CharonY

You mean actual obesity? Perhaps. Though I think I was just overstretching a metaphor a bit here.

This goes a bit towards what iNow said earlier- it depends on how you use it. If you use prompts to ask for different opinions and their underlying reasoning, it could be beneficial. But there is little incentive for it to used that way. What seems more common is that folks use it to a) strengthen their existing conviction (multiple studies focused on spreading misinformation, perpetuating bias and so on) and b) seem to drive cultural homogenization (https://journals.sagepub.com/doi/abs/10.1177/23727322251406591). From a perspective of watching the mind of young folks developing I think the reason is rather simple. We basically provide a buffet that has equal access to broccoli, brussels sprout and kale and carrot as well as pizza, steak, chocolate etc. Then we tell them to only eat what is good for you. The maybe top 1-5% will take the vegetable and they would likely do well regardless of the system. For the rest however, the bottom is falling out and you can barely rouse from their self-induced pizza coma. Some do realize how poorly they are doing, but they still aren't able to stop due to a mix of anxiety, trained incompetence, peer and performance pressure, etc. The only messaging we have left is really that you should eat more of the metaphorical vegetable (and also show them how to do it). But this is about as effective as say no to drugs. Or combating obesity (without wonder drugs).

I think it is well-established that these tools degrade the specific skills that they replace. The issue with AI is that it is perceived as not confined to any one domain and some use it to offload basic thinking and, apparently communication. I have heard reports that some students even try to use it for in-class interaction. In one article I read, a simple question was something like: "What do you think about that book. Did you like it, or not?" Apparently, half the class gave very similar chat-GPT generated answers.

Absolutely. Right now, I see them as an amplifier. That goes for both,competence and incompetence. In most cases, it is not equivalent to an expert, as the latter will more likely tell you why you are wrong. The carefully curated cases outperforming MDs tend to be edge cases where certain specialized abilities (e.g. pattern recognition, case matching etc.) outperform the average MD in controlled test. That being said, I also vaguely recall that some cases, radiologist underperform when they use AI. This could be down to who and how the tool is being used. I believe worst and best-performing radiologists, benefitted list from AI, suggesting that a certain level of competency is required to use it, but also that the boost has a ceiling. Typically there are some discussions on the economic system that in medieval times were based on agricultural/feudalist system. I would think that at least peripherally the importance of the Silk Road in shaping the ancient/medieval times as an early version of global trade. I also recall some discussion on the economics of pilgrimages and crusades, including building areas of worship and hospitality around often fake relics (i.e. early form of tourism). I guess history teachers do have some leeway to focus on what they want, but the one I had back in the day loved to talk more about the social science of history rather than wars.

Actually that is the part where I see the most convincing evidence for a good use- if they are properly curated and deployed in a specific setting (i.e. not the general chatbot for the masses). The reason is that medical knowledge is a mostly contained system, where MDs basically use established frameworks to make diagnoses. For that, if hallucinations can be kept in check, they are frightening good and outperform MDs in multiple contexts. That might be, but in research the big hurdle is to convince folks to give you money to test your ideas.

Well, I got a bigger mug. That is what I meant with "processing" i.e. generating ideas based on existing knowledge, but in my mind that process is finite as new physical discoveries are needed, in my mind. So that is the part I don't understand. Why provide them with power for them to generate the money to pay humans to do things, including discoveries. Why don't we pay folks right now to work on ideas that they are not getting realized because they don't get funding? I.e. doesn't AI seem to be an unnecessary middleman for that process?

So this is something that I am very curious about. On the software engineering side, this is very scary as it suggests to be capable to fully replace humans. In the broader discussion this is also shown as evidence that it is inevitable that it will surpass humanity's capabilities. What is your take on it? I may be entirely wrong, but sounds to me that the assumption is that software capabilities are virtually endless. I do not have the expertise to dispute that, but intuitively I would think that there is some limit. Even if it combines all the existing knowledge humanity has right now, and by being able to process it in a way no human can would discover things that humans either can't or would need to do it collectively and slowly, it does not follow for me that this expansion is limitless. At some point all the knowledge that can be generated based on existing one would hit some sort of boundary. At minimum it would require hardware (or people) to do additional discoveries to push boundaries further outward, and really that is where we are right now in my field, anyway. We do not have a sparsity of ideas or hypotheses, we lack manpower and funding to explore them (and ironically, the funding is tighter as quite a bit gets diverted to AI related fields). Or at least, that is my perspective, but I am curious how it looks like from other angles. A phrase from Gibson comes to mind. "The future is here but not evenly distributed." I think that would fall under cleaning up the mess once it is out? I am wondering a bit whether the paste is really all out, whether we really are thinking properly about cleaning up, or whether we are still at the process of squeezing really hard.

