Home Creators Posts Import Register Favorites Logout

Content

while this chapter didn’t feel like it contributed much to my understanding of AI (not sure if others felt similarly? lmk), a lot of points it made tied back to some of the ideas from my technocracy video via what is basically a breakdown of the limits of data science:

Techno-optimists have argued that statistical tools can replace the scientific method because of the sheer number of correlations they can find in data that no human scientists could conceive of.

So for context, this is how a lot of research in astronomy works these days. The time I started doing astronomy research was around the beginning of the “big data” era of astronomy, in which, by combining the increasingly massive astronomical surveys that were being developed in the field with new techniques in machine learning, there came to be a whole category of astronomy whose research boiled down to “run this data set through this model, see what correlations it spits out, and try to come up with a physical explanation for them.” This was very normalized to me, and I don’t want to say that no interesting results came out of this kind of research, but until reading this I hadn’t thought of it in terms of “replacing the scientific method”, even though it kind of is. Another kind of unfalsifiable astrophysics, if you will, starting with the extremely open ended question of “what’s interesting about this data?” rather than any particular hypothesis.

Of course, while this can work for a lot of astrophysical contexts where the behavior of the system in question is governed by a handful of deterministic physical laws, it doesn’t translate to all contexts:

In the 1950’s MIT professor Jay Forrester pioneered an approach called system dynamics, which tried to extend the use of simulations to social systems. He tried to use simulation to model a whole city, with an eye to solving thorny social problems like urban blight. Here, things went badly wrong. And we don’t have to speculate too much to find out why—a city is far from a physical system.

I feel like this is the technocratic impulse in a nutshell: X method works for stars/particles/whatever, so let’s apply it to human beings—inevitably to unsuccessful ends. Again tying back to my technocracy video, it makes me think of Howard Scott and his insistence that society is metrical, but it also makes me think of something interesting that I saw happen during the early days of covid. Basically a bunch of the astronomers who did this kind of “apply machine learning to big dataset” work started trying to do epidemiology research, regardless of whether they actually had any domain knowledge of epidemiology. From my understanding, most of this research was, at best, pretty much useless—kinda a flimsy gesture of ‘look I’m helping’ from people whose work otherwise has decidedly zero impact on society—and at worst, actively harmful—see, the physicists at University of Illinois who made their campus’s covid re-opening plan, which ended up being a massive failure. I think the examples discussed in this chapter  did a good job of illustrating the importance of domain knowledge in the understanding and possible applying of algorithms to whatever context (i.e., the specific reasons predictive AI would fail in a given situation), and at the same time, the ignorance of some data/computer scientists to that reality. 

In a similar vein, this bit was kinda funny to me:

Before Matt presents the results from the Fragile Families Challenge in academic talks, he usually asks the audience to guess how the teams performed. Some of the most optimistic audience members are computer scientists and data scientists, who have seen how well AI can perform in other domains. When confronted with the disappointing result, this is also the group with the most follow-up questions and ideas for improving predictions. One of the most common questions: Are samples from four thousand families enough?

Like, of course the only possible problem in the eyes of the data-pilled crowd is just that they need more data. I guess that does kind of tie a little bit to the ideology driving our current AI moment. Like, yeah gen AI or whatever isn’t perfect, but if we just steal MORE data and have MORE computing power and MORE advanced algorithms, we WILL eventually achieve AGI—rather than concede any inherent limitations. Interesting to read this particular explanation to why that kind of reasoning is flawed.

next week's reading is finally getting to gen AI, verrrry curious to see what we get there!

