New Feelings

I have not explored AI text or image tools and I don’t have much interest in changing that. But, some of the writing and images that I have seen generated with AI tools are very impressive! The giant robot above is an example. I am not so ludditic as to see these tools as a net loss, as they obviously unlock new abilities as much as they threaten old ones. But personally, finished products have never done much for me. The instant some project or goal is finished, I cease to care about it very much. I didn’t choose this mindset and I don’t necessarily recommend it, but it has proven too stubborn to do much about. It is obviously nice to complete tasks, but I am more motivated by the understanding and insight which is generated as a result of doing and making things. And it seems to me that the AI text and video tools implicitly frame the struggle to gain actionable understanding as simply a means to an end of accomplishing tasks, and thus, a problem to be automated away. I could get behind this if we had an updated vision for “higher level” tasks which transcended things like writing and coding. But to me, writing and coding, as painful as they can be, are not means to an end, they are an external means of refining the process of thought itself. It’s not clear that there are worthy higher level tasks which do not come at the cost of honing basic thought processes in this way (our brains have limitations, at the end of the day). I worry that these tools cultivate power either without a concomitant advance in wisdom, or even at the expense of wisdom. I am curious to find out if we do discover these higher level tasks in the next few years.

This is a way of thinking about engineering ethics that Professor Paul Diduch at CU brought up to me a few months ago. I still haven’t explored my thoughts thoroughly enough to feel organized in my presentation of the idea. But the gist is this—when we talk about ethical dilemmas or social problems, such as climate change or digital technology negatively affecting mental health, it is typical to first explain the problem, then talk about possible solutions. But for the really nasty problems, proposed solutions often don’t seem very actionable or realistic. How many books on the perils of climate change conclude with a prescription to “talk about these issues more,” or “vote,” or “reduce personal emissions”? Do these solutions seem even remotely capable of addressing the enormity of the problem? No, I would argue. The ubiquity of unsatisfactory solutions of this sort demonstrates to me that no one really knows what to do. For many hard problems, it seems that the realistic solutions are ineffective, and the effective solutions are unrealistic. In response to this apparent bind, the logic of “knowable problems” says the following: let’s not pretend like we know what to do, but instead, let’s invest effort in building a strong understanding of why a problem is hard. The nature of a problem can be known without having a solution to it. In fact, characterizing why something is difficult is often quite easy! Engineers are already somewhat comfortable with this way of thinking. For example, we say that non-linear models are more complex and unpredictable than their linear counterparts. Non-linearity is property which is simple to investigate. Thus, we can determine why a problem is hard without having any solution. 

It may seem counterproductive to de-emphasize solutions, but I am not convinced that a hard problem can be solved without first being deeply understood. If we say that climate change is a hard problem because the world economy runs on profit, profit relies on growth, and growth increases emissions, then we have neatly characterized an important aspect of the difficulty of this problem. People already talk about these things, of course. But this analysis makes it seem like the obvious response is a revolutionary overthrow of capitalist economies. The knowable problems paradigm encourages us to ask once again: why is this overthrow a hard thing to pull off? One answer that comes to mind for me is that it is not clear where capitalism “lives.” It’s not like you can go break into a building somewhere and steal it. It’s everywhere and nowhere! Obviously, there are many other reasons too.

I think striving to carefully map out the contours of the ethical problems we are faced with is a more productive approach than paying lip-service to unrealistic and/or ineffective solutions. When you feel that you can wrap your head around something, and that you really see it for what it is, I think an intrinsic motivation to continue thinking about it arises. And more thought about ethical problems seems like a good thing! The belief of the knowable problems paradigm, and let’s be clear, this is an article of faith, is that good solutions are a natural by-product of deep understanding.

A question well-posed is half solved. I think in my research, I have been hesitant to commit to a problem too early because I want to ask the right question. I want to find a “pressure point” in the field, where identifying the problem is the real innovation, and where small efforts have disproportionately large impacts. I suspect these are tough topics to find nowadays. Like firewood at a popular campsite, decades of research has picked science clean of its fuel. But I do not want to do the kind of research where monumental efforts have disproportionately small impacts, though unfortunately, this does seem to be the norm. Piling on complexity for the sake of it is not that appealing. Sure, I like understanding complex things, but I am more motivated by ideas. One intuition that I have had is that pressure point ideas lie at the intersection of fields. So if you can cultivate a broad perspective, say at the intersection of computational mechanics and machine learning, you have a better chance of finding a topic which is imaginative and driven more by ideas than sheer difficulty. I think the equation discovery work of Steve Brunton and Nathan Kutz is a great example of this. In my opinion, this work hinged on asking the right question as opposed to doing something insanely hard from the standpoint of theory or implementation. 

