9/22/2023 the pastry baker and the doctor
I’ve heard linguists say that for all its success, the lack of mechanistic interpretability in LLMs means they don’t really help us understand how language works, and to that extent it’s more of an engineering innovation than a scientific one.
But within a pragmatist conception of knowledge, where truth is grounded in applicability and direct contact with experience, how do we make sense of this? Would AI not be a model of good science? Here is maybe one way of making sense of this: AI technology can produce impressive results in the contexts that it was designed for, i.e. simulating coherent human behavior, but unlike a typical scientific model, it doesn’t provide concepts and methods that you can apply in other contexts, and connect to other domains of knowledge. The “knowledge” it provides is very brittle.
To use an old example from Socrates, it’s like comparing the pastry baker and the doctor: a good pastry baker knows how to produce things which bring people pleasure in the short-term, but they can’t relate that to things like the chemistry of the food and the functioning of the body, and they can’t predict the impact their goods will have on the health of a person in the long term. This is what doctors can do: bring knowledge from anatomy, medicine, chemistry, etcetera to bear on our everyday activities and diet, interrogating and explaining the behavior of the body.
9/22/2023 truth by comparison
An important distinction to make is, there’s truth by operational coherence, which works by direct confrontation with experience, and what I call truth by comparison: sometimes you say “yes, the truth of this statement consists in its agreement with other things we regard as true. So, yeah, global warming is true. Why? Because we already believe in the cogency of our temperature measurement methods, and here are the data produced by those methods.” We’re not directly checking global warming in the sense of truth by operational coherence, usually, but we’re checking it by comparison.
— Hasok Chang, in this lecture about Realism for Realistic People
9/22/2023
But now, what is truth? It’s easy to talk about truth in terms of language, of course: statements are true when they accurately describe reality. But what about reality, truth itself? The best we seem to be capable of saying is this: reality is “the way the world is” (monistic), or “the way things are” (pluralistic). Reality is “what’s true” and truth consists of “what’s real”. But what is it for crying out loud? This is the noumenal question at the bottom of it all, and I have the inkling of an intuition that the reason I have trouble answering it is that the question doesn’t make sense. It’s somehow malformed.
9/22/2023
Here is roughly how I make sense of the correspondence theory of truth: facts are built out of basic grammatical components, each referring to features of the world. You combine them like lego bricks to create complex representations of reality. It’s the idea of a kind of isometry between representation and reality, like making a map of a territory: different symbols represent different things and their spatial relation to one another, and we assess the map’s accuracy and fidelity by comparison with the real thing, using the key and the scale.
I’ve been feeling like at least part of the correspondence theory is built into the very idea of a “representation”: a representation is a representation of something. It is not the thing itself, but it refers to it in a very special way. It says “I resemble the thing that I’m referring to, if you interpret my features in a certain way”. A challenge for anyone wanting to do away with the correspondence theory, I think, is to reconcile with the inherent notions of correspondence in the idea of representation, and in our everyday language.
As I write this, what is at least clear to me is that the correspondence theory is a theory about the relationship between truth and language. Specifically, what constitutes a fact and determines statements as factual. So it likely involves either a silent assumption about what truth is, or a definite theory about it.
9/21/2023
Dominant user interface conventions are functionally useful for website visitors, but also a source of homogeny across the web — maybe the right mindset for a designer (in a commercial context) is that breaking them is okay, even if they cause some initial confusion for a visitor, as long as you provide cues to guide people on how to navigate a page.
9/19/2023 tools both constrain and expand the field of possibility
Some thoughts re: “the designer and the engineer must be able to work by drawing with a stick in the hand”:
There are ideas here I really agree with: we should strive to be independent of proprietary software, and at a basic level, I don’t think you should identify a designer’s ability with creative cloud proficiency. But I don’t think we ought to trivialize the significance of tools either.
