August 3, 2017

War of the Inputs and Outputs


Earlier this Summer, I presented a talk on sub-optimal cybernetic systems at NetSci 2017. While the talk was a high-level mix of representational modeling and computational biology, there were a few loose ends for further discussion.


One of these loose ends involves how to model a biological system with boxes and arrows when biology is a multiscale, continuous process in both space and time [1]. While one solution is to add as much detail as possible, and perhaps even move to hybrid multiscale models, another solution involves the application of philosophy.

In the NetSci talk, I mentioned in passing a representational technique called metabiology. Our group has recently put out a preprint on the cybernetic embryo in which the level of analysis is termed metabiological. In a metabiological representation, the system components do not need to map isomorphically to the biological substrate [2]. Rather, the metabiological representation is a model of higher-order processes that result from the underlying biology.

From a predictive standpoint, this method is imprecise. It does, however, get around a major limitation of black box models -- namely what a specific black box is representative of. It makes more sense to black box an overarching feature or measurement construct than to constrain biological details to artificial boundaries.

A traditional cybernetic representation of a nonlinear loop. Notice that the boxes represent both specific (sensor) and general (state of affairs) phenomena.

The black box also changes through the history of a given scientific field or concept. In biology, for example, the black box is usually thought of as something to ultimately break apart and understand. This is opposed to understanding how the black box serves as a variable that interacts with a larger system. So it might seem odd to readers who assume a sort of conceptual impermanence by the term "black box".


A somewhat presumptuous biological example: in the time of Darwin, heredity was considered to be a black box. In the time of Hunt Morgan, a formal mechanism of heredity was beginning to be understood (chromosomes), but the structure was a black box. By the 1960s, we were beginning to understand the basic function and structure of genetic transmission (DNA and gene expression). Yet at each stage in history, the "black box" contained something entirely different. In a fast moving field like cell biology, this becomes a bit more of an issue.

A realted cultural problem in biology has involves coming to terms with generic categories. This goes back to Linnean classification, but more generally this applies to theoretical constructs. For example, Alan Turing's morphogen concept does not represent any one biological agent, but a host of candidate molecules that fit the functional description. Modern empirical biology involves specification rather than generalization, or precisely the opposite goal of theoretical abstraction [3].

The relationship between collective morphogen action and a spatial distribution of cells. COURTESY: Figure 4 in [4]. 

A related part of the black box conundrum is what the arcs and arrows (inputs and outputs) represent. Both inputs and outputs can be quite diverse. Inputs bring things like raw materials, reactants, free energy, sources of variation, components, while outputs include things like products, transformations, statistical effects, biological diversity, waste products, bond energy. While inputs and outputs can be broadly considered, the former (input signals) provide information to the black box, while the latter (output signals) provide samples of the processes unfolding within the black box. Inputs also constrain the state space representing the black box.

Within the black box itself, processes can create, destroy, or transform. They can synthesize new things out of component parts, which hints towards black box processes as sources of emergent properties [5]. Black boxes can also serve to destroy old relationships, particularly when a given black box has multiple inputs. Putting a little more detail to the notion of emergent black boxes involves watching how the black box transforms one thing into another [6]. This leads us to ask the question: do these generic transformational processes contribute to increases in global complexity?

Perhaps it does. The last insight about inputs and outputs comes from John Searle and his Chinese Room problem [7]. In his model of a simple message-passing AI, an input (in this case, a phrase in Chinese) is passed into a black box. The black box processes the input either by mere recognition or more in-depth understanding. These are referred to as weak and strong artificial intelligence, respectively [7]. And so it is with a cybernetic black box -- the process within the unit can be qualitatively variable, leading to greater complexity and potentially a richer representation of biological and social processes.

NOTES:
[1] certainly, systems involving social phenomena operate in a similar manner. We did not discuss social systems in the NetSci talk, but things discussed in this post apply to those systems as well.

[2] for those whom are familiar, this is quite similar to the mind-brain problem from the philosophy of mind literature, in which the mind is a model of thought and the brain is the mechanism for executing thought.

[3] this might be why robust biological "rules" are hard to come by.

[4] Lander, A. (2011). Pattern, Growth, and Control. Cell, 144(6), 955-969.

[5] while it is not what I intend with this statement, a good coda is the Sidney Harris classic "then a miracle happens..."


[6] 

[7] Searle, J. (1980). Minds, Brains, and Programs. Cambridge University Press, Cambridge, UK.

July 26, 2017

Battle of the (Worm) Bots


There is an fledgling initiative at the OpenWorm Foundation to build a worm robot. This post highlights some of the first steps towards this goal. OpenWorm Robots will be useful for educational purposes and movement experiments/validation. No soft robotics (or miniaturization) yet, but progress is also continually being made on the brains behind the bot. More videos are forthcoming, so stay tuned. Thanks go to Dr. Tom Portegys and Shane Gingell for their efforts.


This is the latest version of the OpenWorm bot, as shown on the OpenWorm YouTube channel.