Comments

Luis Cabral

Before reading this chapter, I saw an analysis of a recent interview that the CEO of NVIDIA did. It's generating some pretty big discourse because (to massively oversimplify what was said) it suggested that even the CEO of NVIDIA (Jensen Huang) was implicating that AGI was not coming anytime soon. The reason why is because when asked a question about selling chips to China, they framed it not being a good business decision because of the "market" rather than what some others were stating in the US government. The sentiment expressed by some government officials and other influencers was that it wasn't a good idea for NVIDIA to sell chips to China because they thought of the chips themselves as "weapons" rather than just chips. Jensen pushed back on this though, stating that China already manufactures many chips and that it was a bit of a ridiculous sentiment to call chips weapons; and, that it was wrong to open the market to China because they are competition that will further innovation away from NVIDIA in an already competitive market. I bring this up because this chapter feels to me as sobering as hearing Jensen essentially confirm that AGI is either a myth or not coming anytime soon. Jensen's idea operates under the assumption that the current state of AI is NOT building towards something inevitable and bigger than what's available but rather, that a cap has been hit and it's time to manage it. To be honest though, it felt like a lot of the issues expressed with predictive AI this chapter falls under a wing of study that my friend and I believe all Engineers, Mathematicians, Data Scientists, etc. should also study under: Humanities. The questions and factors pose were of a humanitarian nature, like the Fragile Families Challenge not taking into account supportive adults outside of the home contributing to positive changes. Or COVID responses not taking into account the stress it was putting on the general population and the amount of agitation growing amongst the public. (Side note, I didn't know that the term Butterfly Effect actually came from meteorology!) With the COVID responses, it highlighted for me the weight that predictive AI can have without addressing the spontaneity and randomness of genetic mutation. For weather prediction, it's been interesting to hear that now our 6-day forecasts are about as accurate as our 5-day forecasts from years ago. Overall, it was a solid chapter with lots of examples that can be used in conversations when critiquing AI amongst friends and family. The point that stuck with me the most was questioning how a successful prediction is determined. Is it if most of the criteria to verify accuracy is correct or is it a positive outcome overall? With other technologies, it's easy to test if something is right or wrong with the scientific method. With predictive AI, it's using past data to predict future outcomes (in some scenarios without deterministic laws at play). So how does high accuracy contribute to positive future outcomes? If accuracy is determined by the standards that humans set, what guardrails would there be to show that the "100% accurate" outcome is not oppressive or biased? To me, no matter how much AI or technology is available, it will always be the expression of a human's or system's intention/design. Predictive AI doesn't seem like it can be used to determine the future accurately when it comes to human outcomes. I know that the world may not want to hear it but to me, if something isn't working it's time to either shift the purpose of the tech or scrap it.

phil e the theyby

I felt this chapter was less "punchy" in terms of ideas I hadn't really considered, but I felt that it made its points well and organized them effectively. I do think its core attack on the assumptions of data science (particularly the assumption that it's possible to obtain enough data to make accurate predictions about the future) was extremely effective, looking at both the particular variables we look at (that sometimes tiny moments can have huge impacts on us) and the "statistical power" argument (it may not be possible to actually obtain the "dataset" we need to make meaningful predictions). I found the part about aggregate data particularly compelling in highlighting how the tools of predictive AI fundamentally don't work when predicting population-level events, because there isn't enough human history to make a meaningful data set (and even if there were, there's so much change that trends from one data point may not carry over to another). "In the past, we relied on the judgment of a record producer or a TV studio. Today we rely on the wisdom of the crowd, the vagaries of the algorithm, and a large element of randomness." (88) I think a lot about the loss of "gatekeepers," not only in terms of the art we're exposed to and the media that creates culture, but also in terms of how we think or what we consider "legitimate" knowledge. Personally, I LOVE the democratization of knowledge and the fact that we can pull from more perspectives. I think that, if we have the skills to reflect for ourselves and parse through the information we're receiving (and its consequences), we're able to be a lot more thoughtful in the actions we take based on the information we engage with. And Dr. Fatima and other similar creators are engaging in that sort of pedagogical process that draws from a lot of different fields and ways of knowing and communicating, but isn't strictly bound to any one of them. That's fucking awesome, and it's the kind of thing I'm happy to give my money to. But at the same time, a lot of "information peddlers" can be extremely predatory, preying on the worst fears, anxieties, and unmet needs of people who haven't been taught those skills for self-reflection and understanding the consequences of their actions. And for all their shortcomings (of which there are many), gatekeepers serve as a guardrail against some of those abuses. We're in an era where we rely on the wisdom of the crowd, algorithmic processes, and randomness to determine what we believe. I don't think it's possible to go back to the era of gatekeepers. But given that, I think it's incredibly important that we're supporting people (including ourselves) in learning how to think and how to parse the information they receive. I don't know the best avenue for doing this (my first impulse, which is to angrily belittle someone when they're *not* being critical or reflective with the information they consume, is actually EXTREMELY ineffective), but I think it might be...modeling? Like, demonstrating not only our own processes critically engage with information, but also how we're sometimes wrong, and what sorts of emotional needs/biases play into that. Dr. Fatima does a good job of this in how she talks through her thought process, and I've seen Hank Green do a pretty good job of this as well. Fun tip for the next chapter: Any time the authors write "gen AI" (or "generative AI," since I don't think they actually write "gen AI" very often), mentally replace those words with "Jenny," pronounced in the way Tom Hanks says it in Forrest Gump.

