphor is so entrenched in biology education. What to do then? We propose two approaches, one for science educators, the other for practicing scientists.

In science education, talk of metaphorical thinking can be turned into a teaching moment. Students (and the public at large) would actually greatly benefit from explanations that contrast different metaphors with the express goal of highlighting the limitations intrinsic in metaphors and analogies. So, for instance, science educators and writers could talk about the human genome by introducing the blueprint metaphor, only to immediately point out why it does not capture much of what genomes and organisms are about; they could then proceed to familiarize their students and readers with alternative metaphors, say the recipe one, focusing on differences with the original metaphor while of course not neglecting to point out the (different) deficiencies of the new approach as well. The goal of this process would be to foster a cautious attitude about metaphorical thinking, as well as to develop a broader understanding of how unlike commonsense modern science really is. On the latter point, it is interesting to note, for instance, that a popular refrain among evolution or global warming deniers is that ‘‘simple commonsense’’ shows that the scientists are wrong, a position that ignores the proper weight of technical expertise in favor of a folk understanding of nature. It is therefore crucial that the public appreciates the limitations of common sense thinking about science.

There is an analogous teaching moment that can be brought to bear when research scientists engage in unbridled metaphorical thinking: we could refer to this as a philosophy appreciation moment. Scientists are notoriously insensitive to, or even downright dismissive of, considerations arising from the history and philosophy of their discipline, and often for good practical reasons: modern science is a highly specialized activity, where there is barely enough time to keep up with the overwhelming literature in one’s own narrow field of research, and certainly not enough incentive to indulge in historical readings or philosophical speculation. Nonetheless, historians and philosophers of science can easily show the pitfalls of metaphorical thinking (by using well-documented historical examples) and even get across to their colleagues some basic notions of philosophy (by analyzing the effects of particular metaphors on the development of specific lines of scientific inquiry). None of this will quickly amount to overcoming C.P. Snow’s infamous divide between ‘‘the two cultures,’’ but it may bring about better understanding and appreciation of philosophy by scientists, and perhaps even help science see new horizons that have been hitherto obscured by a superficially illuminating metaphor.

115 thoughts on “Why machine-information metaphors are bad for science education, part II: the search for new metaphors

  1. Daniel Kaufman

    Clearly there are no purposes or goals without desires. How would one characterize a purpose or goal, other than in terms of something one wants?

    If by ‘purpose’ one means something that is only teleonomic, then perhaps one can make sense of a desireless goal. But if we are speaking strictly and carefully, its not clear to me that the functions one identifies at the biological level of description are properly understood in terms of purposes or goals — again, speaking strictly and carefully.

    Liked by 2 people

  2. Massimo Post author

    brodix,

    “Would “impulse” be an acceptable term for the motivating element of non fauna biology? Is there some other acceptable term”

    Impulse is also usually reserved for organisms with nervous systems. Tendency is a perfectly fine word.

    Like

  3. Daniel Kaufman

    Synred: I was waiting for someone to mention computers, as it leads me to my second point.

    A machine is something that is entirely functionally characterizable, and a functional description, by definition, excludes any mention of substrate. But in the case of living organisms, substrate is crucial, which is why a purely functional characterization is not possible.

    Liked by 2 people

  4. brodix

    Robin,

    ““goal” and “desire” are not synonyms.”

    That has been my point. That we are driven by the fact we desire, rather than any object of our desire.

    “My desire was to understand the process of genetic algorithms better.”

    Your desire was to expand on what you knew, not any particular conclusion that is drawn. That impulse to explore, from the basis of prior results of previous exploration. Expand>consolidate.

    Like

  5. synred

    The substrate changes in organisms too all which is not typical of machines or at least much slower. My computer is gradually changing its substrate as things break and get replaced. (Somebodies’ famous ship too?).

    I like analogy with a soliton, but of course it not too exact.

