Home Creators Posts Import Register Favorites Logout

Content

last chapters! as usual, we will conclude the book with a collaboritve live writing session. I'll explain more details for how it will work later, but for now, if you want to participate, fill out this when2meet so we can set a time.

~~~~

"This is where Te Hiku's story diverges completely from OpenAI's and Silicon Valleys's model of AI development. Intimately familiar with the devastating effects of colonial dispossession, Jones and Mahelona were determined to carry out the project only if they could guarantee three things----consent, reciprocity, and the Māori people's sovereignty---at every stage of development. This meant that even before embarking on the project, they would get permission from the Māori community and their elders, asking them if the endeavor was even something they wanted; to collect the training data, they would seek contributions only from people who fully understood what the data would be used for an were willing to participate; to maximize the model's benefit, they would listen to the community for what kinds of language-learning resources would be most helpful; and once they had the resources, they would also buy their own on-site Nvidia GPUs and servers to train their models without a dependency on any tech giant's cloud."

As I've mentioned in previous posts, something that I'm always thinking about re AI is, is it possible to do without all the garbage we've spent the last weeks learning about in great detail. The Chilean students imagining new ways of designing data centers was a nice little preview of that, but I think the epilogue's discussion of Te Hiku really exemplifies this ideal for me. Language is something that a "large language model" is actually extremely well suited to work with, and small models with specific and limited scope (as opposed to the AGI dream of an everything machine) are what all the examples of useful "AI" i can think of (e.g. scientific research) end up being. And based on the reading, they really did manage to circumvent all the big issues that the big-tech LLMS rely on.

I'm thinking of doing a video on the idea of technology/science being "neutral"---in the context of the AI discourse, I see this coming up in AI defenders saying it's just a neutral technology and the problems only arise in how it's used, and AI detractors saying that it was designed with exploitative premises in such a way that it can never be anything except evil. this is something I think about in more contexts than just AI (I get comments of people say the same things about science as a whole all the time), and I've been wanting to unpack it for a while now. i will say the book in general really drives home the many ways that the development of AI as we understand it today has been deeply intertwined with some pretty problematic practices/ideas, but i really appreciate that she doesn't seem to take this as a complete dismissal of any related technology. it's a hard needle to thread, but the truth is there's a part of me that is just impressed by and interested in developments in computing, and I'm trying to figure out how much of that I can be excited about without falling into the obviously evil parts of it. The Te Hiku example is really doing something for me in that respect, and I wish there were more conversations I saw around how people are re-imaginng this technology to exist outside of, or even push back on its imperial origins. for those who read "A Third University Is Possible" with me, it's giving scyborg energy.

if anyone else has examples of AI being co-opted by non-empire actors, please share!

also to that end, I'm really curious to read the book "Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity" that was mentioned in a previous chapter---anyone keen on it for the next book club?

Comments

across-overalls

I'm a month behind on reading due to developing my own local book club, so I'll just respond to the text above... Non-empire actors: The Tricontinental Institute has developed AI4SS (AI for Social Science), excerpt: What distinguishes AI4SS from the proliferation of AI tools now marketed to academics is its explicit commitment to what we call epistemic sovereignty: the principle that researchers in the Global South should control the means of knowledge production, rather than remaining dependent upon infrastructures and frameworks developed in and for the Global North. https://thetricontinental.org/ai-for-social-science-reclaiming-research-sovereignty-in-the-age-of-artificial-intelligence/ Security Now podcast episode 1093 has a deep dive into the flaws present in AI's foundation however. As I'm behind on reading, I can't say definitively that this book hasn't addressed this, but the reasons why prompt injection is still so successful is mind-boggling. Actually chapter 5 in Empire of AI gives a quick preview when GPT-3 was trained to look through less vetted material by favoring data that "looked" more like wikipedia. The guardrails on top of modern fronteir AIs, with all their labor exploitation post-training baggage, will either remain a programming inefficiency or need to be re-built from the ground up. I'm further stressed by the knowledge that the nearly 1k proposed data centers in the USA as well as the one I'm fighting in my backyard are not driven by any defined need, nor simple corporate competition each needing their own network of data centers, but by a planetary speculation of AGI driven by the tiniest handful of humans. And yeah, I'm interested in Power and Progress lol

Jacques Vincelette

"get comments of people say the same things about science as a whole all the time), and I've been wanting to unpack it for a while now. " ! Revolutionaryth0t unpacks the Cartesian frame concretely this morning https://youtu.be/A3Zrq-_OKhk?t=333

Jacques Vincelette

p382 Napoleon Bonaparte, the French military leader who led a coup to seize control of the French government, installed himself as France’s emperor, (...) “Liberté, egalité, fraternité”—and how it could be reinterpreted and wielded to consolidate his own power. It was ultimately under that banner that Napoleon did the opposite: He restricted freedom, dismissed fraternity—a philosophy based in unity and solidarity—and granted equality only to French men, not women, while reintroducing colonial slavery. French bourgeoisie promoted an opportunistic stooge to obliterate insurrection in Paris and push war as "Glorious Peace" (https://youtu.be/pJRyXZP5OIk?t=470s). So effective the repression that Les Misérables (1862) surpassed Notre Dame (1831), Adolph was greeted with a parade (1940) and Donald with royal treatment (2026).