Woke AF

The goal of educating people about this market term "AI" is important because people do not really understand how AI works. It's not a flaw. For example I know the the base code for Microsoft Office is C++, but I also know that you can use Visual Basic, Python and Javascript to extend and automate many of its functions. Since MS Office only has one job, no one really has to know how it works or what the base code is. This is not as true when it comes to Predictive models. People need to be informed on what these things actually do and why they are not the best at doing it. The whole "Let's guess what will happen." model is dangerous. (YIKES, I just realized that Compass is literally no better then a daily newspaper Horoscope. That's scary.) Mostly because if the people programing these things have bias, based information is they are not going to bother researching. (The author didn't really research Justine Socco, so his information supported a very bad idea without him even thinking about it.) Considering all the "trust me, bro" information that comes with a systematically racist society is this really a shock? Law enforcement using predictive data made by for profit companies, or medical professional relying on predictive application to help with diagnosis, all programed by people with highly extremely racist ideas - like "black people feel less pain" - is going to lead to application that work based on human bias. We are the actual God in the machine, after all. It doesn't have to be racism, it could be classism too. Insurance companies predicting premiums and payout based on how much people pay. (There is paper or video essay about this somewhere, but can't think of where right now. ) This might happen through no fault of their own, but does that make the results any better? For example if you believe that black people are more prone to commit crimes because scholars and professionals are telling you this, or your parents teach you this based on data built on inherent bias, you'll get predictions like "they are poor therefore more at risk. Politicians use predictive models too, which is why it looks like they are not "listening to the people." They have a model they tells them "X amount of this demographic will vote for "Y." And the programmer don't have to be rage a-holes They just have to believe a thing, then how they set up these models will reflect that belief. Then the programmer is creating an algorithm that might do things like have their image search only pull up pretty white women when searching for "beautiful" and pulling up black women when searching for "ugly." (That was a huge oversimplification on that issue, but it did happen with Google.) What happens when someone who believes in their hearts that black people commit more crime and they are the ones coding and programing the predictive models? What happens if they code it that way on purpose. We've all met tech-bros, it's not a stretch. So teaching people what predictive models are, how they work and why they are dangerous AF is important. Consider that knowing how they work might inspire people to innovate ways to fight and beat it, and channels like yours will provide that inspiration because people like your content. Our current government is relying on these predictive models to do awful things. Justified consent because tech bros created the model that implies consent. We've got to know what we're fighting out here.

Laura Robinson

This chapter I thought a lot about the physical cost and limitations of storing and processing vast amounts of data and the presumptions of the more data, more GPUs set that whatever insights they might find will be more valuable, economically or humanisticly, than the inputs even if it does work out, or improves predictions but only in a limited way. Also, theory-ladenness of observation. You don't collect pure sense-data, it would be overwhelming, you collect data based on what you expect to be relevant. To use the example at the beginning of the chapter, you're not going to discover barometric pressure no matter how many thermometers you monitor. The authors seem to think that the limitations to the data is instead just in the practicalities of gathering it and using it, and the uniqueness of personal experiences which is true, but not the whole story.

