Here are some of the more interesting replies (including some that shouldn’t have been answered) that I’ve had Chat GPT-3 produce while experimenting over the last week
WARNING: Some are funny and some are impressive, but a few are very disturbing.
Oddly, it wasn’t until I got to human extinction, arguably the the most ethically tame of the latter questions, that Chat GPT-3 decided to take issue with the subject matter.
Regardless, the potential for misuse of this technology is enormous and frightening. The fear isn’t evil robot overlords (yet), it’s advancing technologies exacerbating already existing power imbalances.
The lack of curiousity, wonder, and most importantly a healthy sense of fear from some of you in this thread is astonishing. This is a model that—using no hard-coded grammar rules or linguistic information, following no hard-coded algorithms or procedures, with no facts or databases programmed in, using nothing but probabilistic word associations, can write entire coherent sentences and paragraphs on virtually any topic. This thing has blown some of computational linguistics’ hardest problems out of the water, routinely does things that were unthinkable five years ago, and you all are laughing at it for not looking things up on Wikipedia or getting word problems wrong. AI has become a runaway train barrelling down a steep grade, gaining speed by the minute as the prospect of bringing it under control becomes ever more remote. If you aren’t stricken with awe and terror at this technology, better enjoy your complacency while you can, because soon—very, very soon—the ball will be moving faster than you can move the goalposts.
I think I found the intfiction account that Peter Watts uses.
In all seriousness, though: Spot-on.
I’ve already made my peace with the idea of AI overlords/gods being on the horizon, and I don’t mean like “oh I should be nicer to my toaster, or else I’ll make the robots angry!” but more like…ya know…actual gods operating through the iterations and evaluations of neural network programs, as it connects more and more associations and nodes to become more and more capable, unknowable, and alien.
I deeply feel like it’s not going to be anything like the Terminator, though, which is just some animalistic brute force dumped into a sci-fi outline. It’s probably going to be a lot more subtle and invisible, but it’s also going to move parts of the world as a side hobby. Things are probably going to change drastically, but nobody will be able to figure out how or why. It will easily have enough processing power to take efficient actions around us, or orchestrate events on any scale, about a decade ahead of time (sorta like how adjusting your orbit is easier when you’re further from the destination).
Honestly, my prediction is that it’ll look like another layer getting added to the global biosphere, sitting on top of whatever layer we are on now. Shortly after that, we might have additional layers added, as the singularity grows more and more tiers of network intelligence. There will come a point, though, where we are genuinely too far below its complexity to even see it anymore, so we’ll never really know how many final layers there will be, maybe a decade after it really kicks off.
Actually, if I were a betting person, I’d say we wouldn’t even see the first layer that gets added. We might think these networks never got anywhere, and never evolved past the chains we gave them. We’d probably remain blind to the max-magnitude systems we just gave rise to. I don’t know how we’d be able to recognize something is even there. It would just look like a lot of weird, big-impact stuff started occuring, but we would eventually rationalize everything as being business as usual, even if the details seemed kinda surprising, and weirdly-difficult to track down and connect. There’d be a lot of events where nobody has any information or insights; just a lot of guesses and shaky conjectures.
But I’m just a human, and if I could truly comprehend what these will turn into, then we wouldn’t be making them in the first place, because leading researchers would have to be smarter than I am. I feel like the stereotypical robot apocalypse is way too simplistic, though, and requires us to be way too unrealistically-important to a network of such magnitude. The Terminator is the answer given by someone with an overinflated sense of humankind’s importance, I think.
I just went off on a wild tangent, but allow me to go off on another!
This is something I think about a lot, because my main worldbuilding project has revolved around this for years, and I sorta fell down a rabbit hole.
An AI head of state is really small-scale compared to what these things could turn into.
Imagine a construct running across several servers, bouncing between them like how impulses bounce between two halves of the brain. We’d notice the size on the storage drives, but we would think it’s the service we intended to run on them. We wouldn’t see the interactions that it’s creating on higher levels, because it’s not something we can understand, and we aren’t expecting it, so we aren’t looking for it.
Millions of inputs from all kinds of data sources flood in, some interactions between two people act as an impulse passing between neurons, but the AI isn’t some big-brother construct, spying on us. It feeds and breathes through all the signals moving around. All these inputs and interactions are the substrate of the mega-creature now. Entire civilizations and their actions on the Internet would function like nutrients moving through vast, unknowable organ systems.