Zooming out a little bit, computational mechanics and machine learning are basically the same thing. This is not that broad of a perspective, really. The question that I really want to know the answer to is whether there are pressure point research topics at the intersection of extremely disparate fields like computational mechanics and psychology, or machine learning and philosophy. I imagine these composite fields have not been picked clean of questions. I wonder how one gets to do work like this, or convinces others that it is interesting?

Bureaucracy is an attempt to reduce life to a series of if-then statements. Sometimes this leads to fairness, other times to frustration.

Will there come a time when scientific advancement levels off? Would it necessarily be easy to recognize this stagnation if it happened? Presumably research would still be occurring at the same rate, it just wouldn’t lead to anything transformative. So to claim that scientific progress was slowing down or had stopped, you would have to survey the various fields of science, and argue that in spite of all of the research output, nothing productive was being accomplished. It sometimes feels like this right now. There haven’t been revolutions in theoretical physics for decades, engineers use the same mathematical models as in the past with fancier computers, we don’t seem close to understanding fundamental questions about the brain…in contrast, the history of scientific progress, specifically the 20th century, is full of revolutionary leaps in our technological power and theoretical understanding of the world. After all, we put safe commercial jets in the sky and went to the moon many decades ago…who really cares if researchers are making progress in understanding and modeling turbulence? Academics all seem to toil away within some sub-field working on a problem that no one really cares about, and that probably has no hope of actually being solved. Is it possible this is a period of stagnation in progress? Even more extreme is the idea that we have already picked the low-hanging fruit of scientific discovery, and that we can never again expect the same rate of progress we have “enjoyed” over the last century. Maybe machine learning and artificial intelligence are the posterchild of our era of innovation. But, in the field of engineering at least, it seems like there is more talk about the profound effects of this new paradigm than there is evidence of them. I think its possible that we look back at a lot of the promises of big data and see them as gimmicks. But if data is not revolutionizing our understanding of the world, then what in science is? 

Maybe we just know a lot of things already, and the problems that are left do not yield to the tools of science for very fundamental reasons. Will we ever be able to predict the fluctuations of the stock market? Model biological organisms from first principles? Perfectly predict the weather? Speaking loosely, these are phenomena which are high-dimensional and non-linear. And maybe no matter how scientifically sophisticated we become, there are obstacles to our knowing about the world that are extremely difficult to overcome. Or maybe something big is in the works we can’t yet see!

This event is hosted annually through the law and business schools at CU and open to all graduate programs, though it was mostly law students (see this link for more info). Teams are given a problem related to policy and tech, and tasked with presenting advice to the government from the standpoint of a business or institution in a mock "hearing." Our problem was to advise congress from the standpoint of google (actually, their parent company Alphabet) on how to deal with the problem of increasingly sophisticated AI "deepfake" images and video. The competition lasts about two weeks from problem assignment to final presentation, so it's short but intense. The interesting and somewhat frustrating constraint was that we needed to present a solution that would be believable coming from google, ie one that wouldn't hurt profit. It would've been inappropriate to recommend some draconian measure that would seriously decrease youtube viewership, or ad revenue from google search. So we were not free to think of and propose an ideal solution to the problem of deepfakes—any solution needed to be more or less aligned with google's business interests. My thought was that recommendation algorithms optimizing for attention plus deepfake video would (will?) allow for even more extreme experimentation with media that is addictive. I personally thought this was more of a problem than misinformation. I think if you are seriously fooled by a deepfake video once, you quickly become skeptical of video as a medium for representing reality. I thought the persistent harm would come from content creators being even less constrained by reality in their search for the most compelling and captivating content, irrespective of its effect on our political or psychological wellbeing. I wrote an article exploring these thoughts about deepfakes, which was published in the Spring 2023 edition of Colorado Engineer Magazine. Our team got second place in the competition. 

