Lovely piece Phil. Personally I have a big problem with their notion of a 2D "concept space" for life. It reminds me of a diagram in Blaise's recent book about human sexuality where he also has a 2D concept space for homosexuality Vs hetersexuality - its one of the silliest uses of math I've ever seen! I'll use both examples to illustrate something i want to say about the abuses of "dimensional" thinking in the book I'm writing about the cultural/scientific/mathematical history of the concept of dimension. fyi: I like Blaise a lot personally! But one can take "math" too far.
Thank you Margaret. The 2D plot is really just a way of representing the data in a visual way that can be eyeballed - the axes don’t really have much physical meaning. I’m not even sure if proximity in this map means so much, as it’s a projection of a higher-dimensional space. Whether dimensionality is a useful way at all of conceptualizing the issue is another matter. I quite like it, which might reflect my condensed-matter physicist’s training in phase diagrams! I advocate something similar for “mindspace” in The Book of Minds, and I think we can usefully think of consciousness as a multi-dimensional attribute too. It’s not that such plots necessarily have a strong quantitative meaning, but they help us to recognize that words we sometimes use as if they mean a single thing - life, mind, consciousness, intelligence - are more complex, and not just things that entities have more or less of. It might also help us to think of development and change in terms of dynamical trajectories.
Phil - the concept of multidimensional representation is the whole theme of my new book. Like you, I Love the idea of phase spaces in physics! But i'm super skeptical of the use of these methods in relation to human-behavior/human-qualities. I'm doing several chapters devoted to explaining why. I agree such plots don't "necessarily have a strong quantitative meaning" but I'm arguing that they have strong propaganda power, which I believe needs to be countered. I understand what Blaise and Levin are trying to do here and I reject it on both epistemic and moral grounds. (I cite Hannah Arendt on this.) You as a physicist have a deep understanding of graphical plotting, but most people don't. I think as a society we're being fed a lot of hooey about the power of dimensional thinking in relation to human behavior. End of Rave.... It'd be much more fun to discuss this over a beer!
I'd love to hear about this Margaret - over a beer would be the ideal way, but I'm very keen of course to see your new book too. Please keep me in the loop!
Really interesting paper – and a great piece. I had two reactions to the paper: some amateur ones on the content; and some “professional” ones on method.
On content, and briefly – I was struck that the history section misses a popular piece of wisdom: that everything that lives must die. By no means sufficient, but perhaps necessary that all living things i) can die and ii) try not to (but must all the same). Admittedly, that just shifts the question to what dying means (and back to ideas in the article, e.g. thermodynamics, agency, etc). But I wonder if at least some of the definitional challenges arise from asking what distinguishes the living from the non-living when we haven’t first agreed what distinguishes the living from the dead. If nothing else, the second question might rule out the option of saying “there’s really no difference in reality”: try believing that on your deathbed.
On methods – I don’t know how interested you’ll be in this, but there may be some contact I think with some of your thinking about metaphor? – some of the rhetoric of this paper makes me nervous. The authors say: “Our application of LLM-driven semantic analysis to the definitional landscape demonstrates a novel approach to consensus formation across disciplinary boundaries. Rather than seeking to identify a single ‘correct’ definition, this approach maps the conceptual territory within which various definitions operate, revealing both their relationships and distinctive contributions.” There are two distinct things here – using LLMs, and mapping a definitional landscape. But lines liket he one above seem to me to imply that they somehow are, or should be, linked.
Mapping definitional landscapes (and versions of that) is not a novel approach at all. It’s my day job, as it is for many other serious qual researchers working in a broadly phenomenological way. No one’s ever asked me to do it for the concepts of scientists, but I have certainly been asked to do it for public understandings of scientific concepts (e.g. biodiversity). And as the authors rightly say, it is all about “preserv[ing] competing perspectives while revealing […] relationships between them, bridging reductionist and pluralist paradigms”.