This is Shane's first iteration, called RoboWorm. Note the sinusoidal movement, and then compare to a biological C. elegans.

July 16, 2017

Wandering Towards an Essay of Laws


The winners of the FQXi "Wandering Towards a Goal" essay contest have been announced. I made an entry into the contest (my first FQXi contest entry) and did not win, but had a good time creating a number of interesting threads for future exploration. 

The essay itself, "Inverting the Phenomenology of Mathematical Lawfulness to Establish the Conditions of Intention-like Goal-oriented Behavior" [1], is the product of my work in an area I call Physical Intelligence in addition to intellectual discussions with colleagues (acknowledged in the essay). 

I did not have much time to polish and reflect upon the essay at the time it was submitted, but since then I have come up with a few additional points. So here are a few more focused observations extracted from the more exploratory essay form:

1) there is an underappreciated connection between biological physics, evolution, and psychophysics. There is an subtle but important research question here: why did some biological systems evolve in accordance with "law-like" behavior, while many others did not? 

2) the question of whether mathematical laws are discovered or invented (Mathematical Platonism) may be highly relevant to the application of mathematical models in the biological and social sciences [2]. While mathematicians have a commonly encountered answer (laws are discovered, notation was invented), an answer based on discovering laws from empirically-driven observations will likely provide a different answer.

3) how exactly do we define laws in the context of empirical science? While laws can be demonstrated in the biological sciences [3], biology itself is not thought of as particularly lawful. According to [4], "laws" fall somewhere in-between hypotheses and theories. In this sense, laws are both exercises in prediction and part of theory-building. Historically, biologists have tended to employ statistical models without reference to theory, while physicists and chemists often use statistical models to demonstrate theoretical principles [5]. In fields such as biology or the social sciences, the use of different or novel analytical or symbolic paradigms might facilitate the discovery of lawlike invariants.

4) the inclusion of cybernetic principles (Ashby's Law of Requisite Variety) may also bring together new insights on how laws occur in biological and social systems, and whether such laws are based on deep structural regularities in nature (as argued in the FQXi essay) or the mode of representating empirical observations (an idea to be explored in another post).

5) Aneural cognition is something that might guide information processing in a number of contexts. This has been explored further in another paper from the DevoWorm group [6] on the potential role of aneural cognition in embryos. It has also been explored in the form of the free-energy principle leading to information processing in plants [7]. Is cognition a unified theory of adaptive information processing? Now that's something to explore.


NOTES:
[1] A printable version can be downloaded from Figshare (doi:10.6084/m9.figshare.4725235).

[2] I experienced a nice discussion of this issue during an recent NSF-sponsored workshop. The bottom line is that while the variation typical of biology often makes the discovery of universal principles intractable, perhaps law discovery in biology simply requires a several hundred year investment in research (h/t Dr. Rob Phillips). For more, please see:

Phillips, R. (2015). Theory in Biology: Figure 1 or Figure 7? Trends in Cell Biology 25(12), 1-7.

[3] Trevors, J.T. and Saier, M.H. (2010). Three Laws of Biology. Water Air and Soil Pollution, 205(S1), S87-S89.

[4] el-Showk, S. (2014). Does Biology Have Laws? Accumulating Glitches blog, Nature Scitable. http://www.nature.com/scitable/blog/accumulating-glitches/does_biology_have-laws

[5] Ruse, M.E. (1970). Are there laws in biology? Australasian Journal of Philosophy, 48(2), 234-246. doi:10.1080/00048407012341201.

[6] Stone, R., Portegys, T.E., Mihkailovsky, G., and Alicea, B. (2017). Origins of the Embryo: self-organization through cybernetic regulation​. Figshare, doi:10.6084/m9.figshare.5089558.

[7] Calvo, P. and Friston, K. (2017). Predicting green: really radical (plant) predictive processing. Journal of the Royal Society Interface, 14, 20170096.

July 2, 2017

Excellence vs. Relevance

Impetus for this blog post. Twitter h/t to Alex Lancaster (@LancasterSci).

In academia, the term excellence is often used in the context of scarcity and competitive dynamics (e.g. publications, career promotion), and as a result can be used quite arbitrarily [1]. In [1], a distinction is made between excellence and soundness. Excellence is seen as a subjective concept, while soundness (enabled through completeness, thoroughness, and an emphasis on reproducibility) is the adherence to clearly defined and practiced research standards. While it may also be true that the concept of soundness can suffer from the same subjective limitations, it is probably an improvement over the current discussions surrounding excellence.

Another term we rarely refer to, but may be of even greater importance, is the relevance of scientific research. In a previous post, I brought relevance theory to bear on potential biases in scientific selectivity. One way to think of relevance is as the collective attentional focus of a given research group, community, or field. Collective attention (and thus relevance) can change with time: papers, methods, and influences rise and fall as research ideas are executed and new information is encountered [2]. As such, relevance defines the scope of scientific research that defines a particular field or community of researchers. Given a particular focus, what is relevant defines what is excellent. In this case, we return to the biases inherent in excellence, but this time with a framework for understanding what it means in a given context.