It is interesting to think about how much computers have changed research in engineering. It seems like very few people work on fundamental model and theory development anymore. After all, there is so much computational work to be done in implementing and piecing together existing models and theories. I think a lot of people would argue that computational work is emphasized in research because there is no longer need for solving small problems or pen + paper calculations. Even if they don’t explicitly argue this, I suspect researchers believe this on some level. But an alternative perspective on the “computerization" of science is the over-used but ever-profound saying that “to a man with a hammer, everything is a nail.” To a scientist with a computer, everything is a computational problem. The computer has emerged as the main paradigm of research, and the most robust way to know things about the physical world. But it is also a bias, and I believe it comes at a cost. To the extent that you spend your time learning and juggling myriad programming languages, jumping through hoops to get different software packages to play nice with one another, and running huge models with opaque innerworkings, you have taken away time from thinking about the physics and mathematics that underlie these heroic efforts of computing. There was a time when engineers were so familiar with the governing equations of their field that they could (occasionally) guess analytical solutions to problems. This kind of insight is totally unthinkable nowadays! Maybe it's true that we no longer have much to do with analytical solutions, but I’m pretty sure deep theoretical understanding of one’s field is as important as it has ever been. Even if this is just to pass on knowledge to the next generation, this understanding is important. When getting your computer to cooperate becomes an end, as opposed to a tool for mathematics to express itself, I suspect it is worth thinking about the trade-off here. Sure, computational work is important for solving practical problems, and often writing code to solve equations builds further physical intuition, but I think too much time spent with the computer distracts us from the fact that all of this work is simply an elaborate untangling of the puzzles posed by partial differential equations. And I think one can learn a lot of important things about PDE’s and their connection to practical problems by being in the habit of moving symbols around on a piece of paper.

The above is taken without permission from here. In fact, my use of this image goes against the explicit wishes of the authors.

In the figure above, mentally replace the phrase “multiscale modeling” with “numerical methods.” The idea of this figure is that when there is limited data but the laws of physics for a system are precisely known, we use numerical methods and not machine learning. An example of this would be finite strain elasticity. The world is not replete with data sets relating forces to large displacements of structural components, but in many cases, we have the right physical models, so we can solve these models with numerical methods. Given that the physics are known, the only benefit of machine learning would be to speed up computations. We are not going to “learn” anything new from the data.

On the other end of the spectrum, we have situations where data is abundant but knowledge of physics is scarce. Here, the term “physics” can be interpreted loosely to mean rules which relate measured quantities of interest. As an example, this might be the domain of recommendation algorithms—huge data sets quantifying online behavior circumvent the need to have explicit rules (for example, that people who like dog videos tend to buy certain products). The data is sufficiently rich to accurately characterize the relationships between variables. In the context of aerospace, we might have inspection data from parts coming out of a manufacturing facility. There may be learnable patterns in the data which indicate that a machine needs maintenance. It may not be necessary to understand the “physics” of this situation, in the sense that we don’t need to know exactly why the data-driven model has made that classification. We only care that the classification is accurate and useful.

The mid-point on the physics-data spectrum is the tricky one. This is where we have rough ideas about what physical models apply, but not their exact form. Similarly, there will be some data describing the behavior of a system, but not enough to train a black-box model. Do we use numerical methods with uncertain physical models? Do we learn surrogate models and use them with caution? An appealing alternative is to use physics and data simultaneously. This is the idea of “physics-informed neural networks” (PINNs) which have emerged from the data-driven modeling community. These models are trained to reproduce training data and satisfy given laws of physics simultaneously. There are two terms in the loss function: one encouraging agreement with training data, one penalizing failure of the network to satisfy a governing equation. The introduction of the physics loss is meant to regularize the problem of learning a surrogate from sparse data. 

But there is a problem with all of this. If we are confident that the governing equation used in the physics loss term is correct, why don’t we throw out the data and solve the problem with traditional numerical methods? On the other hand, if we are not confident that the physics loss term is correct, why would we think that this will usefully constrain the training of the surrogate model? If we are using PINN’s, presumably we are operating in the territory of some physics and some data. Is there any reason to think that forcing a network to satisfy the wrong governing equation helps its performance? I’m sure there are examples where constraining a model with plausible but incorrect physics actually hurts its predictive power compared to a purely data-driven approach. As a simple example, consider the following: we have velocity data on fluid flow, and it is unknown whether viscosity plays an important rule in the observed dynamics. For simplicity, we assume the fluid is inviscid, and use the Euler equation in a PINNs framework to learn a surrogate model to make a prediction for new initial conditions. If the fluid has viscosity, do we think that the Euler equation applies a useful constraint to the model? To me, it is not clear.

Furthermore, solid mechanics poses some challenging problems for PINN’s. If we have force displacement data for a static system, how do we know the physics that govern this response? From a small amount of data, how can we distinguish between 1) linear elasticity with varying material properties 2) hyperelasticity 3) plasticity or 4) damage/fracture? Unlike the fluid mechanics example, these models are not related to one another through the addition of new terms. They are totally different. Does it make sense to learn a surrogate model constrained by the governing equations of hyperelasticity when there is actually damage formation? 