I am really interested to see an example of how LLMs might be used as a tool to support human researchers in this task. What they’ve done is very interesting. My niggles are about how they have then reported it – and wider norms of science I suspect lie behind those reporting choices. Because I missed out a key word in the above quote: they in fact say “revealing quantitative relationships”. Why? That word has lots of meanings, but as a qual researcher I’m used to it being code for “mysteriously objective and beyond reproach, unlike that dodgy interpretative stuff some people do”! And that’s what I also see in this paper: an appeal to LLMs as somehow “objective” in a way a human interpreter could not be.
One key advantage that dodgy interpreters have is that they can engage in reflexive analysis. I’m not convinced LLMs are very good at that yet. But they clearly need to be: as the authors note in the limitations section: “the pairwise correlation methodology, while rigorous, necessarily relies on the semantic capabilities of the LLMs employed, which may introduce subtle biases in how conceptual relationships are evaluated”. Yep. Like a human being doing the same task, in fact. So where is the LLM’s reflexive analysis of these biases? Where indeed is the researchers’ reflexive analysis of their own assumptions and language around the use of LLMS? It's… oh, it’s quantitative research, so reflexive analysis isn’t needed, of course.
This all sounds very critical of the paper, which is unfair of me. I enjoyed it, and as I say found the example of an LLM being used as a tool really interesting. I think the issue here is a cultural one. If scientists really do want to “bridge reductionist and pluralist paradigms”, they really do have to get over their (cultural, if not individual) fear of human interpretation as a rigorous – if never “objective” – method.
Apologies for using your Substack comments to sound off something entirely different. Except... I guess I don't think it is entirely different.
But I’m not convinced that death need by a sine qua non of being alive. Of course, empirically it does tend to be. But the case of the deep-earth “Methuselah microbes” (see work by Karen Lloyd) forces me to wonder if that has to be so - why couldn’t a living being just keep repairing itself more or less indefinitely? One can argue that evolution needs death so that reproduction and natural selection don’t run out of space/resources, but that’s a different question.
Another really interesting article. Wish I had time (or is that inclination? ability?) to read/understand more.
Methuselah microbes - yes, but if the conditions are "can die" and "tries not to", that doesn't preclude an organism that succeeds in the trying part. (I would have to drop the "must all the same" in brackets, though; and engage with the semantics of possibility... fun!)
Thinking about it, the background question here is: what are we trying to do when we're defining a term? Easier to answer that question *within* a discipline, I think. But across disciplines... (the interesting challenge raised by the paper). What could make a definition right(er) or wrong(er)?
To me that has to go back to something that's extra- or pre-disciplinary - call it 'naive', 'lived experience', 'folk', whatever. That doesn't mean a good definition can't change, challenge, or enhance that pre-disciplinary experience - we don't experience the world 'as it really is', andwe should expect challenging boundary cases that lie outside that experience (microscopic viruses, 'life' on other planets, new technologies). But if we're not comparing our definitions at some point to that shared experience of a real distinction that really matters to us, on what basis are we saying "Does that really work as a definition OF life?" (as opposed to, say, is this well specified enough and accepted enough to work as the basis for intra-disciplinary discussion).
Which is what took me to death as the state that maybe defines life (even for immortals). Those pre-disciplinary experiences are of a world where individuals are far more important than types - in stark contrast to a scientific perspective. From this perspective, the term 'living' doesn't primarily distinguish humans from rocks: it distinguishes my husband as he is now from my husband as I know and fear he must one day be. The experienced distinction that grounds our intuitions about boundary cases and definitions is, I think, not living/non-living but alive/dead.
Not claiming this is an original point, btw. And I think similar arguments can be and are being made in relation to terms like 'consciousness' or 'free will'. (My own hazy thinking at the mo shaped by just having read Kevin Mitchell's incredibly good and unhazy book).
Anyway, in light of all of that, I loved some of the points in the article you just shared. Replication for example - it seems to me kind of *obvious* that, even if replication is a feature of all life, that's an empirical discovery, not part of what it means to be alive. (Although it strikes me I'm getting a bit Kantian analytic/synthetic now, which is probably not a good thing, as I vaguely remember from years ago that that doesn't work either...)