As I’ve come to understand it, design is a field constituted by shared conventions, both internal to the designer community, in that we share common tooling and methodologies, and in public, in that we all share visual and cultural reference points. When technology and culture shift, we may pick up and discard various tools and skillsets — hot metal type exchanged for phototypesetting, etc. But this does not mean that our roles can be conceived entirely independently of our tools. Rather, the meaning of the role changes along with changes in tooling.
My perhaps divisive view is that the “embellishment, commercialization, and industrialization” of design is really all there is to design. There is no platonic design mindset or skillset outside of the contingent technologies and cultural practices which ground our work.
Some of the skills that I picked up from my own design education — e.g. the ability to articulate and justify my ideas, the ability to visually order information, the ability to research, etc — are indeed not dependent on any tool, but my role as a designer still hinges on tool proficiency and familiarity with the designed surfaces of this day and age. Typography, and the availability of typefaces, for example, are a technical skill and a tool, but also something integral to my sense of self as a graphic designer.
I can say the same of “engineering” from my background as a web developer — what I’m capable of, what I can even conceive, depends on the constraints of the html/css/js.
In design as well as engineering, I think tools both constrain and expand the field of possibility. None of what we do makes sense without including them in the picture.
9/10/2023
I think there is something to the idea that, in order to persist in a long personal project with no outside pressure, you have to be as in love with your process as with the desired output.
9/9/2023
Finally got around to adding a guestbook to my website! It refreshes daily from none other than this are.na channel, using the magic of eleventy fetch, the are.na api, and netlify scheduled deploys.
9/3/2023
In the past I’ve debated with people about the efficacy of bullets (as designers are oft prone to do in their free time), and as a “pro-bullets” person, something I’ve just realized is that my affection for them comes from their value not as reading aides, but as writing aides. The conventions around bulleted text are such that in using them, there’s less pressure to write complete and well-constructed sentences. That makes them ideal for getting your thoughts out on the page. One might even say, “I think in bullets”.
8/24/2023 update to goby.garden
8/20/2023
Changing the facts can change which facts are intelligible. For example:
If Peggy is a person (fact), then it makes sense to ask “how many toes does Peggy have?” or “what is Peggy’s favorite color?”. But if Peggy is a plant (alternate fact), then these questions no longer make much sense, ordinarily speaking.
8/11/2023
“I think we’ve ended up in a classic situation where people who are fans of tech end up pointing to a new tech thing and saying ‘this is gonna solve all our problems’ when it’s already a solved problem, when the solution is to design streets around people: fund transit instead of designing them around cars.”
— Mingwei Samuel of Safe Street Rebel, on Hard Fork discussing self-driving cars in SF
8/10/2023 from <3
hey nico, just stopping by to say i'm honored to be in your "hyperspecific collection channel" <3 i've been loving this channel for a long time. it's like a happy gold mine.
8/9/2023
“I get asked a lot about learning to code. Sure, if you can. It's fun. But the real action, the crux of things, is there in the database. Grab a tiny, free database like SQLite. Import a few million rows of data. Make them searchable. It's one of the most soothing activities known to humankind, taking big piles of messy data and massaging them into the rigid structure required of a relational database. It's true power.”
— Paul Ford, in this column
8/9/2023
“I've always loved that moment when someone shows you the thing they built for tracking books they've read or for their jewelry business. Amateur software is magical because you can see the seams and how people wrestled the computer. Like outsider art. So much of the tech industry today is about making things look professional, maybe convincing Apple to let you into the App Store to join the great undifferentiated mass of other apps. That's software. When people build their own Airtable to feed the neighborhood, that's culture.”
— Paul Ford, in this column
8/9/2023
“Code isn't enough on its own. We throw code away when it runs out its clock; we migrate data to new databases, so as not to lose one precious bit. ==Code is a story we tell about data==.”
“But programmer culture tends to devalue data. The database is boring, old, staid technology. Managing it is an acronym job (DBA, for database administrator). You set up your tables and columns, and add rows of data. Programming is where the action is. Sure, ==80 percent of your code in Swift, Java, C#, or JavaScript is about pulling data out of a database and putting data back in==. But that other 20 percent is where the action is, where you make the next big world-shaking thing.”