I think one answer to this problem is to parameterize parts of the governing equation along with the surrogate model itself. For the fluids problem, the viscosity could be a learnable parameter. Thus, the model learns the right constraint and a surrogate model that satisfies it simultaneously. This breaks down when the physics are really unknown though. Is there some way to continuously interpolate between linear elasticity and plasticity with a set of learnable parameters? As far as I know, no one has the answers to these questions. But until PINN’s can intelligently reckon with the fact that applying a potentially wrong governing equation as a constraint does not clearly improve training, the usefulness and credibility of these models will probably be limited.

If the hype around machine learning in the world of computational science is justified, then data-driven surrogate models should be able to replace expensive numerical methods for solving partial differential equations. To give the ML world credit, there are other more nuanced uses of data than the wholesale replacement of numerical methods with black-box surrogate models, but this is certainly one use of machine learning that researchers imagine. In many applications, it is critically important that the data-driven model respects the physics of the underlying system. In aerospace, we solve the governing equations of solid mechanics to determine whether a load-carrying part is strong enough for its intended application. In this case, if a surrogate model makes a prediction that does not respect the underlying physics of solid mechanics, it is probably not trustworthy in analyzing the strength of part. This contrasts with an application such as computer graphics, where a surrogate model simply needs to produce outputs that appear realistic—analysis of physical systems in aerospace has implications for people’s safety, so models are subject to extremely high standards of accuracy and reliability. 

In computational models of phenomena like metal plasticity, fracture, and large deformation solid mechanics, we need to solve non-linear systems of equations. Newton’s method is the workhorse for solving these systems, and it does so by taking an initial guess of a solution and iteratively improving this guess until the system of equations is satisfied. In general, there is more incentive to find data-driven surrogate models to replace numerical solutions of non-linear systems because these solves are much more expensive than linear problems. My idea is this—if we are confident that these surrogate models are sufficiently precise to respect the underlying physics of the problems they are trained on, why don’t we initialize all Newton solves with a prediction from a surrogate model? If the prediction of the surrogate is exact, the Newton solver will converge in one step. If the guess is good but imperfect, a small number of Newton iterations will be needed to obtain convergence. It would be interesting to assess the performance of a data-driven surrogate model by the improvement it offered over randomly initializing a guess.

Even if we have no interest in fully replacing numerical methods with data-driven models, it seems that initializing non-linear solves with surrogate models has the potential to speed up computations. It could be that every time a model was run it stored data which was used in continuously refining the surrogate model. This would involve re-training the surrogate every so often. Is there any downside to this? To me, this seems like a realistic use of machine learning in the context of computational mechanics for aerospace applications. Until there is precise error control on machine learning models, it doesn’t seem likely that traditional numerical methods will be replaced in aerospace applications. Unlike in large language models, aerospace applications cannot afford the occasional “hallucination.” But, we could hope to drastically reduce the number of iterations required to obtain convergence in iterative non-linear solvers by initializing with data-driven surrogate models.

The computational mechanics community devotes a lot of energy to techniques for solving problems of materials with microstructure. One example is the FE2 scheme, where the constitutive behavior of the material at every point in the domain is determined by solving another boundary value problem of a square/cube of the microstructure with periodic boundary conditions. This is the standard tool for non-linear multiscale problems, where a constant tensor of effective macroscopic properties cannot be pre-computed. The literature in this area is extremely hard to read and has always left me with lots of questions. I found that using the standard techniques of periodic homogenization, it is not unreasonable to derive the FE2 method for a 1D non-linear bar. This is in my notes on homogenization. In deriving the governing equations yourself, it is easy to see precisely how and why the microscopic and macroscopic equations are coupled. In a sense, non-linear mutliscale problems are more important than linear ones because the microstructure can have a more pronounced effect on the materials response. Linear multiscale problems pickup a slightly unexpected correction to the material properties (unexpected in the sense that direct averaging of material properties underestimates the stiffness), but non-linear microstructures are capable of doing wild non-linear things like buckling. Thus, the introduction of a microstructure could totally change the non-linear response of a structure. This shows up in finite strain elasticity, plasticity, and fracture.

So we see that non-linear problems are more sensitive to the exact configuration of the microstructure, but we know that real materials have apparently random spatial variation. So what good does it do to simulate the non-linear response of a microstructured part if we are uncertain about the exact nature of the material microstructure? The natural way to think about the microstructure in a real part is that it is a random field. In other words, it is probabilistic and cannot be known precisely. It is not useful in practice to very carefully solve the non-linear multiscale problem for one instance of material properties, because we learn nothing about how sensitive the response is to changes in the distribution of material. 