Lovely piece Phil. Personally I have a big problem with their notion of a 2D "concept space" for life. It reminds me of a diagram in Blaise's recent book about human sexuality where he also has a 2D concept space for homosexuality Vs hetersexuality - its one of the silliest uses of math I've ever seen! I'll use both examples to illustrate something i want to say about the abuses of "dimensional" thinking in the book I'm writing about the cultural/scientific/mathematical history of the concept of dimension. fyi: I like Blaise a lot personally! But one can take "math" too far.
Thank you Margaret. The 2D plot is really just a way of representing the data in a visual way that can be eyeballed - the axes don’t really have much physical meaning. I’m not even sure if proximity in this map means so much, as it’s a projection of a higher-dimensional space. Whether dimensionality is a useful way at all of conceptualizing the issue is another matter. I quite like it, which might reflect my condensed-matter physicist’s training in phase diagrams! I advocate something similar for “mindspace” in The Book of Minds, and I think we can usefully think of consciousness as a multi-dimensional attribute too. It’s not that such plots necessarily have a strong quantitative meaning, but they help us to recognize that words we sometimes use as if they mean a single thing - life, mind, consciousness, intelligence - are more complex, and not just things that entities have more or less of. It might also help us to think of development and change in terms of dynamical trajectories.
I was once asked why so many problems get analysed using 2x2 matrices. The only reason I could think of is that a piece of paper has two dimensions.
I think thats essentially right. We all learn about x/y coordinates in school and thats exactly what we can do on a sheet of paper.
Phil - the concept of multidimensional representation is the whole theme of my new book. Like you, I Love the idea of phase spaces in physics! But i'm super skeptical of the use of these methods in relation to human-behavior/human-qualities. I'm doing several chapters devoted to explaining why. I agree such plots don't "necessarily have a strong quantitative meaning" but I'm arguing that they have strong propaganda power, which I believe needs to be countered. I understand what Blaise and Levin are trying to do here and I reject it on both epistemic and moral grounds. (I cite Hannah Arendt on this.) You as a physicist have a deep understanding of graphical plotting, but most people don't. I think as a society we're being fed a lot of hooey about the power of dimensional thinking in relation to human behavior. End of Rave.... It'd be much more fun to discuss this over a beer!
I'd love to hear about this Margaret - over a beer would be the ideal way, but I'm very keen of course to see your new book too. Please keep me in the loop!
Really interesting paper – and a great piece. I had two reactions to the paper: some amateur ones on the content; and some “professional” ones on method.
On content, and briefly – I was struck that the history section misses a popular piece of wisdom: that everything that lives must die. By no means sufficient, but perhaps necessary that all living things i) can die and ii) try not to (but must all the same). Admittedly, that just shifts the question to what dying means (and back to ideas in the article, e.g. thermodynamics, agency, etc). But I wonder if at least some of the definitional challenges arise from asking what distinguishes the living from the non-living when we haven’t first agreed what distinguishes the living from the dead. If nothing else, the second question might rule out the option of saying “there’s really no difference in reality”: try believing that on your deathbed.
On methods – I don’t know how interested you’ll be in this, but there may be some contact I think with some of your thinking about metaphor? – some of the rhetoric of this paper makes me nervous. The authors say: “Our application of LLM-driven semantic analysis to the definitional landscape demonstrates a novel approach to consensus formation across disciplinary boundaries. Rather than seeking to identify a single ‘correct’ definition, this approach maps the conceptual territory within which various definitions operate, revealing both their relationships and distinctive contributions.” There are two distinct things here – using LLMs, and mapping a definitional landscape. But lines liket he one above seem to me to imply that they somehow are, or should be, linked.
Mapping definitional landscapes (and versions of that) is not a novel approach at all. It’s my day job, as it is for many other serious qual researchers working in a broadly phenomenological way. No one’s ever asked me to do it for the concepts of scientists, but I have certainly been asked to do it for public understandings of scientific concepts (e.g. biodiversity). And as the authors rightly say, it is all about “preserv[ing] competing perspectives while revealing […] relationships between them, bridging reductionist and pluralist paradigms”.