    Like

  6. Alan White

    Instincts have goals in some sense, but many demonstrations of such in the animal kingdom certainly have no basis in desire. I see no sense in saying spiders “want” to weave webs in order to kill prey–which in terms of desire would be a complex meta-desire to weave webs in order to fulfill the primary desire for food. If one wishes to call this teleonomic (I certainly would), then it is a very complex heaped form of teleonomic behavior. It’s this kind of complexity that draws anthropomorphist explanation like a magnet. (I’m not contradicting Dan, just offering support to be clear.)

    Liked by 2 people

  7. davidlduffy

    Hi Dan. I don’t think that “functional description, by definition, excludes any mention of substrate” quite works. The functional description includes all those relevant properties of the substrate that allow it to implement the function in question. For instance, engineers are quite interested in properties of materials.

    Like

  8. synred

    I wrote a story with thinking plants called ‘Causality’. The plants our the programmers of a simulation that turns out to be us. It’s really about QM. With no readership, I haven’t bothered to proof it, but the premise is kind of cute.

    Like

  9. Coel

    Massimo,

    I’m curious. Can you explain to me why you are so invested in the machine metaphor? What’s the deal?

    That’s a peculiar person from someone who has written two blog posts and a formal paper on the topic, and so must think it is worth discussing.

    But, to answer, I argue for “machine” talk (and again, I don’t consider it a metaphor, I regard it as literal) because I see it as helpful rather than unhelpful.

    I also think (as I said in Part 1) that denying the similarities between human-designed functional machines and the functional “machines” that result from Darwinian evolution, would hand rhetorical advantage to the creationists.

    Obviously you think the opposite on these points.

    Liked by 3 people

  10. brodix

    Sorcatic,

    Thinking and feeling are not the same. I realize we have this top down view of our drives, but they evolve bottom up.

    Denying some biological urge seems as motivated as assuming it has some self reflective, platonically deified component.

    As I pointed out in the previous thread, we can assume platonic form, but step anywhere near life as anything other than emergent from atoms bumping into each other and scientismists run shrieking.

    Like

  11. Philip Thrift

    “But in the case of living organisms, substrate is crucial, which is why a purely functional characterization is not possible.”

    That is certainly the case in biocomputing. Substrate is crucial in the operational semantics of bioprogramming where (for example) “synthetic genes circuits could be programmed to multiplex molecular biomarkers in vivo and compute appropriate therapeutic outputs”, and “medical molecular payloads in live animals” can be delivered.
    https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5009805/

    Like

  12. brodix

    I don’t assume to know any more than is evident. There are sentient life forms. We don’t know what the source of that sentience is, but taking it as a bottom up axiom or process, than assuming it is illusion, or proposing an ideal source, would allow a logical understanding of biological interaction, where it is competing desires, rather than presuming ideals or extremes as inevitable conclusions and goals.

    It would also refute the logic of a top down ideal as source of both sentience and biological complexity.

    More the new born baby than the wise old man.

    Look at how the world is being run. Does ego or wisdom seem to be the primary motivation? Wouldn’t a philosophy taking the obvious into account seem reasonable?

    Like

  13. Massimo Post author

    david,

    “The functional description includes all those relevant properties of the substrate that allow it to implement the function in question. For instance, engineers are quite interested in properties of materials”

    That’s surely the case, but a functional description of a system concerns itself only with the role played by each part in that system, not with what the parts are made of. That’s a big reason I’m not a fan of functionalism in philosophy of mind, because I think — like the engineers — that the materials matter.

    Liked by 2 people

  14. Massimo Post author

    Coel,

    “I argue for machine talk (and again, I don’t consider it a metaphor, I regard it as literal) because I see it as helpful rather than unhelpful.

    I also think (as I said in Part 1) that denying the similarities between human-designed functional machines and the functional ‘machines’ that result from Darwinian evolution, would hand rhetorical advantage to the creationists.”

    Well, to regard machine talk in biology has literal is bizarre, and I doubt even the most entusiastic fans of the metaphor would regard it as such.

    Maarten and I have made a pretty good, documented argument for why the metaphor has been more damaging than not.

    And you are absolutely wrong about the alleged advantage the metaphor gives us against the creationists. They love it.