For most problems of practical interest, the microstructure is simply a means to an end. It is used to determine the constitutive relation on the macroscale. And in real parts and materials, the microstructure is uncertain and characterized statistically. Thus, I wonder why we don’t use the techniques of homogenization to upscale the constitutive behavior of uncertain microstructures in an “offline” way, and then forget about the FE2 approach. It seems that practical multiscale problems are equivalent to single-scale materials with stochastic constitutive relations. These uncertain constitutive relations come from non-linear microstructures, but could be pre-computed, interpolated, and stored in some way. Then, uncertainty quantification techniques could be used to understand the sensitivity of the structural response to configurations of the material. This approach seems to be more clear about the fact that 1) the microstructure is really just a fancy constitutive relation and 2) that in non-linear problems and real materials, the response is sensitive to the details of the microstructure but these details are only known probabilistically. To reiterate, I think there is a strong connection between multiscale problems on real parts and materials with uncertain constitutive relations.

At one point a few years ago, I was interested in thinking about what humor is. What makes something funny? Why would a social animal like us humans evolve this behavior? I guess the goal was to understand what purpose humor served in our social relations from an evolutionary standpoint. I don’t actually know anything about evolutionary biology, but I wanted to have a sense of what humor was all about. After all, it is a pretty strange thing. It is hard to define what is funny. But in practice, we are extremely motivated to explore what is funny through making jokes with our friends and acquaintances. 

There are a lot of stupid theories of humor out there. For example, that jokes serve as means of relieving internal tension. This is called the “relief theory” and is pretty obviously wrong. Another common search result is the “incongruity theory” of humor, which basically states that something is funny when it is out of place. Examples of this would be absurdity, or situations which defy our expectations in some way. This seems a little bit closer to the truth, except that it doesn’t explain why certain incongruities are funny and others are not. One webpage supported the incongruity theory of humor with the example of using a toilet brush as a tooth brush. This is incongruous because our expectations for the word “brush” are context dependent, and can be brought into conflict with one another. The only problem is that this situation is not that funny. 

A good summary of theories of humor can be found here. Towards the end of the list, there is a theory titled “humor as play.” I don’t remember anything about what this section said, but the title alone was the nudge I needed to form my own theory. Incongruity, absurdity, and defied expectations seem to be key elements of things that are genuinely funny. But not all incongruity is funny. What separates the two? I think one answer is that funny incongruities are ones that run the risk of genuinely confusing us. No one mixes up their tooth brush and toilet brush, but if a comedian effectively dramatizes the confusion/chaos that arises from holding inconsistent beliefs about something plausible, it is funny. It is funny because they are showing us something insightful about how we operate. Comedians dig into our language and thought about the world looking for inconsistency, hypocrisy, confusion, and misconceptions. Then they unearth these findings, act them out, and we laugh. I think the laughter response is a sign that something true, and possibly important, has been brought to the surface. Sure the word brush points to two very different objects, but this is not important to mention because no one gets into trouble as a result of this. And this is where the idea of play comes in—I think play in the context of sports or games, viewed from my amateur evolutionary perspective, is about showcasing physical skills, decision making under pressure, an ability to compete and cooperate, etc. We are attracted to these things because these are obviously desirable attributes, and engaging in play both hones and advertises these attributes. So my argument is this—humor is play, but instead of the medium for play being games (on a board or a sports field), it is language and thought. Humor plays with our thinking about the world. It makes a game of finding inconsistencies, confusion, and strangeness in our relationships to the world. Instead of showcasing physical process and social skills, I believe humor advertises to other people that we are thoughtful, discerning, and self-aware. I see it as a very intellectual form of play. When someone excels in a sport, it communicates they are physically healthy, disciplined, capable of understanding and following rules (thus socialized), and so on. When someone is funny, I think it demonstrates a mastery of language, a broad awareness of self and others, and a certain kind of wisdom. Things are funny when they get at some truth. 