— Paul Ford, in this column
7/26/2023
In re-designing the goby database architecture to address earlier limitations, I’m learning new things about what my original idea entails logically (things which the old database architecture failed to capture adequately).
The idea that I should be “discovering” new things in trying to map a logical schema is really amusing to me — it’s logic, so in a sense it’s all there from the start, waiting to be deduced from fully available information. But of course there’s only so much I can hold in my head at a time, so things never occur to me or I assume they’ll work out at first. As a result, I have moments of ‘discovery’ when I realize implications of what I’m trying to do.
7/25/2023
Commercially viable websites often seem to have the weight of the world on them: they’re a big investment and the public face of an organization, and to that end they have to be extremely robust, functional, refined, and cohesive.
Accordingly I’m finding it’s a joy to be able to work on small, one-off editorial web projects, where the room for play and experiment is much broader, although the design decisions still have to be well-considered.
7/22/2023 from
re: spinny cloud on mix-blend-mode: multiply
yes! I was thinking of setting it to multiply
but the way the white background changes was compelling hh
7/16/2023
The ICA has so far resisted the drive towards lowest-common-denominator programming that countless institutions adopt to attract the highest audience numbers. And likewise, we have aimed to avoid the bland, diluting effects of design-by-committee industry defaults that precipitate the easy slide into monoculture, instead championing the idiosyncratic and particular. In the ongoing drift from physical to digital means of channelling information, holding specific information in suspension with a generic template is not getting any easier, but it is a necessary job lest we are shortly made redundant ourselves...
— Stuart Bailey discussing the design of the ICA identity and the role of templates in the digital age in Revue Faire no45
7/16/2023
An emphasis on the non hierarchical, parallel programming of the ICA's five main strands of activity, again to emphasise the multiple arts while acknowledging the fact that they run at very different speeds. In a typical season, there are anything up to eight different films playing in the cinemas every day, maybe one talk or learning event a week, one live event a month, and a daily exhibition that runs for around three months. A useful visual metaphor here was a motorway, with vehicles moving at different velocities along its various lanes, enacted most prominently on the website.
— Stuart Bailey discussing the design of the ICA identity in Revue Faire no45
7/16/2023
When we say the only way to get better at something is practice, I don’t think “practice” can be conceived as merely doing it the same way over and over — practice makes permanent, not perfect, as is said.
What we want to capture when we say that, I think, is that by doing something enough, doing it poorly many times over, you (hopefully) start to scrutinize and validate your own methods, course-correcting over time.
7/11/2023
There’s an idea from Dan Michaelson that has always stuck with me, that a website can be a really interesting place to begin an identity system, because it asks you to think about how (typo)graphic elements apply to different content types and formats. But in recent work I’ve been realizing that the constraints of web aren’t always generative — it feels like they sometimes confine my thinking to treatments that are easy to templatize and build in code. I’ve been conversely deriving joy from starting first with static compositions, and then challenging myself creatively and technically to adapt them to dynamic surfaces like the web.
7/10/2023
“It's as if now we're at a moment where there isn't even the semblance of a coherent project of shaping consciousness. Instead, we're in an increasingly chaotic field. There is this acceleration, or this abruptness, and this kind of incoherent flood of completely incompatible, constant information.
So rather than building a coherent set of assumptions and ideas, we are inhabiting a world in which exchange and conversation and communication is... I don't want to say prohibited... but disrupted, in ways I can't think of as having any historical precedents. The diminution of any sort of historical memory on the part of an average person has being going on for some time, since the 20th century. Now there's a combination of damaged memory with the inability to absorb and assimilate how the present is configured.”
— Jonathan Crary, in an interview about his new book Scorched Earth in Real Review issue #14
7/10/2023
“I think if everything we were doing was extraordinarily complex we would go crazy, and I think if everything we were doing was extraordinarily simple we would also go crazy, so I think the key is always having a range of complexity and also a range in the scale of projects. Some projects we do are two years long…and it’s great to have that going on but it’s also nice to have the kind of thing where you can do the whole thing in 3 months, start to finish.”
