Fundamental Insights: But so late in coming
A few "sparks" that have inspired my thinking lately, but seem so obvious upon reflection
As I start to settle into San Diego (we moved into a house! that we bought(ish)!), I’m finding my thinking a bit more expansive and relaxed. Perhaps its the math that is unmooring me from previous presumptions or opening me up to the possibility that unlikely people and places can hold profound insights, but this is what I’ve been thinking about over the past couple of weeks.
St. Augustine vs. Aristotle: The purpose of government
I know. Old school debate. Literally what I went to college to study. But, in listening to my Great Courses’ History of the Catholic Church while washing dishes, I heard this important distinction: St. Augustine’s view of the world was that we were born into sin, and everything that came after worked toward minimizing this sin. Thus, the purpose of government is to prevent further sin - that the main focus of civil institutions is to support religious goals - and to legislate and to govern well is to protect people from sinning more. This is juxtaposed to Aristotle’s understanding of government, which argues that community is the center - the “political animal” line is a bit misquoted, but the basic idea is that people work together to build community, by the mores, laws, culture and so forth that stem from our common humanity. The purpose of government, then, is to facilitate community belonging.
What this insight means to me is that the GOP extremists view of squashing liberty isn’t actually, to its adherents, about squashing liberty, or even legislating a particular view of morality (even though the effect is to do just that). Rather, it is to use the power of government to protect us from sinning more. This means that it is logically consistent to be pro-business, pro-freedom, and so forth, as long as individual sins and temptations for sin can be controlled. Literally, government’s purpose is to help us get out of original sin, and that’s it. When viewed from this perspective, I find it much easier to see the actions of Christian nationalists as part of a coherent ideology. Why is coherence important for such abhorrent ideals? Because it is not effective to dismiss their ideas and their political strategy just as “incoherent” and “contradictory” - to intervene in ways that may be actionable requires seeing where the heuristics for this viewpoint emerge. In other words, understanding logic, however faulty, strengthens the power to challenge it.
Intersectional thought makes no space for assumptions of power that are themselves rooted in racist/discriminatory/stereotypical foundations.
Via Franklin Foer’s essay in the Atlantic, “The Golden Age of American Jews is Ending,” there is this paragraph that just sums up exactly why anti-semitism has generally been disassociated from conversations about IDEA, DEI, anti-discrimination and so forth. Basically, when we assume based on people’s identity that they are the oppressor, sometimes those assumptions are in fact deeply problematic and themselves rooted in discriminatory thoughts - The problem is that the stereotypes and the anti-semitic tropes that tend to be associated with Jews falsely calm they have more power than they actually do (Jews run the banks, Hollywood, international politics, cause global wars, etc.). Via Foer:
At its core, the intersectional left wanted to smash power structures. In the American context, it would be hard to place Jews among the ranks of the oppressed; in the Israeli context, they can be cast as the oppressor. Nazi Germany definitively excluded Jews from a category we now call “whiteness.” Today, Jews are treated in sectors of the left as the epitome of whiteness. But any analysis that focuses so relentlessly on the role of privilege, as the left’s does, will be dangerously blind to anti-Semitism, because anti-Semitism itself entails an accusation of privilege. It’s a theory that regards the Jew as an all-powerful figure in society, a position acquired by underhanded means. In the annals of Jewish history, accusations of privilege are the basis for hate, the kindling for pogroms. But universities too often ignored this lesson from the past. Instead, they acted, as the British comedian David Baddiel put it in the title of his prescient book about progressive anti-Semitism, as if “Jews don’t count.” - Franklin Foer
The idea of Jews as oppressors is actually a very anti-semitic, deeply problematic way of thinking about Jewish people that also elides differences between country of origin, religion, race, class, and so forth. So, when an intersectional lens is applied to Jews, the idea they have power and thus are oppressors is itself a deeply problematic assumption. There is no way around this intellectually that I see in the main ideological construction driving the contemporary progressive left. (This is also (maybe) why it is so hard to account for the folks who grew up in trailer parks but happen to be white, too, and their privilege is pretty complicated. )
Tech people do not see siphoning insight without credit from across the web and across human knowledge as problematic. I was fretting about whether stealing code and querying ChatGPT was cheating on myself (more than anything else) and maybe cheating myself out of knowledge I ought to have. But there is a fundamental cultural difference that I think people working in the world of words do not understand about technical cultures that I think was put so well, from an old DC connection, Luke Peterson, and a supremely important insight in, yes, a LinkedIn comment:
If I have learned anything from analyzing the digital media consumption behavior of audiences around the world for the last 10 years, it is that coders have always found help from other places. Is copy-pasting ChatGPT-written code worse than copy-pasting code found on StackOverflow, Reddit, or Quora? The pros swiftly migrated from those places to OpenAI last year. Should we tell your students that creative prompting is the new creative googling? Do they know enough to not run “DROP TABLE infile” if ChatGPT gets frisky? - Luke Peterson
In other words, it is logically consistent, culturally appropriate, pragmatic, and totally legit to just soak up the knowledge of the internet and repurpose it for yourself (and theoretically, the tools you will build for humanity). This is why OpenAI, ChatGPT, Bing, and their engineers do not hesitate for a second to think about how the foundation of knowledge produced by others is the source of future knowledge for all - and so ideas of copyright, idea ownership, monopolies on content from which new content is created - this all is fair game when we extend the logic of how coders code to how Open AI and other generative AI LLMs are building their databases. Even the word “generative” suggests that there is something that came before to inspire something to come next - and when this is the underlying logic of how this AI will be trained, it totally makes sense that authorship and idea ownership is a secondary consideration, if a consideration at all.
Big data prediction cares more about matching than causality
In my new “Spring” semester class at UCSD, Causal Inference, we discussed something quite important - that when you imagine an if→ then statement, we are assuming that causality actually is happening, that it cannot be ruled out by some other alternative explanation, and that there is a temporal order. However, the way that algorithms work when it comes to surveillance capitalism has a lot less to do with causal order and even temporal order. What matters most is likelihood of association (literally maximum likelihood optimization via what ultimately comes to be understood as some form of learning that minimizes the gaps between predictions and various data points for one particular instance of data.
In other words, based on a bunch of inputs from you, what is happening in the background is thousands of different algorithmic estimations of the best most likely thing -the single best prediction of what the next bit of data might be. What this means is that the order doesn’t matter. What you click on, what you do, in an instance, it doesn’t matter if X causes Y, or one action happened before another - it’s really about congruence not causality.
I am starting to understand this a little bit, but the most basic form is a support vector machine - the basic idea is that if you have some slice of the data (points in space) — to use this training data to then predict the best way to maximize the largest minimal distance between points — this is called - no joke - a “hyperplane” with the best hyperplane showing the largest minimum margin (I know so weird and tongue twister) - between these points as that will give you the best prediction of the next set of points/data in space. The figures below are better and stolen from Fernando Iglesias Garcia’s Open Source Computer Vision Post
If you think about this from how logic might inform practice, the OUTCOME doesn’t matter as long as the outcome is predicted correctly. There is a valueless universe of the best ways to match our preexisting behavior with some best prediction of our future interest/behavior/attention - and literally, at the core of technical production, the formulas could not care less about whether/how you came to the starting point and whether/why/how you might arrive at the next destination, only that it is best prediction of what might happen next. There are no values, only scores of prediction that work and don’t work.
OMG it’s a formula - but if M is the margin you are trying to get, you are trying to get the likelihood of various Betas (coefficients, points in space) subject to the data before, for all of the existing data that came before. It’s literally taking a likelihood equation and getting the min vs. the max. but, yes, maximizing the minimums. So weird.
Graph below, from Garcia:
This strikes me as important as it explains why sometimes you see ads after you bought something, but more important, that essentially, any of these companies - Facebook, DoubleClick, Amazon, Google - they don’t care at all “Why” you might be doing something or even what might be motivating the behavior. All that matters is getting you to the most likely next thing to steal your attention away from whatever else you might do, online or offline.
Really, the takeaway is that all that matters is whether the matching is best and maximized, not whether the causes that got you to where you happen to be make some sort of logical or coherent sense, or temporal order, or even reflect some sort of systematic reason. There are no values, only optimal predictive approaches.
These insights, I think, take me a little bit further into thinking about how to comprehend the magnitude of what it just might take to think seriously about how to recover the project of American liberalism, pluralism, and tolerance. It will just take me a bit to get from those points above to …. joke … the maximum minimal margin of competence and knowledge to write the darn proposal for a new book.