    Like

  15. Coel

    Massimo,

    Well, to regard machine talk in biology has literal is bizarre, and I doubt even the most entusiastic fans of the metaphor would regard it as such.

    I suspect you’d be surprised! 🙂 (And, as per comments in Part 1, it fits the straightforward OED definition.)

    Maarten and I have made a pretty good, documented argument for why the metaphor has been more damaging than not.

    Most of your argument and documentation is about the metaphor “blueprints”, not directly about the term “machine”.

    And you are absolutely wrong about the alleged advantage the metaphor gives us against the creationists. They love it.

    It’s not that the comparison to “machine” directly gives us an advantage — yes they do indeed love it — it is that they would love it even more if you tried to deny the comparison (such that they would accuse you of ideological blindness) rather than accepting it head on with “yes, functional machines are exactly what natural selection produces”.

    Liked by 3 people

  16. Philip Thrift

    “in philosophy of mind … the materials matter”

    A chemicobot is debating a silicobot about which is fully conscious: “You can’t be fully conscious because you don’t have the right chemicals for feeling like I do, like endorphins.”

    Like

  17. Massimo Post author

    Coel,

    “I suspect you’d be surprised!”

    I can’t wait to see the peer review paper showing that most biologists take machine metaphors non-metaphorically. What was the reference, again?

    Like

  18. Philip Thrift

    “yes, functional machines are exactly what natural selection produces”

    But it’s not quite like that, or that simple today.

    “As models of living bodies, synthetic biological machines are less misleading. They will undoubtedly get more impressive, maybe even, outwardly, more lifelike. But it’s a good bet that their success will depend on a lot of ingenious backstage engineering, devoted to the technofixes needed to stabilize the functioning of the DNA-work over ever larger ranges of environments, and to buffer the devices from outrageous fortune’s slings and arrows. Nevertheless, expect changes ahead, perhaps even—if the ambitions on display in that Leeds storefront are anything to go by—benign ones.”

    Click to access 701b20e8403ac5b88f8c597fd82e2e8f9cf4.pdf

    Like

  19. brodix

    Massimo,

    “Uhm, no.”

    Could you explain that to Socratic then.

    E.O. Wilson described the insect brain as a thermostat. That is sensory. They register temperature gradients, like our fingers do, when we touch a hot stove. Necessarily it takes quite a bit of feedback to get to the level of introspection we would have, but it is evolution and only evolution which leads from them to us.

    The real question is quite a few layers below them.

    Like

  20. synred

    If you have ‘bugs’ or damage to the hardware, machines will produce incorrect results which one might call hallucinations. Our machines being simpler produce simpler mistakes.

    Like

  21. Massimo Post author

    “Of course I’ve never suggested that they do.”

    You did, my friend. Same line of argument that brought you to that conclusion that Siri is (somewhat) conscious. Indeed, I’m pretty sure you use a thermostat as an example.

    Liked by 1 person

  22. synred

    Coel: I remember you saying writing something along the lines of thermostats having ‘rudimentary consciousness’ … I don’t remember when and that was not the exact phrase.

    At the time, it struck me as pretty extreme even for you…

    Liked by 2 people

  23. Coel

    Massimo,

    You did, my friend. Same line of argument that brought you to that conclusion that Siri is (somewhat) conscious. Indeed, I’m pretty sure you use a thermostat as an example.

    No, that’s roughly the opposite of what I was saying. I have never thought for a moment that Siri is at all conscious. I’d consider that idea ludicrous.

    What I was doing was considering whether Siri understands, and in doing that I was taking a stripped down and functional account of what is necessary for “understanding”. Thus, if Siri can correctly respond to the request “what is the time?” then Siri “understands” that sentence (but not necessarily anything more than that specific sentence), and that’s all there is to it.

    It seems to be philosophers (not me) who regard conscious self awareness as a necessary part of “understanding” and similar concepts. Thus you interpreted me as saying that Siri was conscious but I very explicitly stated that that was not what I was saying.

    Liked by 1 person

Comments are closed.