I get that. But perhaps because of the way I am trained to think about systems, I am always a bit baffled and disappointed that there is a kind of fatalism associated with that. I get what you are saying, it is there and we need to deal with it. But among decision-makers what I see is more akin to, the toothpaste is out, so there is nothing we can do. And then they are surprised that they keep accumulating toothpaste and that it doesn't in fact clean itself. I guess I am seeing a lot of parallels to other, much slower moving issues, like say global warming, where issues were not only predictable, but actually accurately predicted, a menu of actions were laid out, and then mostly ignored until the issue got so bad that folks then resigned to it, with no real game plan to address it meaningfully. Why not create a better tube? What about effective clean-up system. Or perhaps we can even develop improved metaphors? Increasingly I feel that we are using our brains mostly to please our egos and/or get rich, rather than solving real-life problems. Understood. I have no doubt that your industry is changing. From the outside, it reads to me like the industrial revolution on steroids. I am not entirely sure regarding my position, but I always used to be a tech enthusiast, both in private as in professional life, although technological changes in science undergo slightly different rhythms (typically more hardware than software, with quite a few notable exceptions). But with the cracks in systems and society I am seeing (and again, quite a bit I fear is the old man syndrome) I am increasingly drawn to the human side of things and getting increasingly skeptical regarding the impact of tech in our lives. Not in a Luddite sense of way, but I think one of dimming optimism which slowly turns into pessimism. Or as my wife put it, it feels like we are not aiming for Star Trek (next generation that is) but instead for the wost black mirror episodes. The way the system is moving, it feels that humanity plays a big role in diminishing its role. Which weirdly is also echoed by the US government, which makes me question my sanity before I have my sixth coffee.

There also seems to be a gold rush situation where communities are trying to draw in data centers by providing tax benefits and other things. It does seem a bit short-sighted to me, as I suspect it is not clear how much money they are going to ultimately bring to those communities (beyond the investment in building the infrastructure).

I think it is a matter of perspective. I have no doubt that the impact on coding is seismic. But in my corner of the world, it has been (so far) unable to accelerate the type of science that matters to me, but, in balance, has starting to create a host of kids who are increasingly useless. The folks I see and interact with, are those who are on the hype end. I.e. thinking that it can already replace critical thinking. It might come to that, but not yet. And this is where I see the hype. If I was a coder, I probably would already be switching jobs or try to be the guy who they keep until retirement to keep the agent army running. I guess, the point I am trying to make is that in certain areas, AI are cool (e.g. able to replace administrative assistants), and very impactful when things are mostly digital. In other areas, such as higher education, they are clearly disruptive to traditional learning, but so far have little positive impact. Those who do well might be doing better, for the rest the bottom is falling out. It is hard to be overly enthused in that regard. Research as a whole will have quite a bit of an impact, though most notably it is n computer science at some point and social science, where literature work is somewhat dominating. It is getting more reliable in things like cleaning data, which is important in many areas, but not doesn't free up the time you are wasting trying to teach college kids how to operate a book. I have no doubt that things will change, but at least for some of us it doesn't live up to what we have deal with right now.

So the argument against solar was from what I understand the ability to scale production up and down as needed. Solar was considered to be too tricky. But I have not seen really a quantitative argument. My guess is that there are more considerations and up-front costs that they would rather avoid, but this is only my gut feeling. I am sure there are some studies/report on cost/benefit somewehre. The closed loop argument is also a bit tricky, and I only skimmed some of the issues, so am not really knowledgeable on that front, either. But from what I understand, closed-loop system cost more energy to run, so depending on how the power gets in, the ecological costs are moved further upstream. Otoh there are ideas of also using the heat e.g., for greenhouses and other buildings. I am pretty sure that folks have or are currently doing heavy calculations on the overall burden (carbon emission, water usage etc.). Ultimately, someone has to pay the price.

Not sure whether I already brought it up but college kids are renting smart glasses to cheat in exams. Reports are mostly from China and Japan, but I have heard of cases elsewhere anecdotally.

They are not only popping up in the US. Canada and the EU is working hard on the idea of AI sovereignty (which would be the charitable way to to describe it- it is to me not clear what kind of agreements are in place to actually enforce it properly). Most that I am aware of will rely on LNG due to perceived flexibility and faster implementation. Some have proposed closed-loop cooling systems, but I have not seen details how exactly that works. The one in Utah will have a much more massive impact. However, given the rate they are popping, it is hard to overlook that the mid-sized ones will likely also have considerable issues. That being said, a positive element is that some of the data centers will connect their large capacity to the grid, and the promise here is that they can easily scale use up and down and either consume overcapacity or feed into the grid when needed, thus providing more stability.