Black EnviroNerd

Sounds like an interesting and timely video idea! Also very interested in reading Power and Progress! The only example I can think of to your question is AI being used to ridicule the powerful. I’m thinking of when Elon was heading DOGE and someone put out the AI vid of Trump licking his feet. Granted the propaganda wars cut both ways. Another example I thought of was a tool Yale developed to generate comments to put into regulatory dockets. https://pper.app/co2/ Not the biggest fan of this since it might replace my job one day and kind of freaks me out a bit because of that. Additionally I think communities' perspectives having to be translated by an AI feels a bit problematic to me. Also cuts both ways as AI generated comments have been used to flood dockets to kill environmental regs. https://www.govtech.com/artificial-intelligence/ai-generated-comments-swayed-california-air-decision

AlexTheTutor

1) For me, the question “is Empire the proper metaphor to use?” led to many other questions. For example, using the metaphor ‘Empire’ conjures the idea that it was destined by the stars for there to be a large corporate structure controlled by the few to do its best to impose its will on the many. If we accept this version of reality, then anyone worried about money can buy AI stocks: you will be destined to be among the few. But, the dot-com bubble and many other events in history show this to be too simplistic and reductionist. For this reason, it may be a good idea to read “Power and Progress” to provide a greater historical context, it gives examples (printing press, steam engine) to show that many times these new technologies can and will be used by the powerful to gain more power, but that nothing is inevitable, and anyway, the other questions of humanities… 2) Is Empire the proper metaphor? Is there a better metaphor? What is a metaphor? What is Meta for? A society without a cultivated sense of humanity will begin to lose its humanity, is that what Meta is for? Before any more advances in technology, it is necessary to stop and think what kind of world we want for the future, a world with a sense of humanity or one that eliminates all empathy and creative thought? I imagine a world in which no one uses AI to summarize a book just to get a grade, to earn a degree, to earn a job, to earn money, to have money to buy things, but… instead someone reads a real book and on page 700 something, someone stops and considers the importance of God (or in general a sense of spirituality), the following text emerges: ‘You better think about the extension of civic rights, or even of keeping down the price of meat. You will show your love for humanity more simply and directly by that, than by philosophy.’ I answered him, ‘Well, but you, without a God, are more likely to raise the price of meat, if it suits you, and make a dollar on every penny.’ (Brothers Karamazov) Then, you consider the last time you bought meat (or eggs), and you consider the few who are trying to shove new forms of technology down our throats, and you consider the lack of humanity in these decisions, the lack of spirituality: before all the water dries up, the spirit must have already been dried for many years 3) But, of course, AI is doing wonderful things too: St Jude’s hospital is doing research on Parkinson’s disease and is using AI to accelerate this research; genetic markers are being analyzed using AI to identify disease and develop new screening protocols to improve disease prevention; more improved Gaussian models can create more detailed pictures of molecules, and therefore greater resolution of cellular biology and consequently greater advances in pharmacology to develop better medicines; in cases in which drugs already exist, it should be perfectly possible to leverage AI to completely eradicate tuberculosis from the face of the earth… 4) All that is required is the will to do so… the humanity to drive the motivation… in this way, behind any great advancement in science must exist a poet, a poem, and poetry

Jacques Vincelette

p134 "specific assumption: that AGI, despite being amorphous and unknowable, was also inevitable." p396 “There’s always this false logic around data work: What is the minimum that we can pay these people? That comes from a colonialist logic. Culprit project inevitability to double down on fateful decisions. Grace in ambiguity of poetry and rhyming rhythms.