Laura Robinson

On the one hand, Huang may be backing away from AGI as a possibility, on the other hand NVIDIA has been promoting "world models" as an alternative use for their AI GPUs, which are all about making predictions about chaotic systems.

phil e the theyby

Regarding the “theory-ladenness of observation,” I think there’s something deeply human about longing for AI to remove the theory and give us an “objective truth.” As a species, we rely so deeply on stories to make sense of the world. But when the modern world constantly shows us that the stories people in power use are actually bullshit and are super harmful, it makes sense that people long for something tangible that removes that deeply fallible, “human” element to it. I think the authors do a really good job at demonstrating why predictive AI can’t work, but they don’t really point to how to meet that human need for clarity and certainty (or at least not yet, and that doesn’t really seem to be their focus) once you shatter the illusion of AI being able to hand us the divine, biblically-accurate narrative about how the world worlds. And maybe that’s our job? I think that constructing meaningful narratives about the world that make sense but also have room for self-reflection and humility feels like it’s kind of a core idea underlying Dr. Fatima’s (and a lot of other left-leaning video essayists’) work, even though it’s not explicit and each person focuses on a different corner of that broader problem.

Luis Cabral

That is true as well! I just read another comment about belief and it seems that for Jensen and possibly other AI companies, it's not about delivering a promise but more about playing into the belief that a certain promise can be delivered. Umar's comment stated, "They just have to believe a thing, then how they set up these models will reflect that belief. " As well as the chapter (and other comments) pointed to an attitude of "more data will mean more accuracy". Even if it's not the main purpose of the authors' of this work, I keep finding myself asking lots of follow up questions about ethics and solutions to the problems presented by predictive AI.

Dr. Fatima

love the connection to theory-ladenness of observation Laura! yes that's absolutely at play here, nice catch also can confirm trying to help people make sense of their experiences is definitely a central goal for me Phil, glad that comes across clearly lol

Dr. Fatima

interesting, can you share a link to the interview/analysis Luis?

phil e the theyby

@Dr. Fatima Ok but like, making sense of our experiences specifically in the context of the collapse of trust in global grand narratives, where the commonly-used alternative is “hand over the control of narratives to Big Data (or whatever demagogue is politically aligned with you) because the world is too big and scary for us to understand.” So my statement is slightly less shallow than it sounded :P

Anthony Perez Eisenbarth

Chapter 3 is the most technically substantive part of the book so far, and as someone who works in data science professionally (alas as Adorno says, “There is no right life in the wrong one”) and comes to this from a heterodox economics background, I find myself in the unusual position of being simultaneously validated and horrifyingly amused. The Fragile Families Challenge is the centerpiece, and it is genuinely illustrative of the problem. 160 competing research teams, ten thousand variables per subject, the most sophisticated machine learning methods available, and the best models barely outperform a four-variable regression model using race, maternal education, marital status, and prior GPA. Decades of sociological theory encoded in a handful of variables beats the data-pilled approach handily. As someone who spends a meaningful portion of my professional life building predictive models, I can confirm that this finding is not surprising to anyone who has actually tried to predict human behavior with real data rather than benchmark datasets. The gap between what machine learning can do on a carefully curated competition problem and what it can do on a genuinely messy social outcome is enormous, and most practitioners know it even if the vendors selling these systems prefer not to advertise it. What I find somewhat ironic is that this is essentially the same critique heterodox economists have been making about mainstream economics for decades: that the fetish for sophisticated technique tends to displace serious engagement with the underlying data-generating process and the institutional conditions that produced it. The book gestures at this without quite landing it, which by this point is a familiar pattern to me at least. The deeper issue, which neither the authors nor most of their readers will recognize, is that the Fragile Families result is not merely a finding about machine learning. It is a confirmation of something Alasdair MacIntyre argued in After Virtue fifty years ago about the fiction of social scientific expertise more broadly. MacIntyre’s claim was that the authority of the social scientist, and the corporate manager (the MBAs and the like) and technocrat who deploys social scientific knowledge, rests on a claimed stock of law-like generalizations with predictive power. That stock, MacIntyre argued, does not exist and cannot exist, because human social life is structured by sources of irreducible unpredictability that no accumulation of data or sophistication of method can overcome. Narayanan and Kapoor seem aware of this. But what the Fragile Families result demonstrates empirically is precisely what MacIntyre demonstrated philosophically: that the social world does not yield to the covering-law model of prediction that would be required to justify the authority claims of the expert. This connects to something I am actually working on right now in my own research on the so-called credibility revolution in economics, the movement toward natural experiments, instrumental variables, and quasi-experimental designs that has dominated empirical economics since the 1990s and is often presented as having finally put the discipline on rigorous scientific footing. The credibility revolution improved internal validity within narrow empirical designs. It did nothing to resolve the deeper problem MacIntyre identified, which is not about method but about the nature of social causation itself. You can have an impeccably identified estimate of a local average treatment effect and still have no law-like generalization capable of supporting the kind of policy engineering authority the profession claims. The gap between a well-identified regression discontinuity and a basis for managing a social system is precisely the gap MacIntyre was pointing at. The data science version of this problem is visible in the Fragile Families result: 160 teams with every method available, and the social world remains, in MacIntyre’s phrase, “in a very literal sense out of our, and indeed anyone’s, control. No one is or could be in charge.”

