Biology is often seen as an experimental science, with theory serving to generalize patterns from empirical data. But history reveals that theory plays a more generative role in discovery itself.
Fascinating, wonderful writing! I listened to it and then came back to read it again. Thank you Ulkar.
I am still left with the question: What counts as a theoretical understanding? Suppose a machine is able to predict with great accuracy across varying contexts because it has represented the mechanism in detail, but the human using the machine cannot see how this mechanistic "understanding" is represented. Should we call this scientific advancement, or does science require a human to be the one understanding it?
Thanks so much for the kind feedback! Glad to hear you enjoyed the piece.
Yes, I'm still grappling with that question myself. I think if a machine can build a model of and predict a mechanism of a real-life phenomenon, that would count as scientific advancement, and I am open to granting that that may constitute an understanding for the machine itself. But I also think we cannot afford to do away with human understanding, especially in biology.
From an experimentalist's perspective, I can't help but think of the approach of "rank-ordering" parameters in a sloppy model. Molecular biologists are often performing titrations (log-scale searches on parameter sensitivity) and having limited resources (needing to prioritize which titrations are most important or sensitive).
I very often encounter friction between theorists and experimentalists. Theorists are wanting to fit the parameters to the data, where as experimentalists are typically seeking an accurate rank ordering before trying to fit any parameters, or even just fully satisfied with a rank ordering.
Intuition about the rank-ordering of sloppy parameters can come from many places, for example, when shopping for a reagent, you might find that many are advertised to be tolerant to temperature. An experimentalist would use this to infer that for some products, the temperature parameter may be very stiff (in some regime).
As mentioned, Poisson processes (diffusion, decay) are one set of phenomena that lead to these kinds of sloppy systems. Could there be non-Poisson biology, and could it then have compressible laws?
The notion that biology is a purely inductive science is a historiographical illusion fostered by an excess of phenomenology. Breakthrough moments in the history of the life sciences have almost never emerged from the incremental accumulation of data, but rather from theoretical frameworks that postulated the existence of structures long before experimental tools could detect them.
A few paradigmatic examples:
Erwin Schrödinger (1944): His insight into the "aperiodic crystal" in What is Life? defined the physical necessity of a stable, high-density informational code a decade prior to the structural resolution of DNA.
Alan Turing (1952): Reaction-diffusion morphogenesis predicted the dynamics of chemical gradients in embryonic development long before molecular biologists isolated and characterized individual morphogens.
François Jacob and Jacques Monod (1961): The operon model was not conceived as a mere curve-fit of enzymatic data, but as a logical and cybernetic construct of genetic feedback postulated out of pure systemic necessity.
In complex biological systems, characterized by high-dimensional, intrinsically sloppy parameter spaces—pure empiricism drowns in combinatorial titration and nonlinear variability.
Here, theory plays its most authentically generative role: rather than modeling reality by retrofitting parameters to pre-existing curves, it identifies a priori the physical, thermodynamic, and geometric constraints that prune the space of possibilities.
It is theory that dictates which parameters are stiff and which are negligible.
This is precisely what I verified experimentally between 2015 and 2020 at Sarapee Khuamung, Chiang Mai, investigating the biophysical interaction between coherent light frequencies and mitochondrial bioenergetic response.
Approached strictly through empirical screening, the project would have collapsed into an intractable combinatorial space, navigating endless titrations across spectral bands, irradiance levels, pulse frequencies, and exposure windows.
Instead, an upstream theoretical framework grounded in biophysical field dynamics isolated a priori the governing stiff parameters required to modulate mitochondrial chromophores and systemic energy states.
Experimentation was never a blind trial-and-error search for patterns in biological noise; it was the deductive, targeted validation of strictly constrained physical boundaries.
Arguably, information theory is one of the most successful approaches to understanding biology. Yet, not a word about information theory in this article.
Fascinating, wonderful writing! I listened to it and then came back to read it again. Thank you Ulkar.
I am still left with the question: What counts as a theoretical understanding? Suppose a machine is able to predict with great accuracy across varying contexts because it has represented the mechanism in detail, but the human using the machine cannot see how this mechanistic "understanding" is represented. Should we call this scientific advancement, or does science require a human to be the one understanding it?
Thanks so much for the kind feedback! Glad to hear you enjoyed the piece.
Yes, I'm still grappling with that question myself. I think if a machine can build a model of and predict a mechanism of a real-life phenomenon, that would count as scientific advancement, and I am open to granting that that may constitute an understanding for the machine itself. But I also think we cannot afford to do away with human understanding, especially in biology.
From an experimentalist's perspective, I can't help but think of the approach of "rank-ordering" parameters in a sloppy model. Molecular biologists are often performing titrations (log-scale searches on parameter sensitivity) and having limited resources (needing to prioritize which titrations are most important or sensitive).
I very often encounter friction between theorists and experimentalists. Theorists are wanting to fit the parameters to the data, where as experimentalists are typically seeking an accurate rank ordering before trying to fit any parameters, or even just fully satisfied with a rank ordering.
Intuition about the rank-ordering of sloppy parameters can come from many places, for example, when shopping for a reagent, you might find that many are advertised to be tolerant to temperature. An experimentalist would use this to infer that for some products, the temperature parameter may be very stiff (in some regime).
As mentioned, Poisson processes (diffusion, decay) are one set of phenomena that lead to these kinds of sloppy systems. Could there be non-Poisson biology, and could it then have compressible laws?
The notion that biology is a purely inductive science is a historiographical illusion fostered by an excess of phenomenology. Breakthrough moments in the history of the life sciences have almost never emerged from the incremental accumulation of data, but rather from theoretical frameworks that postulated the existence of structures long before experimental tools could detect them.
A few paradigmatic examples:
Erwin Schrödinger (1944): His insight into the "aperiodic crystal" in What is Life? defined the physical necessity of a stable, high-density informational code a decade prior to the structural resolution of DNA.
Alan Turing (1952): Reaction-diffusion morphogenesis predicted the dynamics of chemical gradients in embryonic development long before molecular biologists isolated and characterized individual morphogens.
François Jacob and Jacques Monod (1961): The operon model was not conceived as a mere curve-fit of enzymatic data, but as a logical and cybernetic construct of genetic feedback postulated out of pure systemic necessity.
In complex biological systems, characterized by high-dimensional, intrinsically sloppy parameter spaces—pure empiricism drowns in combinatorial titration and nonlinear variability.
Here, theory plays its most authentically generative role: rather than modeling reality by retrofitting parameters to pre-existing curves, it identifies a priori the physical, thermodynamic, and geometric constraints that prune the space of possibilities.
It is theory that dictates which parameters are stiff and which are negligible.
This is precisely what I verified experimentally between 2015 and 2020 at Sarapee Khuamung, Chiang Mai, investigating the biophysical interaction between coherent light frequencies and mitochondrial bioenergetic response.
Approached strictly through empirical screening, the project would have collapsed into an intractable combinatorial space, navigating endless titrations across spectral bands, irradiance levels, pulse frequencies, and exposure windows.
Instead, an upstream theoretical framework grounded in biophysical field dynamics isolated a priori the governing stiff parameters required to modulate mitochondrial chromophores and systemic energy states.
Experimentation was never a blind trial-and-error search for patterns in biological noise; it was the deductive, targeted validation of strictly constrained physical boundaries.
Great essay
Arguably, information theory is one of the most successful approaches to understanding biology. Yet, not a word about information theory in this article.