— Adam Michaels of IN-FO.CO, on Scratching the Surface
6/25/2023
I think philosophy papers, produced as they are under the pressure of answering all possible objections and staying within a well-defined scope, sometimes trade clarity and persuasion for accountability and exhaustiveness.
I don’t know whether that’s a good or a bad thing, but it at least informs my decision as a non-academic about what kinds of philosophical prose I ought to invest the most time in reading.
I wonder if one solution to this would be the approach taken by Edwin Curley in Behind the Geometrical Method, which, as the description states, is “actually two books in one”: first, a clear and explanatory account of Spinoza’s arguments geared towards first-time readers of the Ethics, and secondly a text comprised entirely of the footnotes to the first section, which defend its interpretation of Spinoza in richer detail for other academics.
6/25/2023 “out-thereness”
When I think of “objectivity” or the question of what makes something an “object,” I often intuitively lean on the idea of “out-thereness”: that is, the idea that objects are those things which are out there, being as they are independently of us grasping them so.
This is, I’ve found, an incredibly fraught notion. But maybe one worth holding onto and expounding nonetheless.
6/25/2023
Through the cyclical retracing of lines of inquiry that this platform enables, one observation I’m making is that my interests over time have gravitated towards the interplay or tension between what is ‘objective’ and what is ‘subjective’ (or how we should even conceive of those properties).
Thinking particularly of the trains of thought represented by
and
6/25/2023 Wittgenstein’s style
“The constant characteristics of Wittgenstein’s writing are close reasoning and strong imagination. But the book has also the character of great variety of tone: this is a rare character, and particularly rare in philosophical writing. (The only other examples I can think of are some of Plato’s dialogues). You get long passages of very sober, straightforward enquiry and argument; then a burst of breathless dialogue (always, of course, between himself and himself); sudden turn of humour, passages full of passionate feeling; pronouncements reached after perplexed enquiry, which have the air of being written with that feeling: And that settles everything; pieces of delicate, accurate characterisation of some particular temptation; remarks that are like a grasp or cry of realisation. And you get certain themes, certain moods recurring and recurring with different variations. I have long been tempted to compare this book with a musical composition; but hesitated to do so, until I found it elicited this reaction independently from someone who read it de novo.”
— Elizabeth Anscombe, in a BBC Radio Talk discussing her English translation of Wittgenstein’s posthumously published Philosophical Investigations
6/25/2023
“The final product of all this work has very remarkable literary qualities; once it is published, I think there will be no more wonderment why Wittgenstein –who spoke English well –wrote in German. It was horribly difficult to translate. I doubt whether much of a reflection of its style would be possible in English at all; at any rate it was not possible for me to achieve it. In general, German has possibilities of a homeliness –the very epithet sounds horrid in connexion with English –that is not in the [slightest] in conflict with the highest literary style. For an example, you only need to look at Gretchen’s lines in Faust when she comes in after Faust and the Devil have been in her room. Wittgenstein’s German is at once highly literary and highly colloquial. Good English, in modern times, goes in good clothes; to introduce colloquialism, or slang, is deliberately to adopt a low style. Any English style that I can imagine would be a misrepresentation of this German. All I could do, therefore, was to produce as careful a crib as possible. I bent over backwards to write in a spare and compressed English, since the German is spare and compressed; and in part I the translation turned out several lines a page shorter than the original. (This was right, because English is a shorter language than German”
— Elizabeth Anscombe, in a BBC Radio Talk discussing her English translation of Wittgenstein’s posthumously published Philosophical Investigations
6/25/2023
“‘I spend more time than you perhaps could ever understand, thinking about questions of style,’ Wittgenstein once said to someone. The state of the MSS and TSS that he left behind him are a witness to this. He wrote an enormous amount; he threw away a good deal, and what is left is a formidable quantity of MSS written from 1929 onwards. He would often write first in small notebooks, then transfer siftings from these to larger ones, then further-polished-siftings to still larger ones. Then he would dictate to a typist. Then [he] would cut up the typescript and throw a lot away, and try different arrangements of the rest: for he always wrote in the form of isolated paragraphs capable of rearrangement. A great part of the material of the Philosophical Investigations exists in two other quite different arrangements, each brought to a final form and ready for the printer, and each elaborately cross referenced; for he hoped at one time to supply his ‘Remarks’ as he called them, with cross references to every other one with which he saw a fruitful connexion. His MSS. sometimes contain remarks written over and over again in various forms. They always contain a huge number of variant readings, with variant punctuations; and if you read carefully through each possibility, you notice how sharply aware Wittgenstein was of small rhetorical differences.”
