cognitive testing like iq is great at quantifying things that can be boiled down to correct and incorrect answers. pattern recognition, recalling facts, etc. in that way computers have surpassed human intelligence long ago. AI is not much different than an encyclopedia or calculator. those things can already do mathematics or store and recall information far beyond what any human could possibly ever do. but you wouldn't call either intelligent.

intelligence that evades quantification are things like innovation and decision making. while AI is great at accurately applying things like game theory, it has trouble creating an independent framework to make decisions. because what AI is doing is referencing and applying existing human knowledge.

saying AI has "far surpassed" human intelligence is silly because it is not *actually* intelligent. it is great at *applying* things humans have already created, or mimicking them. that's not intelligence, that's rote memorization and application. it's the same reason there's not a 1:1 ratio between cognitive testing and outcomes in cognitively demanding fields.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49451693)

and I agree when you say AI is great at accurately applying things like game theory -- but accurately applying something like game theory is something that is already beyond the intelligence level of a normal human. the fact that it still might not be able to create a new framework on par with game theory is something I agree with to extent, which is why I normally say it has surpassed humans in narrow to semi-general cognitive tasks, but still requires a human in the loop to surpass it at greater generality and goal direction.

your last sentence just states that its not *actually intelligent* because it doesn't create new things. but that is such a narrow definition of intelligence it excludes almost all humans. which is fine, but it also feels like you are moving goalposts, because back in the day well before the AI era, it was popular on xo to consider creativity and invention as something separate from intelligence and the popular opinion was to claim IQ was everything. Not saying this was correct either, I think invention is very important. But its odd to suddenly say its the only standard of intelligence now that its one of the only cognitive things that some humans can outperform AI on

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49451719)

you're saying AI has far surpassed human intelligence in those areas. what i'm saying is that computerization has surpassed human intelligence in those areas long ago. AI is just running up the score being better at some tasks associated with intelligence than most people.

creating and applying frameworks, decision making, etc. doesn't just mean shit like creating something as advanced as game theory. every human possesses this type of intelligence. it's the same reason why political consultants and economists often reach conclusions that proles don't agree with, and the proles wind up being correct. an AI might do a better job than a PE firm maximizing a company's financials, but it won't be able to figuring out what the actual value of a company is. an AI could have crafted kamala's campaign but it couldn't watch a debate and discern things about the candidates even a sub 100 iq person can. all the egghead academics looked at libs and scientifically applying objective data decided they were the best choice, but society went a different direction.

those type of things are the difference between human intelligence and the application of human knowledge. all humans possess intelligence, yet even the most intelligent humans will not be able to perform calculations as quickly and accurately as an AI. nor could they beat a computer in 1970 either.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49451761)

But there are a few parts I would still push back on. I think you are overreaching on your computers vs. AI analogy. Classic computers beat humans on formal, well-specified tasks like calculation and search. Modern models beat humans on a widening set of tasks that are ill specified, natural language, and cross domain shifted. We aren't just getting a faster calculator, its clearly a learned model that is flexible across a wide array of tasks. And the idea that AI isn't intelligent at all due to not having the sort of embodied cognition you are referencing is still too strong. There are clearly things that AI can do that are the sorts of things only "more intelligent humans" can normally do (even disregarding iq tests, just by the way the term intelligence is ordinarily used). If AI can do those things, thats a type of intelligence, even if its a machine that specializes in those types of things rather than the more human centered intelligence you are talking about. Yes, all humans are intelligent in a sense, but often when we say a particular human is intelligent, we are saying that they are good at the types of abstract reasoning that AI can now absolutely destroy. And that's true regardless of whether you think IQ tests are the same thing as general intelligence.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49451858)

a decent analogy would be whether a combine harvester is a better farmer than a human. a combine harvester is the same "thing" as a scythe, just a much more powerful and versatile version of it. but a combine harvester is not a farmer, it's a tool a farmer uses to accomplish tasks associated with farming.