The difference is that both are specific tools, whereas the AI is presented as an universal tool able to replace critical thinking capabilities. Ironically, those that have it, probably won't use it that way. Folks that lack it though... Well there are now lawsuits and it will be interested to see if that is going to change how the companies operate their models. It is also in their interest not to put proper safeguards on the system. What has been argued is that it would cripple their capabilities. Adding on top of that that the company spokespersons and leaders repeatedly mentioned how it will eventually be able to solve all our problems, it goes a little bit beyond a a viral moment of a neat tool, IMO. The hype at least feels endless, with the stated goal being AGI. Mechanistically it feels more that they want as much customer use data as possible to generate something that will become profitable rather than merely useful. And the move fast break things attitude, well, it breaks people on the way.

The scaling argument makes perfect sense, though I suspect there will be some nuance regarding what activities require the support of agriculture and which not. I am guessing that in most cases it wouldn't be a yes/no answer, but rather a matter of scale. We do have evidence of very early crafting and arts, but more complex arts really could only develop once food wasn't the key limiting factor of survival, I would guess. But regarding wars, there are (oral) records of First Nations in North America. While some have developed agriculture, others were largely dependent on hunting. I would suspect that the scopes of such conflicts were a bit more limited, but could be interesting to follow up. That being said, I suspect that it really depends on what we consider a war. If that is any large scale aggression between communities, that has likely happened throughout our history (well and our ancestors, considering that our chimpanzee cousins are doing that, too). Military specialization (e.g. making shields and building weapons specifically against humans) was also very prominent among First Nations, including hunter communities, as they developed a highly sophisticated system to sustain themselves rather successfully (which is one of the explanations why some First Nations didn't really develop large-scale agriculture).

There is one more thought on this, now that I think about it. I have been talking with researchers, who have collaborations with China. What I found interesting is that that it seems that in China, AI is intended to be used as a tool and they put a lot of money into operationalize AI, e.g. for robotics or to solve very specific questions. Even in the educational sector their implementation of AI seemed far more supporting learning (e.g. dedicated tools to reinforce training elements, rather than giving answers). Meanwhile, in the West AI is often framed as a thinking tool with the ultimate goal to develop it out into AGI. I found the perspective quite striking, and to me the Chinese approach seemed more grounded. Or at least I have an easier time to wrap my head around it without having layers of hype on top. I am curious, how do you see it? Edit: I should add that I am aware that the Chinese path could, at least in part be the result of the government being afraid that it could be a tool being used against then, but it still (to my mind) represents itself as more rational model, regardless of the underlying motivation.

Ah I read "supporting" in the text as a form prerequisite. My bad.

Intuitively I would have thought that language would predate agriculture. There are societies who largely live from hunting and have developed fairly complex societies. Though there are limits in community size and specialization, and associated forms of technology development, of course.

I think there are two elements of it. Historically, the development of abstract language was probably the biggest change in human history. Other developments, such as writing had huge impact and offloaded some of the effort of oral memories, but those are still entirely human activities. Even when writing affected memorization, writing itself has become a human activity. Here, the activity is offloaded wholesale and there can be entire loops without any input from humans and the role for humans keeps getting smaller. That, I think is entirely new and we really don't know what to do with it. The stated goal of AGI is basically to make human thinking obsolete. Nowhere, in this scenario do I see what the place of human then would be. Sometimes they throw in abundance or related ideas to it, but those are more independent economic discussions, only peripherally related to AI.