— Elizabeth Anscombe, in a BBC Radio Talk shorty after the posthumous publishing of Wittgenstein’s Philosophical Investigations
6/25/2023
“A certain amount of current philosophical discussion concerns itself with linguistic usage. This is a direct result of Wittgenstein’s teaching that in a great many cases, in which we speak of ‘meaning’, though not in all, it can be defined thus: the meaning of a word is its use in the language. But ==use is not usage; he did not wish to base anything on idiom==. ‘I distinguish,’ he wrote ‘between the essential and the inessential features of an expression. The essential features are the ones that would make us translate some otherwise unfamiliar form of expression into this, our customary form.’ What is or is not correct English usage is of no conceivable philosophic interest; nor does it matter if I choose to use words in an extraordinary manner, so long as it [is] clear what I am saying.”
— Elizabeth Anscombe, in a BBC Radio Talk shorty after the posthumous publishing of Wittgenstein’s Philosophical Investigations
6/25/2023 The Tractatus haunts the Investigations
“The Tractatus is of the greatest possible importance for understanding the ‘Philosophical Investigations’. W. came to realise this and wanted the two books bound up together, which will, I hope, be done in a purely German edition. It is important because Wittgenstein clearly remained in love with the thoughts of the Tractatus, though he attacks and most powerfully undermines them. The Tractatus haunts the Investigations.”
— Elizabeth Anscombe, in a BBC Radio Talk shorty after the posthumous publishing of Wittgenstein’s Philosophical Investigations, which she translated into English
6/14/2023
Are moral questions matters of fact, matters of convention, or a secret third thing?
6/14/2023
Recognizing that journalists inevitably carried personal biases and blind spots, Lippmann called for controlling them by professionalizing journalistic processes and, in particular, embracing lessons from the scientific method. He entreated journalists to focus as much as possible on facts and to actively pursue evidence that could challenge, rather than simply confirm, their own hypotheses. In this conception, words like objective and impartial are not a characterization of an individual journalist’s underlying temperament, as they are so often misunderstood to mean, but serve as guiding ideals to strive for in their work. “The idea was that journalists needed to employ objective, observable, repeatable methods of verification in their reporting—precisely because they could never be personally objective,” Tom Rosenstiel, coauthor of The Elements of Journalism and one of the leading defenders of the model, explained in 2020. “Their methods of reporting had to be objective because they never could be.”
— A.G. Sulzberger, in his piece Journalism's Essential Value
6/14/2023
I think we need to lose the thought, first of all, that the universe is a chicken.
Hasok Chang on the correspondence theory of truth and Plato's “slice nature at its joints” metaphor, from this talk
6/14/2023 a practical example of operational coherence
Every time I think about [GPS], it’s just incredible, how does this thing work? But it does. What they have to do, if you don’t know, is coordinate a whole set of geosynchronous satellites which they fly using good old-fashioned Newtonian mechanics. And then they put atomic clocks on these satellites that track the exact differences of time that different signals from different satellites take to go to a place and come back. And then these atomic clock readings have to be corrected using both general and special relativity — not together, separately — because there’s time dilation due to speed and there’s the change of clock rate depending on the place in the gravitational field. And then the signal gets beamed down to you and me, and we just walk around with our phones pretending that the world is flat.
All of these material, conceptual things have to be brought together in an incredibly harmonious way in order to make this whole activity of satellite navigation work. So roughly speaking what we’re talking about [with operational coherence] a harmonious fitting-together of elements and aspects of an activity that is conducive to the successful achievement of the aims of the activity in question.