I am really interested to see an example of how LLMs might be used as a tool to support human researchers in this task. What they’ve done is very interesting. My niggles are about how they have then reported it – and wider norms of science I suspect lie behind those reporting choices. Because I missed out a key word in the above quote: they in fact say “revealing quantitative relationships”. Why? That word has lots of meanings, but as a qual researcher I’m used to it being code for “mysteriously objective and beyond reproach, unlike that dodgy interpretative stuff some people do”! And that’s what I also see in this paper: an appeal to LLMs as somehow “objective” in a way a human interpreter could not be.
One key advantage that dodgy interpreters have is that they can engage in reflexive analysis. I’m not convinced LLMs are very good at that yet. But they clearly need to be: as the authors note in the limitations section: “the pairwise correlation methodology, while rigorous, necessarily relies on the semantic capabilities of the LLMs employed, which may introduce subtle biases in how conceptual relationships are evaluated”. Yep. Like a human being doing the same task, in fact. So where is the LLM’s reflexive analysis of these biases? Where indeed is the researchers’ reflexive analysis of their own assumptions and language around the use of LLMS? It's… oh, it’s quantitative research, so reflexive analysis isn’t needed, of course.
This all sounds very critical of the paper, which is unfair of me. I enjoyed it, and as I say found the example of an LLM being used as a tool really interesting. I think the issue here is a cultural one. If scientists really do want to “bridge reductionist and pluralist paradigms”, they really do have to get over their (cultural, if not individual) fear of human interpretation as a rigorous – if never “objective” – method.
Apologies for using your Substack comments to sound off something entirely different. Except... I guess I don't think it is entirely different.
On death: there were a few comments here, though not so much of that survived the edit.
https://www.quantamagazine.org/how-life-and-death-spring-from-disorder-20170126/
But I’m not convinced that death need by a sine qua non of being alive. Of course, empirically it does tend to be. But the case of the deep-earth “Methuselah microbes” (see work by Karen Lloyd) forces me to wonder if that has to be so - why couldn’t a living being just keep repairing itself more or less indefinitely? One can argue that evolution needs death so that reproduction and natural selection don’t run out of space/resources, but that’s a different question.
Another really interesting article. Wish I had time (or is that inclination? ability?) to read/understand more.
Methuselah microbes - yes, but if the conditions are "can die" and "tries not to", that doesn't preclude an organism that succeeds in the trying part. (I would have to drop the "must all the same" in brackets, though; and engage with the semantics of possibility... fun!)
Thinking about it, the background question here is: what are we trying to do when we're defining a term? Easier to answer that question *within* a discipline, I think. But across disciplines... (the interesting challenge raised by the paper). What could make a definition right(er) or wrong(er)?
To me that has to go back to something that's extra- or pre-disciplinary - call it 'naive', 'lived experience', 'folk', whatever. That doesn't mean a good definition can't change, challenge, or enhance that pre-disciplinary experience - we don't experience the world 'as it really is', andwe should expect challenging boundary cases that lie outside that experience (microscopic viruses, 'life' on other planets, new technologies). But if we're not comparing our definitions at some point to that shared experience of a real distinction that really matters to us, on what basis are we saying "Does that really work as a definition OF life?" (as opposed to, say, is this well specified enough and accepted enough to work as the basis for intra-disciplinary discussion).
Which is what took me to death as the state that maybe defines life (even for immortals). Those pre-disciplinary experiences are of a world where individuals are far more important than types - in stark contrast to a scientific perspective. From this perspective, the term 'living' doesn't primarily distinguish humans from rocks: it distinguishes my husband as he is now from my husband as I know and fear he must one day be. The experienced distinction that grounds our intuitions about boundary cases and definitions is, I think, not living/non-living but alive/dead.
Not claiming this is an original point, btw. And I think similar arguments can be and are being made in relation to terms like 'consciousness' or 'free will'. (My own hazy thinking at the mo shaped by just having read Kevin Mitchell's incredibly good and unhazy book).
Anyway, in light of all of that, I loved some of the points in the article you just shared. Replication for example - it seems to me kind of *obvious* that, even if replication is a feature of all life, that's an empirical discovery, not part of what it means to be alive. (Although it strikes me I'm getting a bit Kantian analytic/synthetic now, which is probably not a good thing, as I vaguely remember from years ago that that doesn't work either...)