No, it is both. Folks do have positive experiences, though at least in my neck of the woods it depends on who you talk to. For example, for those doing more teaching it is considered more a pain. For certain types for researchers they are a good copywriter. For students they are the best thing ever- though the pain of learned incompetence will come much later. But the viral stuff comes from the having AI infused in every electronic device, and loosely the following talking points: don't worry about cost and resource use. AGI will solve all our problem, so don't even think about regulating the system the benefit will outshine all possible negatives. So really, don't think about regulating it also: here is your email/pdf. Do you want me to summarize it? You really don't want to do all the work, now, do you? look, it is just a harmless chatbot. Don't think about what folks can use it for. After all, it is pretty much too late an fait accompli. There is really no use to discuss ethical or other use at this point anymore. It came in fast, and while the companies at the beginning seemed to stress ethical use, it moved so fast in integrating so that there is little to no thought about the consequences on any level. We are in the midst of a great experiment where we are going to figure out, for the first time, what happens if we take an aspect that we often use as the defining factor of humanity, and offload it to an external system for efficiency's sake. There have been cataclysmic developments in the past, such as the invention of writing and other physical record-keeping. But those happened over a long time frame. Now, the companies are pushing for a massive acceleration, by having a popular product so that folks get used to its use, without thinking of consequences. The last technical development I think of with similar impact was the combination of cell phones and social media and that was still way slower than what we see here. And still, we are only starting, probably way too late, to do something about the former. In my mind, and seeing the last few years of students, it is like giving free candy to everyone not thinking about the incoming diabetes crisis. Edit: I also think the term "evangelize" is exactly right. And that worries me, too. Mixing religious fervor with something that is being embedded in almost all aspects of life is something that I am skeptical about. And this is from someone who always had a deep love of tech and what it can do. But largely, I was thinking about it in this framework: But AI is not sold us a tool, and certainly not a precision tool. And in fact it is not used like that, either. It is being used to offload the process of thinking. It has been used to make folks feel less lonely. It feels emotional and intellectual gaps. I think where people are right is that at least some folks are not thinking about it like a tool. It has become an emotional and intellectual crutch.

Yes, but the framing is different. With a jack hammer, you are supposed to learn how to use it before doing so. And if you mess up, you often have somewhat immediate consequences. With AI, it is more marketed as something to not make you think or learn, it will do it for you. Also, it is everywhere, normalizing even the stupidest interactions. I think I would be at least as much annoyed if someone constantly shoves a jackhammer into my face and tells me to use it for everything. My point is perhaps, it is being sold as precisely no like any other tool.

Perhaps even worse. It is not only a stochastic parrot, it is also a stochastic parrot in a mirror. It creates and illusion of something that is not really there but seems realistic enough that the user will project their own thoughts on it. Then, by having their thoughts reinforced what they consider to be external, but, as mentioned in the previous post it fundamentally is mostly a conversation with yourself. This in itself is not necessarily bad, as it can help shaping your arguments. But it falls apart if folks don't realize that because of the way they are using it, it is not really an external agent, it is there to react to your prompts. I see it quite a bit with my students who use it to gain confidence in their reasoning, but it fails to grasp the gaps in the reasoning, and very frequently results in overinterpretation and ultimately false conclusions. The utility of this tool unfortunately scales with expertise.

I will add that especially with regard to journal articles- they rarely provide answers as such (except for very specific things). They add evidence of varying quality to discussions. How they then contribute to answering a a question depends on the expertise of the person who uses the papers. And interestingly, in some fields with with restrictive or at least well-defined knowledge frameworks (much of medicine and engineering, for example) AI will likely perform as well or better than humans. Conversely, in other areas with significant gaps (much of cellular and molecular biology), the undocumented expertise of humans is what differentiates it from AI. I.e. someone working in the field is much better at evaluating the strength of presented evidence, often due to undocumented cues. Finally, the single biggest issue I see is accountability. A reported/journalist who keep getting things wrong can be easily classified as competent or incompetent in the field. Similar to those who write journal articles, in the community folks can get a sense which groups are really good at delivering high-quality research, and who puts out everything that crosses their minds. For AI that doesn't work. Some models are better than others, but even if they are great in one area, they may suck in a different. And each update can make them better in another area, but break in yet another. Ultimately, I think it boils down to how we trust anyone or anything. We can direct trust to individuals, as we can look at specific track records, hold them accountable and/or directly interact with them. Humans are entities that we somewhat understand, if only by extrapolating with knowledge about ourselves. AI are largely opaque, they might change at any given minute and are fully beholden to their owners who can change output at their leisure (with Musk's Grok being a prominent example). I trust Steve with vector modeling. He seems to understand it really well. I don't trust Steve with understanding cell lines. With AI you have to extend the trust to the company and the the whole concept of how AI generates answers.

Why? The math department has Steve. If you need maths you go to the basement and bring him coffee and treats. You just have to follow protocol. First you ask question and Steve explains. Then you just stare at him. And he will explain again. Just slower. Then, very importantly, you have to look like a deer in headlights. He will sigh and then show you. I think he secretly likes that. He also likes sugary treats. Steve is a good guy and doesn't overheat like our computer. This is because Steve is in the basement and hides from the sun. And the outside. His wife said we shouldn't feed Steve after midnight. Steve might be a Mogwai. I prefer Steve over computer.

Troll/propagandist, there is little difference of that nowadays. A tell is characterizing Navalny's anti-corruption actions as "going out of control". From the viewpoint of Putin it might be true, but for an observer, not so much.