— Philosopher of science Hasok Chang, in defining his term 'operational coherence' in this lecture
5/16/2023
It’s not that I think LLMs are conscious — only, I haven’t seen any good arguments for why on principle they couldn’t be, although this is often treated as a hard fact.
5/4/2023 how AI shifts the relationship between intelligence and human worth
“I think these things [LLMs] are clearly, whatever they are, intelligent. They working with information in a problem-solving way. I’ve been wondering, why are we so scared of giving up the term ‘intelligent’? Why are we so afraid that something else might get called intelligent? And I think it has to do with how much ==we’ve made that the dominant way in which value humanity==, particularly in a secular dimension. Why is it okay that we treat cows and chickens and the natural world in the way in which we do? Well, ‘we’re smarter, I guess?’…So then if you give that up, ==if you believe these things are intelligent and maybe they’re going to be more intelligent in so many dimensions than we are, then you’ve lost something profound.==
But if that’s not how you value humanity, if you don’t think the worth of a human being is their intelligence — which on some level we clearly don’t, I mean children are wonderful not just because they might become smart one day… — ==there is something about how much we have dehumanized ourselves that I think is getting laid very bare in AI discourse==. If we have such a thin ranking of our own virtues and values that these programs can destabilize it so easily given how limited they are, I think it’s getting at something which is a little discomforting, which is that we have valued human beings very poorly. And ==it would take a lot culturally, and call a lot that we have done into question, to value ourselves and other creatures in the world differently==. But if we don’t then we have no defense against the psychic trauma of this thing that we’re creating.”
[...]
“I think there is a lot of barely submerged guilt and shame, and truly vicious judgement of ourselves in a lot of the conversation [around AI]. I think some of the fear is that if you created something smarter than we are, it would treat us the way we treated everything else.”
— Ezra Klein, in his interview with Erik Davis
5/4/2023
An interesting discussion of the supposed spontaneity of creativity in Ezra Klein's interview with Erik Davis:
EK: “To be McLuhanites for a minute, if the medium is the message — if the medium encodes certain ways of being and thinking that change the people who use it — what do you think the message of the AI chatbot medium is?”
Erik Davis: “Wow that is an extraordinary question. I must admit I’m spinning a bit here...”
EK “Let me try one on you: something that I think is very present in the way people are thinking about AI is the idea that the output is what matters, and that the work of knowledge and of creation is this kind of, ‘you run a search on the information in your head, and then spit out the output’. There’s much more, I think, if you pay attention to yourself as a human being, mystery in that process. For instance, ==the work of writing a bad first draft is not just a waste of time on the way to a good fourth draft. It is often an intellectual space in which you realize you shouldn’t be writing that draft at all, in which you realize you should be doing a totally different piece, in which you realize something you never thought of or that something which isn’t in the ‘training set’ for that draft is actually relevant here.== And it’s that mysterious intuitive sense that leads to great work or oftentimes just decent work... That idea that we can just outsource that — I think it’s a way of thinking ourselves as computers, as opposed to the more slightly mysterious creatures we are...”
Erik Davis: “It’s remarkable the way in which [AI] is an invitation to let go of a certain space of the unknown, the mysterious, the novel, the unpredictable in our own minds. And perhaps one scenario is that it becomes clear that these [machines] are insufficient, and so we honor that aspect of ourselves even more. But there’s also the possibility that it wasn’t necessary all along. And it is easy to imagine a situation where we become used to off-loading more and more decisions, and ==thereby accepting to ourselves that we too are predictable machines==. So I think part of that message has to do with prediction and pattern, and what is in us that is not unpredictable, that is not pattern. Can we isolate that? Can we put our finger on it?...”
continued on are.na...
This page caps at the ~40ish latest entries — partly because it would be laborious to implement a dynamic feed, and partly because I like the idea of this page being finite, like a journal ran out of pages — you can see earlier blocks in the full channel on are.na.