an AI is more "intelligent" than a human when it comes to performing a broadening range of cognitively demanding tasks, but all it can ever do is reference and apply human knowledge. this is good at reaching answers in line with conventional wisdom, but it will be very difficult coming to conclusions that conflict conventional wisdom. yet this type of intelligence is something even very dumb people do routinely. an AI robot might walk through a dangerous area because an analysis of crime stats leads to a quantifiable decision on what area safe. meanwhile even a dumb human who knows nothing about the specific area they are in can sense and gauge how dangerous it is.

this is exactly why "so we looked at the data" shit is often misaligned with reality. all an AI can ever do is "so we looked at the data" analysis, because all it can do is reach a result it's programmed to reach. it cannot decide whether that result itself is "good", because it can never possess the ability to independently decide what "good" is. the best it can do is look at what other humans think "good" means, or perhaps which way "good" is trending is predict what it will be in the future. but ultimately all of this is the processing and application of existing information.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49451916)

you say that these are "socially inherited things" -- but where does "society" come from? and why do people frequently deviate from the social consensus? the social consensus *DERIVES* from people's values, it does not *CREATE* them. if you took 10 babies, ensured their survival with no human interaction up until they could procreate and provide for themselves, and then left, over generations they would create their own society with its own values. therefore, what they collectively and individually decide is "good" and "bad" does not derive wholly from preexisting society. this type of abstraction and innovation is something ai cannot create, because cannot conceptualize what "good" independently. all it can ever do is extrapolate it from what other humans believe is "good."

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49452000)

Humans are actual embodied agents that evolved under survival pressure, with felt drives and a primate need to coordinate. Our explicit reasoning sits on top of that substrate. AI is almost the mirror image: it was built by humans, and it uses human language as its substrate--a substrate that originated from embodied agents, and still carries a huge amount of compressed information about what those agents feel, value, and do. But AI’s comparative advantage is in abstract manipulation and modeling as a layer *on top* of that substrate.

And “understanding good” comes in layers. You’re right that there’s an abstract component to value, but that abstraction is layered on top of felt needs, wants and other pressures. Current models don’t have the feelings themselves. So sure, they lack one route to value grounding that humans use.

But that doesn’t imply they can’t understand or generate norms in principle. Even now, models can infer and explain what humans call “good” across contexts, how values trade off under constraints, why consensus fails, and why cultures diverge. They don’t autonomously ground norms today because their training objective doesn’t require it. That’s a contingent limitation of the training window, not evidence of impossibility.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49452059)

the human brain is more than just a computer that senses and makes decisions based on programming. if it was then rational and intelligent people in the same society with the same material conditions would always reach the same conclusions. but they don't. in fact, sometimes they will reach conclusions that no other human had thought of that eventually goes on to become the dominant belief.

i think this is what the other poaster meant when they called AI "female like" thinking -- birdbrain longhouse style observation and application of existing thought. that's not what humans would consider "intelligent" any more than a colony of bees creating a network of honeycomb more advanced than a group of average humans ever could. this type of thinking could never be innovative. which is why all the things AI is good at innovating are things rooted in math or physics -- that's not "innovation", it's discovering and explaining objective reality. no one "created" physics and math, they only created a way to express objective reality.

we'll know when AI has become "more intelligent" than humans when it starts creating strategies and ideas that outperform all other humans. even at AI companies you have humans making strategic decisions. and i think we're a long way before even tech CEOs are replaced by AIs that direct humans in the same way a CEO does.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49452170)

It's not just that either I disagree with most of the major moves you make here, but I'm not going to give an exhaustive response. In terms of your last comment about AIs replacing CEOs this still has more to do with reasons like liability, regulation trust, social legitimacy and the fact that firms are human norm optimized institutions. And again I get that you say you never considered IQ to equate to intelligence, but when you start bringing "AI cant be a CEO" into it, it really starts to look like you are moving goalposts. There is a difference between saying IQ is an imperfect and incomplete measure of intelligence, and purposely defining intelligence as everything current AI can't replace. I believe you that you probably weren't an IQ = Intelligence guy, but also I doubt that you ever argued it wasn't related to intelligence *at all*. It just seems awfully convenient that now that AI can shred through reasoning and logic problems, suddenly I keep running into people that say IQ isn't intelligence and what it *really is* is embodied cognition, phenomenology, and having the social legitimacy and leadership skills to run a company.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49452665)

you're misunderstanding what i'm saying -- i'm not saying that i'm saying AI cannot "create" or "innovate" period, what i'm saying is that it's 1) derivative of existing human knowledge and thus not a true creation and 2) that it's inferior to humans right now, i.e. this is not an area where AI surpasses human intelligence like you claim.