I printed out my Heidegger reading in preparation for the Thanksgiving flight home to Pennsylvania. I was looking forward to having this time blocked off to devote some serious, undistracted attention to “The Question Concerning Technology.” After all, Heidegger is one of those names that is shrouded in mystery, probably as the result of being referenced somewhat regularly, but always in an ambiguous way. Two minutes into beginning the reading after the plane took off, a well-meaning but irritating undergraduate at the Air Force academy sitting next to me asked “is that an academic article?” and in hearing that it, in fact, was an academic article (about technology), he proceeded to inquire if he could read along with me. While very unwelcome at first, I surrendered to the absurdity of this strange arrangement and we began to read…

Unfortunately, we were both very confused throughout the reading of this very academic article. I am disappointed to report that reading “The Question Concerning Technology” unequivocally confirmed my suspicion that Heidegger is an author whose primary purpose is to elevate the reader’s standing at snobby dinner parties, rather than to communicate anything of utility. To me, this essay reads like a parody of philosophy. Though I am open to hearing it, it is very hard for me to imagine a compelling argument in defense of this style of writing. I feel confident you know what I mean by this style if you have read the essay. One argument I can imagine is that writing of this sort was more common in the past but has aged out, leaving modern readers a bit in the dark. This piece is not very old though (1950’s), and were there not myriad novelists from this generation who knew how to communicate clearly and viscerally in a timeless manner? I suspect a more common defense of this style would be that it is in service of, or otherwise necessary for, rigorous philosophical thought. But do we really believe that such labyrinthine, indirect, alien prose could possibly be in service of clarity of thought? Philosophy is not quantum field theory. I refuse to believe that profound ideas about deeply human questions must hide behind such opaque writing. In gazing into a work of philosophy, I expect to see myself and my humanness focused and reflected back. Gazing into this work, I see confusion, pretension, arrogance, and emptiness.

Why do we feel the need to apologize for works like this? Why do we apply entirely different standards of judgment to historical pieces of writing (and movies) with reputations like that of Heidegger? Do we honestly think that if it were published today, “The Question Concerning Technology” would be heralded as insightful? According to lots of lists, Alfred Hitchcock movies are the best ever made. Given that the overwhelming majority of movies have been released since Hitchcock, most of which have benefited from decades of technological and artistic innovation, this seems unlikely from the standpoint of probability alone. Yet, there will always be someone (maybe someone reading this right now!) who defends claims of this sort. I repeat: why are we so enamored by outdated, difficult, or otherwise clunky works of the past?! It seems like we, like many groups of people throughout time, are compelled to elevate our ancestors to a mythological status. We see something sacred and pure in the past, taking the authority of intellectual ancestors for granted, while feeling perfectly at home criticizing contemporary popular culture.

Perhaps what people mean by Alfred Hitchcock movies being the best is that looking back, they are some of the most influential. To this point I would respond: if that is what you mean, you should just say that…But on this note, I see this historical lens as a more productive way to engage with works of the past, and maybe one that can be applied fruitfully to Heidegger (though I am still skeptical of this). In addition to thinking about a work through its influence, one can also apply the standards of its own time to that work. Perhaps it was extremely novel or forward-looking at the time of publication, even if it no longer feels that way to us. I think reading Freud is a good example of this: the fact that many of his ideas feel boring and self-evident is a testament to the massive influence they had. Though his work was very unusual at the time, it has so thoroughly integrated itself into our understanding of the psyche that it feels obvious. In order to appreciate Freud’s cultural and historical import, we need to look past the fact that it is somewhat unpleasant to read, and in fact see that this unpleasantness (because it seems obvious) is actually an incredible testament to its importance! Turning back to Heidegger, instead of debating whether his writing was enlightening from a personal standpoint, maybe we will discuss the important ways in which it influenced the discussion around technology at its time of publication. 

Post-script: I believe that no ideas are irreducibly complex, and that a primary goal of writing should be accessibility and clarity. I think it is possible to write to give readers a felt “sense” of the ideas, and I would like to believe that thoughtful communication about complex ideas can still target intuition without sacrificing rigor. When I read papers where material is presented in the most abstract, mathematical, detached ways, I wonder whether the authors have any intuition for what they are talking about. Is this just inert symbol manipulation, or do these people have a gut-level relationship with their work? When I read someone like Douglas Hofstadter, or watch videos from Grant Sandersen’s YouTube channel “3Blue1Brown,” I am in awe of the depth and breadth of their understanding. These two individuals are master communicators of complex ideas, breaking seemingly intractable intellectual hurdles into an interplay of familiar moving parts. They tell stories, give examples, speak clearly, only introduce abstraction as needed (and feel the need to justify it when they do), and are explicit in their desire to connect with their audiences. This is what constitutes good “philosophy” to me–the ability to make complex/abstract/mysterious things feel relevant to everyday life, and then to demonstrate that these complex/abstract/mysterious things can be apprehended with relatively “non-mysterious” theoretical tools. It is disappointing that so many intellectuals (scientists, philosophers, etc.) do not feel the pull to be storytellers in the way that Douglas Hofstadter and Grant Sandersen do. It is not surprising that these two enjoy fairly diverse audiences, whereas Heidegger is relegated to specialized academic circles and the most masochistic of dinner party-goers. 