There is also an interesting relationship between soundness and relevance. For example, the stated goal of venues like PLoS One and Scientific Reports is to evaluate manuscripts based on methodological soundness rather than merely on field-specific relevance. To some extent this has eliminated issues of arbitrary selectivity, yet reviewers and editors from various fields may still surruptitiously impose their own field-specific conventions to the review process. Interestingly, soundness itself can be a matter of relevance, as the use of specific methodologies and modes of investigation can be highly field-specific.

Sometimes relevance is a matching problem between an individual researcher and the conventions of a specific field. Relevance can be represented as a formalized conceptual problem using skillset geometries [3]. In the example below, I have shown how the relevance of a specific individual overlaps with what is considered relevant in a specific field. In this case, the researcher has expertise in multiple areas of expertise, while the field is deeply rooted in a single domain of knowledge. The area of overlap, or Area of Mutual Relevance, describes the degree of shared relevance between individual and community (sometimes called "fit").


How relevant is a single person's skillset in the context of a research community, and how do we leverage this expertise in an inclusive manner? The mutual relevance criterion might provide opportunities in cases where there seems to be a "lack of fit". Understanding the role of collective attention within research communities might allow us to consider how this affects both the flow of new ideas between fields and the successful practice of interdisciplinarity.


NOTES:
[1] Moore, S., Neylon, C., Eve, M.P., O'Donnell, D.P., and Pattinson, D. (2016). Excellence R Us: university research and the fetishisation of excellence. Palgrave Communications, 3, 16105. doi:10.1057/palcomms.2016.105.

[2] Wu, F. and Huberman, B.A. (2007). Novelty and collective attention. PNAS, 104(45), 17599-17601. doi: 10.1073/pnas.0704916104.

[3] First introduced in: Alicea, B. (2017). A peripheral Darwin Day post, but centrality in his collaboration graph. Synthetic Daisies blog, February 16.

June 18, 2017

Loose Ends Tied, Interdisciplinarity, and Consilience

LEFT: A network of scientific disciplines and concepts built from clickstream data. RIGHT: Science mapping based on relationships among a large database of publications. COURTESY: Figure 5 in [1] (left) and SciTech Strategies (right).

Having a diverse background in a number of fields, I have been quite interested in how people from different disciplines converge (or do not converge) upon similar findings. Given that disciplines are often methodologically distinct communities [2], it is encouraging when multiple disciplines can exhibit consilience [3] in attacking the same problem. For me, it is encouraging because it supports the notion that the phenomena we study are derived from deep principles consistent with a grand theorizing [4]. And we can see this is areas of inquiry such as learning and memory, with potential relevance to a wide variety of disciplines (e.g. cognitive psychology, history, cell biology) and the emergence of common themes according to various definitions of the phenomenon.

Maximum spanning tree of disciplinary interactions based on the Physics and Astronomy Classification Scheme (PACS). COURTESY: Figure 5 in [5].

The ability to converge upon a common set of findings may be an important part of establishing and maintaining coherent multidisciplinary communities. Porter and Rafols [6] have examined the growth of interdisciplinary citations as a proxy for increasing interdisciplinarity. Interdisciplinary citations tend to be less common than within-discipline citations, while also favoring linkages between closely-aligned topical fields. Perhaps consilience also relies upon the completeness of literature inclusion for people from different disciplines in an interdisciplinary context. Another recent paper [7] suggests that more complete literature citation might lead to better interdisciplinary science and perhaps ultimately consilience. This of course depends on whether the set of evidence itself is actually convergent or divergent, and what it means for concepts to be coherent. In the interest of not getting any more abstract and esoteric, I will leave the notion of coherence for another post.


NOTES:
[1] Bollen, J., Van de Sompel, H., Hagberg, A., Bettencourt, L., Chute, R., Rodriguez, M.A., and Balakireva, L. (2009). Clickstream Data Yields High-Resolution Maps of Science. PLoS One, 4(3), e4803. doi:10.1371/journal.pone.0004803.

[2] Osborne, P.  (2015). Problematizing Disciplinarity, Transdisciplinary Problematics. Theory, Culture, and Society, 32(5-6), 3–35.

[3] Wilson, E.O. (1998). Consilience: the unity of knowledge. Random House, New York.

[4] Weinberg, S. (1993). Dreams of a Final Theory: the scientist's search for the ultimate laws of nature. Vintage Books, New York.

[5] Pan, R.J., Sinha, S., Kaski, K., and Saramaki, J. (2012). The evolution of interdisciplinarity in physics research. Scientific Reports, 2, 551. doi:10.1038/srep00551.

[6] Porter, A.L. and Rafols, I. (2009). Is science becoming more interdisciplinary? Measuring and mapping six research fields over time. Scientometrics, 81, 719.

[7] Estrada, E. (2017). The other fields also exist. Journal of Complex Networks, 5(3), 335-336.

Printfriendly