phil e the theyby

This was a really interesting read, and it was cool (if upsetting) to hear the parallels to economics, as someone who’s not particularly informed about that field. I think the thing that’s interesting to me is that, even if you believe in a truly deterministic universe, the perspective shared by the authors of this book and by MacIntyre (based on your description) is that *you can never use those deterministic processes to make predictions, because the rate at which conditions change is so rapid that it’s impossible to meaningfully generalize from one situation to another*. And that doesn’t mean that all predictions of social phenomena are completely useless (I’d argue that the current evolution of fascism has been predicted for decades by certain folks on the left), but it does mean that those predictions are almost always really tentative and subject to a LOT of error.

Jacques Vincelette

p. 61,"systems were structured ways of predicting the future. They were based on the positions of the stars, or tarot cards (...)". I graduated class of 1997 Mechanical Engineering with a minor in Mechatronics with hopes build space robots. I settled for simulation, with reliance on Monte Carlo models and game theory for group behavior. Predictive computer models have worked like elaborate tarot decks, and black box neural-networks offer another obscure layer.

Black EnviroNerd

Similarly to Dr. Fatima I’m not sure how much this chapter really expanded or challenged any ideas I had toward predictive AI. I think the examples were pretty good at establishing and proving the authors’ main point that AI can’t predict the future. The main question coming to mind for me is if the act of attempting to predict the future, or more particularly human behavior a conservative exercise at its core? There is a sort of underlying cynicism to me about the idea that there’s a certain set of variables that an algorithm can use to predict individual outcomes. To jump off the point the authors used in previous chapters as to how these exercises can give a sense of certainty in an otherwise uncertain world. I think extending this idea further is that these predictive algorithms can also give a sense of control especially when you fold in decision makers who’ll cite these results to justify their actions, whether they be austerity measures or mass incarceration. I can recognize the good intentions that may go into making these algorithms but the fact that they’re so easily co-opted gives me pause and makes me question the validity of performing these exercises. Especially when these results are used to make critical decisions about people (who goes to jail, who gets resources in college, who gets treated for a medical condition etc...)

Laura Robinson

I'm not sure it's exactly predicting the future that is the purview of the Conservative, what are theories of change but predictions of outcomes for systems of people based on actions taken. I have heard that statistic about the percentage of people who need to organize for something for it to be inevitable from lots of people who are pushing for genuinely progressive if not quite leftist ideas. What I would suggest is sorting people into a hierarchy of worthiness which can be done with or without some veneer of scientific prediction tends toward the reactionary. But I won't deny that this rhetoric works much better for certain audiences than naked stereotype and bigotry.

AJ Biffl

That's a really cool question! I think you're definitely onto something - prediction fundamentally requires the view that people's behavior in the future will largely be the same as it was in the past - no allowance for large scale growth, development, evolution, change. It is fully seated in the assumption that the world at its largest scales, and the people in it, will remain constant, the past conserved. And then beyond that, yes I agree it all fits into this headspace of justifying hierarchical decision making. Building models to predict human behaviors rests on the assumption that such an entity exists that can decide and adjudicate and enforce and regulate based on those predictions. Predictive models are gauged on their practicality, and so require an apparatus to put their results into practice. The brain justifies the hand, the hand feeds the brain

AJ Biffl

MacIntyre's ideas sound really cool! Does he elaborate on exactly why such a complete set of predictive laws is impossible? I wonder if it's related to the phenomena the authors described about social media videos, that if there were a predictive metric then people's behaviors would change to target the metric and it loses its predictive power You mentioned "data generating" processes and institutional effects, could you elaborate on that?