Post-post-script: I would like the record to reflect that I appreciate the opportunity to engage with a difficult thinker like Heidegger. At this point in life, the likelihood that I will pick up an essay like “The Question Concerning Technology” on my own accord seems to be going down. In spite of all my protestations, I enjoy furthering my thoughts about the challenges and rewards of engaging with historical works. Like an angsty teenager hellbent on rejecting everything put in front of them, I feel that I further develop an intellectual identity through articulating my aversion to philosophy of this sort. And even if the only outcome of reading Heidegger is to double-down on a personal commitment to clear, simple, and connective communication about ideas, there is certainly benefit to this. I think commitments like this are more compelling and credible when they are arrived at through personal experience and trial-and-error, as opposed to being pure articles of faith. Personally, I think a philosopher or scientist communicating to the public should first show that the world is more complicated than you think, and then secondarily demonstrate that this complexity can be broken down into digestible parts. It is the job of the thinker to find and articulate complexity, but also to “tame” the intellectual wildernesses they inhabit. Obviously some problems are just hard, but this “taming” could simply be communicating a sense of the right way to conceptualize that difficulty (this is the “knowable problems” concept). The governing equations of solid mechanics are linear, whereas those of fluid mechanics are non-linear. This is a simple mathematical property which explains why fluids do some crazy things that solids don’t, even without offering a “solution” to that problem. There is much more to say here, but the point is this–complex things do not prohibit simple formulations. This is especially true when what we are trying to formulate is not a “solution” to the problem, but rather answer the question “why is this thing so complex?” I think when problems resist solutions, addressing this question is a worthy “backup” plan. In our attempts to make sense of the world, I see our job as to find the complex among the simple, and the simple among the complex. This interplay is productive from the standpoint of generating knowledge, but also satisfying aesthetically.

I am seeing that one of the benefits of studying myth is that you become attuned to a large number of stories. These stories can be used to frame thought about whatever thing you are trying to make sense of. With more stories available to you, you can more carefully tune and optimize the lenses you apply. If we are going to play the potentially dangerous game of making sense of the world with stories, and particularly those of myth, we should at least be careful to apply the right stories. And there sure are a lot of stories out there that could apply. (If you are concerned that stories are not a sufficiently rigorous tool to make sense of things like a company or nation, I would encourage you to try to describe these things “scientifically”). I think someone slips into an ideological view of the world when they have limited stories that they use indiscriminately. This can be intentional or accidental. But having a breadth of stories available as potential interpretative frameworks allows one to be more deliberate and granular in making sense of things. I wonder if, in the face of a dramatically modernized world, there are some situations which are truly novel, for which old stories do not apply? The burden of identifying and articulating these situations falls on contemporary novelists and filmmakers, I suppose. One example I can think of is Lord of the Rings, which explores the danger and corruption of excess power. Can you imagine hunter-gatherers worrying about having too much power? These stories seem to be those of a society anxious about the new and ambiguous power of technology and science to form and destroy the world. Though people have worried about the influence of new technologies for a long time, technological power on the scale of the atomic bomb (for example) poses unprecedented existential risks. I wonder if this myth is fundamentally modern?

It is hard to think of the word that feels right to me for the concept that “psychic entity” points at. Possible synonyms: worldview, belief, idea, schema, concept, mindset, perspective, zeitgeist, culture… even “vibe” seems to get at this thing. First off, how have we not developed language to speak critically about our basic commitments and orientations to the world?! It is clear that religious and non-religious people “see” the world differently, yet it is difficult to carefully articulate what that means and what effects it has. The notion of different “sight” can be illustrated simply by showing that different starting assumptions lead to different interpretations of the same experience, but this does not get at what it is like to hold these differing worldviews. One example might be a near-miss accident which the religious person attributes to miracle or Providence, and the non-religious person to luck. How do these first principle beliefs/assumptions influence our experience of the world? What does it feel like to “see” Providence vs. luck. Do answers to these questions influence how seriously we take certain sets of beliefs? I occasionally wonder what it would be like to treat beliefs as the stuff which mediates between the outside world and feeling/action. In other words, don’t judge beliefs based on their truth content, rather on the results that they produce. Results would be in terms of personal, internal experience, action, and relations to other people. I think it is reasonable to view mythologies through this lens–as sets of beliefs which are optimized (constrained by the basic structures of the human mind) for desirable psychological/social results as opposed to “truth,” by which I mean scientific truth.