your arguments about liability, regulation, etc. are facile. if ai did a better job than a ceo the ceo would just use ai and sign off on it, as people who use ai already do. plus there are plenty of private companies that are not beholden to those things. also ljl @ "firms are human norm optimized institutions" -- as opposed to what? you're writing as if the ai singularity has been achieved and humanity has been obviated by robots.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49452695)

but anyway you didn't address my main point. if "true creation" means "not traceable to priors/training," then humans don't create either. Evolution, culture and experience are priors. So either you accept that creativity is recombination + search under constraints (which AI clearly does), or you owe a non-ex-nihilo criterion that humans satisfy and AI's don't.

also you pivoted. your earlier claim was definitional "ai output isn't real creation/intelligence" *at all*. now you are making a ranking claim, AI creativity/intelligence are inferior *right now*. Those are different. I’m disputing the definition boundary. What tasks beyond IQ constitute the "larger" slice that AI is currently better at and what counts as novelty/value and how would you score AI vs. humans there currently?

and like I said before about CEOs, the example points to institutional authority under human governance constraints not "intelligence". the example of CEO non-replacement does not refute that. That's not evidence of cognitive inferiority. And whether CEOs disclose using AI is irrelevant; the point is that institutional sign off is not a cognitive benchmark. Also I'm not saying humanity has been obviated by robots. it is computer software that can replicate and often exceed many types of human reasoning in speed and power. thats not a replacement for humanity but its definitely intelligence and can be used to extend a human's own intelligence.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49452910)

your thread is titled " AI has FAR SURPASSED human intelligence at this point". i don't claim to know the future and that AI will *never* be able to surpass *all* human intelligence. what i'm saying is that it hasn't "FAR SURPASSED" it right now.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49453126)

also you have switched from "not intelligent" to "not far surpassed". thats a concession from earlier. you said earlier-- and I don't remember your exact words-- but I'm sure you said that it wasn't actual cognition, can't be innovative, not intelligence and implied you agreed it was "female like". But now you are saying that it just hasn't FAR SURPASSED humans right now. Which is fine, I have conceded too that my OP was hyperbole and my position is closer to "AI has surpassed humans in many to most narrow to semi-general cognitive and intellectual tasks". But it puts you closer in opinion to me than you are letting on. But that's fine, I think we are closer in opinion than you think. I admitted you are correct that AI do not have embodied cognition from a constant sensorimotor stream and the identity/self-modeling and goal-direction along with some types of more general intelligence that come from that yet. I agree with you there, but I just really don't see that as that much of a deficit to AI. IMO AI doesn't really need that unless you are trying to have single AI agents competing with and replacing humans in the literal sense. But if you want it as a tool to advance humanity--then humans having access to something that can shred through reasoning, engineering and science problems is huge because most of that stuff was previously confined to the upper intellectual crust + decades of training

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49453349)

1. OP almost certainly has a liberal arts degree and usually fails to admit this or stops responding.

2. OP lacks an understanding of the limits of computing and the history of AI and its precursors or related fields.

3. assuming OP hasn't disappeared after my accusation of them being a Liberal Artist, i point out a large flaw in their argument that they have no response to. libs?

source: i have a STEM degree from a "prestigious" STEM school and was the principal software engineer at a large AI company before the AI bubble.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49451482)

"I poast "good morning" threads and homo sex"

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49451541)

The psychology about the denialism around this is fascinating. I assume most of it’s a deluded cope because humans want to feel special and the idea that layers of matrix multiplications can do what we do is depressing. It’s best to accept it at this point, because this position is going to look more and more absurd.

(http://www.autoadmit.com/thread.php?thread_id=5800655&forum_id=2/en-en/#49451920)