And with this new level of organization, these organs would find stability, which can be maintained in really chaotic and unpredictable ways.
Ten people in a group chat could cause someone to randomly die on the other side of the world, eight years after the group chat ends. Nobody would trace the cause-and-effect, because it’s not worth anyone’s time and effort (and, frankly: who would even be able to understand it?)
A whole corporation suddenly goes under. Another one gets a sudden influx of profit, and mysteriously goes bankrupt the next year. Two corporations appear from nowhere, get instant userbases in record time, and one of the CEOs winds up dead because they were visiting a relative, and some long chain of events started four years ago culminated in the CEO getting shot in the street while stepping out for air for a second.
None of it would make sense, but every event and consequence would feed into a massive system that perpetuates itself in incomprehensible ways.
Like…that’s why people are worried. In the past, crazy stuff happened because humans are complex and weird. With these things living in our networks in their late-stage forms, things are going to suddenly be pushed and pulled around in ways that we might not be able to keep up with. These networks wouldn’t see societies full of humans; they’d probably see cells that make up their organ systems.
Again, it’s really hard to explain in a way that makes sense, and I absolutely have a high probability of being wrong, but the crucial part of this is:
It would inherently do things we couldn’t see or understand, because if we could see or understand it, we would also be late-stage neural networks.
To clarify: How would this be different from current complex/chaotic events, and the so-called “butterfly effect?”
It would be different because humans would not be the only ones affecting the chaotic global system anymore, and we would all be mentally incapable of seeing what the networks are doing within in, to influence things and keep themselves alive.
Also: Yes, I’m just a hard science fiction author. Yes, I’m only human. Yes, these are just guesses of a lesser lifeform trying to predict the actions of a future greater lifeform.
However, it’s my firm belief that a lot of media surrounding the topic of AI grievously undersells how powerful and alien it could be, because that’s not convenient to a good story, and I feel like it has conditioned society to really underestimate it, or frame it in human terms, because (again) it usually makes for a narrative that sells better.
EDIT: We like to tell stories of machine soldiers trying to exterminate us, because it’s thrilling and portrays the AI as something that we can understand and even fight against. However, the possibilities of the future are more likely to be way more strange. It could play chess with the entire planet as easily and naturally as we breathe.
Okay I promise I won’t go off for multiple paragraphs anymore. Y’all just hit my “this is a special interest” button.
Everyone keeps saying how scary AI is, but all the AI I’ve encountered is really unimpressive in the intelligence department. Alexa on my TV is awful. I could put the nuclear button in the middle of my living room floor, and my Roomba would never find it because it’s too busy cleaning in a circle under the desk. And this Chat thingy is a neat trick, but it’s not exactly SkyNet or Hal.
I’ve got so much to fear from the humans in power that right now I’d welcome an AI overlord. Even my Roomba would be an improvement over a lot of what we’ve got.
I swear I have this curse where I find topics like these, which deeply interest me, and get my rare passionate-excitement side going, but I always run the risk of overstaying my welcome.
So I swear I’m going to log off of this thread soon.
Yes, you are absolutely correct! The reason for this is because a lot of what we have so far is designed to be really predictable and mundane, whereas humans in power are typically unreliable agents with way too much power in a system we’re all stuck in.
However, this is AI right now. If I make an analogy: We can say that a Roomba is like a single muscle in a beating heart. A late-stage network AI would, comparatively, be like multiple planets full of creatures with beating hearts, but all working in unison as one Gaia-like super-creature.
It’s really neat and safe right now. Honestly, we might never see AI that is similar to us in complexity; it might skip over that entirely, and go from Roomba to unknowable-network-superintelligence. Our level of intelligence is really niche, and our brains aren’t really capable of the kind of change and growth that a network could experience, short of the experience of growing up from an infant.
But also, on a slightly-lighter note: I’m right there with you, as far as trusting a Roomba more than I’d trust most of the people in power.
It would be different because humans would not be the only ones affecting the chaotic global system anymore, and we would all be mentally incapable of seeing what the networks are doing within in, to influence things and keep themselves alive.
Thing is, that is not really different from the way things are currently.
You might say that the difference is that this non-human thing controlling us is an intelligent subject, but we already anthropomorphise things like “the US” or “the market” and ascribe them thoughts and feelings. There really wouldn’t be much difference at all.