I guess the question I am asking is: what is it like to hold certain worldviews? Maybe we could label this line of inquiry “phenomenology of belief,” indicating interest in the felt consequences of different lenses through which to interpret experience. I worry that many of the terms above (worldview, perspective, concept) are much too cognitive/abstract/individual in that they ignore the role of “outside” things like the physical environment, technology, danger, etc in contributing to experience. To even begin to think through this question, it is necessary to make these “worldviews” (psychic entities) visible.

At this moment in history, I think a phenomenology of belief is important because we have great power to shape our physical, social, and cultural worlds, but I think we are seeing that physical comfort is not a reliable proxy for general well-being, and thus is not a good choice of “optimization variable.” We need to be aware of the worlds we collectively imagine and create before they can be judged and formed. So the question of how to make these vague, abstract things available to the senses is important, and challenging! I think the best starting point is to motivate the importance of formulating some kind of model of a system of belief. I have briefly tried to do this–it is important to become aware of psychic entities because we have power to shape them, and the psychic entities of the modern world do not reliably lead to holistic well-being. The next step would be to de-naturalize certain assumptions about the world: why do we believe that the continued production of new technologies is a good thing? Is this based on evidence or faith? 

I thought about the question of how to make psychic entities available to the senses over the course of last week. Stories and art seemed to be the primary answer, which I found interesting because I had never thought about art as one of the only vehicles for making the contents of the unconscious apprehensible. Great artists do seem to be uniquely in touch with the spirit of the times, so that through their insight and technical artistic skills they can bring out things we all feel but can’t articulate. My concern is that this artistic insight is not sufficiently conscious in order to understand the psychic entity it makes visible. If we want to exercise control over what psychic entities we let into our lives (and choose how we relate to the non-negotiable ones), it helps to have a more analytic understanding of these forces. This was my motivation for trying to map out relevant psychic forces and their interconnections in class. It is an attempt to explore how “vibes” (it is unfortunately hard to think of a better word!) emerge out of how an individual/culture answers concrete questions such as: What is the relationship of the individual to family? How much education is required for dignified work? Where and why do people gather in community with one another? How are the future and past conceptualized? Out of this tangled network arise fantasies, stories, experience, art, desire, innovation, and more generally, a “sense” of what life is like. I wonder if attempting to make this map would be a useful exercise for identifying strengths and weaknesses in given cultures. This could be cultures of science, modern religion, tech, indigenous people…

I find the incessant discussion around consciousness tedious and academic. Before launching into this, I would like to be clear that I really enjoyed these two readings. The first was the James Alan Gardner story from this anthology. The E.S. paper was a nice example of how publications in scholarly journals need not be pedantic and lifeless (hopefully this freedom is not specific to philosophy, it would be interesting to see science papers pull this off). That being said, I think that consciousness is such a popular target of philosophy simply because it is so accessible. Pretty much everyone has some awareness of the strangeness of the mind, the self, free will, etc. In contrast to many other topics philosophers might set their sights on, the question of consciousness requires no pre-requisite training to appreciate. My suspicion is that many other real and potential philosophical questions require more specialized knowledge to even formulate (let alone address). Questions of consciousness are to philosophy as pizza is to catering–not risky, creative, or novel, but generally agreeable. 

Consciousness is a word, and like all words (perhaps apart from symbols in formal systems), its meaning is unclear. Words are tools we invent to facilitate communication about the world, and it's probably fair to say that they stick around when they succeed in that aim. We should not be arguing whether reality conforms to our words; this seems to be placing the real phenomena and the word used to describe it on equal ontological footing. I don’t know exactly what the word consciousness means, but it has certain metaphysical connotations which make debates sound more spiritual than academic. For the putatively conscious entity in question, we might pose more practical questions such as: Can it be held responsible? Does it feel pain? Does it have any internal experience to speak of? Is the whole greater and/or different than the sum of its parts? I think what people often mean by consciousness is “like us humans” in the sense that one could enumerate all the apparently relevant features of human consciousness, but then hesitate to classify a strange entity which meets these criteria (such as a nation) as conscious. This is simply because it is not human-like. And I think that’s fine–we are tricking ourselves into thinking reality should have neat conceptual boundaries simply because our use of words implies this.