AJ Biffl

This chapter definitely felt like a rehash of the technocracy video, but it's a discussion that's not going anywhere. This "eight billion problem," there's real tension there. On the one hand, patterns exist, patterns can be noticed - but on the other hand we're dealing with the most complex and chaotic system imaginable. It does feel like the question needs to be "should we" and not "can we," because "can we" can always be met with "we can try" and then they ignore all the examples the authors gave for reasons that we actually can't

AJ Biffl

Love the shout about theory-ladenness of observation! The thing about the physical costs brings up an interesting point: the larger the computer system, the larger the dataset, the larger its entropy - the larger the energy, time, effort, coordination, organization, administration, required to make it function. The odds of errors and bugs and faults goes up, the weight of bureaucracy and admin goes up. And for each piece of apparatus you add, you also need to add apparatus to coordinate and organize that apparatus, and apparatus to coordinate THAT apparatus, and so on

AJ Biffl

Have you seen Hank's video about 'media revolutions and our terrible moment'? He sort of gets into the fact that this is an entirely new era of mass communication, and we don't have the psychic guardrails as a society to resist all the BS that floats around on social media. There's a pretty consistent connection between new media forms and populist political movements - with the old gatekeepers gone, the old psychic and legal regulations gone, there are no old taboos and no checks on truth

Anthony Perez Eisenbarth

MacIntyre was a phenomenal Marxist turned radial Aristotelian philosopher. He had very unique perspectives. And I would say it is somewhat related to the book's points, but it's carried further. MacIntyre identifies four distinct sources of irreducible unpredictability in human social life, each of which operates independently of the others and together make the project of social scientific prediction in the law-like sense almost impossible rather than merely difficult. It’s a perspective that he shares with other philosophers like Nancy Cartwright, John Dupré , and Ellery Ells. Cartwright calls the world "dappled" a patchwork of context-specific regularities, not a single clockwork universe. If even physics only gives us laws that hold inside engineered experiments, why would social life, with its reflexive agents and shifting institutions, yield anything stronger? Dupré adds that reality is a process, not a collection of stable things; disorder and flux are the rule, not the bug. And Eells formalizes the consequence: causes raise probabilities, they don't necessitate outcomes. So we can explain why something happened, but we cannot predict it with law-like certainty. But back to MacIntyre’s four points. The first is the impossibility of predicting radical conceptual innovation. New ideas, new frameworks, new ways of understanding situations genuinely alter what people do, and they cannot be predicted in advance because if they could be predicted they would already be known and would therefore not be innovations. The second is what MacIntyre calls the game-theoretic character of social interaction, which draws obviously on game theory, but MacIntyre’s use of it is in more in line with the way John Roemer or Samuel Bowles and Herbet Gintis make use of it rather than say finance or econ bros. In most consequential social situations, each participant is simultaneously trying to render others' behavior predictable while keeping their own behavior unpredictable to those others. This coordination failure is a structural feature of strategic interaction that better information cannot resolve. MacIntyre illustrates this with a wry example: a commercial board-game version of the Battle of Gettysburg. A moderately good player taking the Confederate side could win the battle yet the actual General Lee, far more intelligent (presumably than the average player), lost. Why? Because the player knows from the outset what Lee could not: the precise time scale, the exact units that will appear, the fixed terrain boundaries. In real life, no one knows they are fighting the “Battle of Gettysburg” until after it’s over. The shape of the situation is conferred retrospectively by outcomes, not prospectively by a rulebook. That’s why computer simulations that import past determinate situations into future indeterminate ones inevitably fail: they mistake the retrospective clarity of history for prospective predictability. The third is the sheer contingency of individual lives, the accidents, encounters, and chance events that alter trajectories in ways no model could have captured in advance. The fourth is what MacIntyre calls the unpredictability of each agent's own unmade decisions: people genuinely deliberate, and the outcome of genuine deliberation cannot be specified in advance without eliminating the deliberation. Social media metrics is a good example of the second and fourth sources operating together. The moment a metric becomes known and consequential, behavior reorganizes around it, which is what Goodhart's Law or the Lucas Critique in economics describes. It is a structural feature of social systems in which agents observe and respond to the rules being used to evaluate them. What I mean by data generating process comes from Bayesian statistics applied to social processes. Academia really did a number on the way I explain things so bear with me. In Bayesian statistics and machine learning, a “data generating process” is the assumed underlying mechanism that produces observed data. It’s typically formalized as a probability distribution. The key assumption is stationarity or at least exchangeability: future data are generated by the same process as past data, possibly with known types of change (e.g., drift, regime switching). When we train a machine learning model on historical data, we are implicitly assuming that the patterns in the data reflect the same stable underlying regularities that will persist into the future. But social data is not generated by a stable underlying process in the way that, say, physical measurements are. It is generated by specific institutional arrangements, specific power relations, specific historical conditions that produced specific behaviors from specific populations at specific moments. When Optum trained its health risk model on healthcare expenditure data, it was training on data generated by a racist healthcare system in which Black patients received systematically less care than White patients with equivalent health needs. And when the institutional arrangements change, as they do continuously, the patterns the model learned cease to hold even in the domains where they previously described something real. The data-generating process is a product of history. Although if MacIntyre is right, then the very idea of a "data-generating process" in social science is merely metaphorical or mistaken if by “data-generating process” we mean a stationary or law-like mechanism of the kind assumed in Bayesian statistics and machine learning. Since by that line of argument there is no process that generates social data in the way atmospheric pressure generates temperature readings.