Dr Jennifer Burgess

I managed to catch up and finish! I liked the note she ended on but overall the general feeling it has given me is one of being much more scared for the future than when we read AI Snake Oil (which made me feel better about it all!). I started looking up some of the people involved and clicking on how they're all linked to each other and there is such a closed little circle of people who are either related/friends from university/in relationships. They don't need a sycophantic chatbot when they have each other to constantly confirm their biases. And the people in this book are all MY AGE which is even more terrifying (I am very much the millennial cliché of just not feeling like an adult). I hope that there might be a natural stopping point in how much further LLMs can be developed (because I don't think that they're limitless or the road to artificial consciousness or whatever), if we can get to a point where their capabilities and use are a bit more stable perhaps we can have space to think about how we actually adapt to this. So whilst Power and Progress might seem a logical next step, I'm not keen on reading any more Daron Acemoglu (especially not such a long book as well). I read Why Nations Fail and thought his theory was unbelievably weak for someone so lauded. I've recently started reading "Goliath's Curse - The History and Future of Societal Collapse" by Luke Kemp (researcher on "existential risk") which so far I am impressed by (nuanced, lots of critique and balanced thinking...so far).

Dr. Fatima

i also went down a rabbit hole of googling some of the people in the book and it was wild seeing them just exist as these people in our world. wild that this handful of randos we can like, read the tweets of, is controlling so much rn

phil e the theyby

I finished this book several weeks ago (I was VERY into it), and apparently reading ahead + several life things getting messy at the same time is a bad combination for actually doing book club posts :P But I'm here now! One idea I keep coming back to is how impossible it feels for an entity to change what it's designed to do. At the end of the day, OpenAI was always a private company focused on developing AGI and entrenching its power and control of it. When that's the purpose, whether they're for-profit or not is largely irrelevant. All of the values-driven statements in the world don't change that this is the fundamental purpose of the company. I feel the same is true of Anthropic; the purpose of Anthropic is to control the technology, so even if the founders have slightly different principles and fears, control is what the company aims to do at the end of the day (although Anthropic still sucks WAY less than OpenAI, and I certainly won't argue otherwise). Institutions that are designed to hold power will always work to retain power, regardless of what other principles they endorse. A couple of gross quotes from "Reckoning": "'One of the failure modes for AI as an industry is basically people losing trust and comfortability in giving up their data" (393) Um....if that's one of your industry's failure modes (that people want agency over their own data), we have a HUGE problem. "The best way to manage it would be for OpenAI to double down on its PR, entrench its relationships with governments, and hold steadfast to its convictions in its vision." (398) Both of these paint the picture that the manipulation of public opinion and use of state power is so central to AI's functioning that the whole thing (as it's currently being pushed within our economy) absolutely collapses without that ability to extort data and infrastructure. And I think that's what the anti-AI crowd is largely reacting to: this intense push for control, and the obvious attempts to manipulate public opinion. I did really enjoy the sections about Te Hiku in the epilogue that Dr. Fatima mentioned, because it shows the potential for this to be "just another technology." But then, if it operated in this form, most of us never would have heard of it or paid much attention to it, just like all the previous versions of "AI" highlighted in AI Snake Oil. The only reason we care about these sorts of stories of other types of GenAI more than stories of any other new technology that does a cool but niche thing is because we're contrasting these instances of ethical GenAI to ChatGPT (and the like) that have overrun our society at a terrible cost. I thought this was a phenomenal book overall; I think it did a brilliant job of painting a picture of the soil OpenAI grew from, particularly the ideological and cultural elements that pushed it forward. It's a bit sad to me that my takeaway is "wow, this whole thing entirely emerged from the same ideological and economic roots as every other problem in our modern world; time to attempt to address that in the same ways I've already been trying to address all of those problems." That said, this book did hone in a lot on the SPECIFIC ideological and economic factors that created this monstrosity, and I do feel a lot more knowledgeable about AI and how different groups of people see it. That helps me empathize with and support people with a wider range of different views around AI, and I also feel more equipped to educate people on this topic. So while the takeaways largely felt pretty bleak, I still see this as an incredibly valuable book.