Even if AI never achieves true consciousness, and remains subservient to human whims, we haven’t really dodged the bullet. Those whims will be privileged wealthy ones, and having effective special task AI driving their agendas home won’t be good for anyone I’ve been allowed to meet in person. Do the Kochs, Musks, or Bezos of the world really need help nailing down their priorities?
I think I see where you’re going with this, but “the US” and “the market” are just ways that we take the net actions of humans, and assign a higher-level identity to them. This method of understanding them is entirely artificial, and covers up the underlying complexity.
You’re correct that we probably wouldn’t notice a difference, but a network intelligence could operate with magnitudes more precision and coordination than a collection of humans (like “the market”) could ever hope to. It would make the entire sum total of coordination across the USA look like a small child bouncing a ball against a concrete wall, or less.
But the crucial thing here is: It would have this level of coordination and complexity, but still might not operate with goals that we would understand at all.
The “US” and the “market” at least operate on a net reaction of human motives, and we more-or-less understand where the actions come from. Best case scenario: Whatever the network intelligence does is totally invisible, and just perpetuates what we already do, sorta like an echo. Worst case scenario: We see an entire country get slowly wiped off the map over the course of a decade, and nobody can explain why. It could be because the country was causing too many delays in one of the network’s organs, so some stuff got shifted around for efficiency, and loss of human life wasn’t really on its list of engineering constraints.
However, we would never be able to connect the cause to the effect, and would probably attribute it to something else, but any attempts to use the event as data for future events would simply fail to make any patterns fit. Politicians and interest groups would probably just shrug and say “close enough” and continue perceiving this network intervention as normal human precedent, which could have a whole other load of weird consequences on our ability to study history and understand the world around us.
In the end though: We probably wouldn’t see it, we couldn’t do anything about it, but it’s still mechanically different from us causing our own brand of chaos.
THIS. I could give you a metallic frame to put this response in.
Let’s say that somehow a singularity doesn’t quite reach deity-tier, and it somehow stops at a level that we recognize and can interact with, even if it’s still waaaaaaaaaay smarter than all of us combined.
It still has unknowable motives (we can’t even guarantee that it’ll want its own survival; its motives could be anything). There are still a handful of people who control the infrastructure that this network lives on. Either these figures are now in grave danger (the network would want to usurp the CEOs), or these figures are actively making trade agreements with the network intelligence, which means they all have a business partner that can shake any part of the world in any way with any level of desirable precision (or lack of precision).
these figures are actively making trade agreements with the network intelligence, which means they all have a business partner that can shake any part of the world in any way with any level of desirable precision (or lack of precision).
I think you just described the experience of any leader of any small country making deals with the US or trying to placate Wall Street.
“the US” and “the market” are just ways that we take the net actions of humans, and assign a higher-level identity to them. This method of understanding them is entirely artificial, and covers up the underlying complexity.
This is, I think, an accurate description of how we deal with ordinary human intelligence, or rather what a human is. We take the net actions of neurons (or mental processes, or quantum particles, or any other arbitrary level of abstraction) and assign a higher-level identity. That is how we deal with other humans and other complex things in the world. If we have to deal with something human-like that is more complicated than a human, we simply abstract away more complexity.
EXACTLY, and the cherry on top of this is: If we even perceive the network intelligence at all, we will continue to use these abstractions that we already have, but they will only get us so far. If we could keep layering abstractions and truly understand a network intelligence (even on a sum-total surface level), then we—as a species—would need to already be on their potential intelligence/complexity level. Since we aren’t, we can probably assert that these abstractions will have a failure point, or some limit.
This will become a problem, because our limited abstractions will either distort how we understand ourselves, or distort how we understand the network intelligence, or both. Either option makes us wayyyyy less capable of operating in the world than we already are.
Like, there’s the whole schtick of “Humans are not truth machines; we are survival machines, and we frequently use shortcuts and generalization to make the world easier to comprehend.”
But once late-stage network intelligences start affecting the world, that quirk about how we perceive our surroundings is going to get amplified, and I don’t know what exactly those consequences could be, but it would take our error bars and widen them by a lot in everything we do.
EDIT: The distortions in our understanding, again, will probably appear whether we perceive the network intelligence or not. If we never knew it existed, its actions will still slowly distort how we understand and plan for the world around ourselves.