Anthony Perez Eisenbarth

Yeah, MacIntyre doesn't think all social science predictions are useless just that there are no universal laws covering social life that are out there to be discovered.

AlexTheTutor

“The limits of my language are my limits of my world”, and so if I choose a large language model, my limits would be quite large, but in what sense? If it were infinite, I would know how it all ends, and the knowledge of knowing how it all ends would change the ending, which later would change the ending once again once the secret gets out. Thus, saying nothing of water and silicon, there are limits. The world is finite. The model itself limits what can and cannot be said. Suppose I choose a model full of errors. Should I then claim to have infinite incorrect knowledge? Copy, Paste, Repeet. The model gives “better” results based on how much time it has to practice and how much data it is given. As if speaking of limits in calculus, given infinite time and data, then is it possible to translate “better” into “best” or even “perfect”? But, we have already dismissed the idea of infinity as childish. Now, what threshold are we to say that the AI model provides “good enough” information about the future when it needs certainty in its data about an uncertain future? If 2 pendulums attached to one another creates a chaotic situation, what of life? How are we to predict life: a thing with countless pendulums connected to one another, governed by uncertain laws of human nature. Thus, AI is the kind of religion in which we must believe that infinite power of knowledge is not only possible but... inevitable. AI is nothing but a giant math equation multiplying with other math equations to create a final result. Is our faith in infinity based on ones and zeros? Perhaps in the future, scientists will deem the scientific method obsolete. And, what of religion? If we can push out science, then we can also highjack faith itself to convince enough people to invest into a clear and singular future. Thus, the prophet said unto its people: Invest all your money in the future of technology, for surely all bubbles in our infinite universe will grow to infinity. This time is different. Hear nor the pop, fear not the crash. Lend us your money, my child. AI so conceived is nothing more than a prophet’s desire to reach infinity. And, much as using GPS to always figure out where to go as one part of the brain shrinks due to lack of usage, so too can AI, if used all the time to outsource all of thinking, will shrink the entire brain until the perpetual child can do nothing but vote for his own imprisonment.

phil e the theyby

I had forgotten about that video, but I did see it! I’ll have to watch it again, because I remember it doing a good job of putting this moment in historical context. Like, this is something that’s happened before (if in a different form), and it’s our responsibility to try to build those guardrails and steward humanity into living with these new forms of communication rather than being subsumed by them.

phil e the theyby

I would argue that what Karl Marx was doing was essentially trying to predict the future, just without the assumption that the existing system is completely static. I do think trying to predict the future is pretty core to the human experience (and that this is a good and adaptive thing); we’ve just transitioned from using the movement of animal herds or the stars to using data and mathematical models. In a lot of indigenous cultures, time is seen as cyclical, with certain cycles or patterns repeating throughout history. In a certain sense, that is predicting the future, just with less specificity than modern systems. For me, the distinction between a more authoritarian view of the world and a more liberatory one isn’t whether you’re trying to predict the future, or even if you’re trying to have some sort of agency in it. It’s whether you’re trying to use that knowledge to dominate other beings and shape them to your worldview, or if you’re using it to honor other beings for what they are. A colonial framework says that humans (typically, White men specifically) have dominion over other beings, and shape the world into their image of what it should be. A liberatory framework still makes predictions and aims to influence the world based on those predictions, but with the aim of honoring the autonomy of other beings, and helping other beings to be who they are and thrive in that, even when it doesn’t adhere to what we might want.

Black EnviroNerd

Appreciate you all for humoring my musings! More thoughts after reading the replies and a night's sleep! I do agree with Laura and Phil and think you guys are correct in that just simply trying to predict the future isn't inherently conservative. However I think AJ's reply was more getting at my intention and thought process. AJ I love how you described it., "prediction fundamentally requires the view that people's behavior in the future will largely be the same as it was in the past - no allowance for large scale growth, development, evolution, change. It is fully seated in the assumption that the world at its largest scales, and the people in it, will remain constant, the past conserved. " Narrowing down the idea a bit, attempting to predict individuals behavior does seem to be more of a conservative practice to me due to the assumptions that AJ highlighted that go into it. The idea of predicting behavior feels a bit cynical to me and I imagine appeals more to those predisposed to have more negative/conservative beliefs about human nature (I.e the conservative belief that humans need law and order to be moral actors). Additionally even though the predictions can be used for things I would generally agree with and find progressive they require a system that will use them in for lack of better words the "right" way. For example I'm thinking of the example at the beginning of chapter 2 with the predictions of students academic success. As much as the practice may have opened the students up to beneficial interventions (to the extent the model even correctly identified them as needing them), it also opened them up to more misguided interventions (anecdotally all lot my black friends were advised to switch their major in college whenever they had a setback) and outright exclusion (in the case of Dean looking to force them out). Whether the system is a liberal one or a fascist one I think it does warrant pausing and questioning if these predictions will ultimately serve us or these oppressive hierarchies. Considering how one of the main modes in which these hierarchies thrive is divide and conquer, more often than not I feel these predictive behavioral AI models have more utility for conservative interests.

phil e the theyby

@Black EnviroNerd I think you're DEFINITELY onto something here. There's something about trying to predict what an individual will do that kind of assumes that you know something more about them than they know about themselves, typically because you feel you have some sort of knowledge about a "higher order" (whether that be "the way the world works," "human nature," "God's plan," or anything else). Then, based on that knowledge, you're asserting that you know the correct steps or actions to take, which is either what the individual should do, or what should be done about them on a societal level. I'm not sure I'd even describe it as "conservative" (since I'm trying to move away from using the label of "conservative" to mean "everything I don't agree with"), because it can be used in ways that are extremely disruptive to existing social systems and are trying to fundamentally structure society in a new way (I think of "conservative" as often involving or invoking hierarchy, but that its priorities are more on sustaining the existing or a previous social order than necessarily enforcing hierarchies). I think I'd probably describe it as "authoritarian"; it's a way of imposing external control over someone and assuming that you know what's best for them, often without their knowledge or their consent. I think this description applies even to examples I wouldn't necessarily oppose on a moral level (for example, I don't think I'm inherently against trying to predict who is likely to smoke and why, and based on that trying to create supports that make them less likely to smoke). Even if I see the potential social outcome of that (fewer people as smoking), that's still me imposing my view of the world onto other people, who may not agree with it. Thanks for your really thoughtful and interesting elaboration; it's definitely an idea I want to keep reflecting on and engaging with.

The Baiterek Life System. Mind Gym.

Hello, I believe I understand the causes of the problems discussed in this publication. The main problem is that our life is created! Planet Earth and the people on it were created by Higher Living Cosmic Beings. They created humans to raise their own fully functioning offspring. On Earth, our Creator-Parents, using special software devices, create numerous problems solely so that their children (tachyons) can develop and improve. I discuss this in detail in my books, "Notes on the Universe, Baiterek."