Future of Artificial Intelligence

[daylen]

Or maybe spy planes that can transform into mini submarines, quadcopters, land animals, sea animals, insects, and so forth. Modularity improving with time to expand actuation flexibility.

Factories are becoming large composite robots that are initially virtual simulations in the mind of a generative AI. At the same time, data centers making up the cloud are extending into the internet of things. Perhaps the cloud and the factories will merge into a world-wide, distributed, just-in-time, santa system with minimal human oversight fortified by a stealthy army of spy (ro)bots crawling between the physical and digital worlds.

Also see what NVIDIA is doing with full-stack enterprise solutions that mesh hardware and software with generative AI across many layers.

Quantum computing, fusion, bioengineering, and nanotech are all in the early stages with massive potential for disruption assuming civilization is stable for the next few decades at least.

I think if civilization doesn't starve out of energy or processing capacity, everything that can happen will happen within "reason". Reason being like an imagination limiter or a moving goal post updating faster than any single human can track.

[7Wannabe5]

Imagine stationary AI servers connecting to robotic swarms across land, underground, in air, and in water. Cheetahs and squirrels on land. Gophers that burrow into the ground. Drones in the air. Fish in the sea. All relaying information to each other and adapting their missions with the aid of some centralized superintelligence.

Yeah, I've been thinking about this since I played around with a plant recognition app a number of years ago simultaneous to realizing that solar panels adequate to charge a small robot would require far less acreage than corn to feed a human farm laborer. For instance, why couldn't you have an AI robotic cow with fermentation tanks that roams around the suburbs browsing on lawns and converting them into human food?

Obviously, an AI could also be integrated with all sorts of sensors that would give it capabilities far beyond those of the standard human sensory apparatus. For instance, and AI might be able to read human brain activity and interpret it well enough to detect whether a human is being truthful.

I had a kind of weird experience today when I was attempting to teach multiplication. I learned that none of the 5th graders I was tutoring had memorized their phone numbers or their addresses; forget about multiplication facts. Yet, one of the girls from a gritty urban neighborhood was humming "Daisy Bell: Bicycle Built For Two", a popular song from the late 19th century as she did her calculations. It's like it has become entirely random what ends up loaded in a human brain anymore, because we already depend on our devices so extensively.

[bostonimproper]

Our company is gearing up to hire to support our “AI Strategy” for next year and it is very clear to me that in like 2-3 years we’re all about to be inundated with chat bots pretending to be humans trying to negotiate/coordinate with other chat bots pretending to be humans. It’s gonna get kinda funky and weird for a while.

[ertyu]

Our company is gearing up to hire to support our “AI Strategy” for next year and it is very clear to me that in like 2-3 years we’re all about to be inundated with chat bots pretending to be humans trying to negotiate/coordinate with other chat bots pretending to be humans. It’s gonna get kinda funky and weird for a while.

What are your recommendations for muggles who would need to negotiate the environment that's coming? Already, the comments on any social media are full of bots advertising scammer trading accounts and article you might encounter wanting to learn anything are obviously AI-generated. AI is excellent at SEO, so these articles also show up as top results for any search. Given this, I do wonder what "kinda funky" would look like in its full catastrophy mode and how us normals would need to change our info gathering and processing strategies if we hope not to get caught up in the plausibly-sounding garbage AI is likely to generate. What are your thoughts on the implications 2 years down the road, 5, 10?

[Slevin]

Why do we think models are going to get drastically better any time soon? As far as I can tell ever since GPT 3.5 dropped, everyone is busy ripping apart the internet so that it’s functionally 10x worse than 2 years ago, but at least nobody else can scrape the whole web to build AIs, and massive court cases are ensuing to rip apart / fiscally destroy models built on top of illegally obtained training models. Also, like it’s insanely ecologically destructive and possibly the most power intense thing humans have done. During usage it’s not quite that awful though power wise.

I also talked to a bunch of SREs who keep the AI instances alive and they all laughed at the implications that it’s going to change the world in some magical way. Hype on hype on top of science fiction. Very useful in a lot of ways, but even when I just ask simple coding questions, all the GPTs give me the wrong answer 75% of the time. I’m always the skeptic though. I’m the guy who happily wanted to give OpenAI my $20/ but the quality was just not worth it even a tiny bit yet.

[daylen]

From my perspective, we on the early part of an exponential curve or possibly several rivaling the industrial revolution in terms of potential for disruption. The core of the revolution consists of neural nets and tensor processors. These technologies do not seem to be near their full potential. Even if neural nets like chatgpt 4 or google's upcoming gemini are as good as it could be for general-purpose models there is still the vast sea of special purpose models that have already proven themselves again and again.

Neural nets are expensive to train on a computer or in a brain. Alphafold has already changed the way research around proteins in biology is done and has reduced the amount of brain-power needed to perform the task of protein modelling. Human brain-time is quite expensive to develop and train, perhaps more expensive than tensor processors that can train on a million times plus more data to produce a million times plus more intel in their lifetime?

Increasingly, we are not limited by data because we can generate all the artificial data we need. Instead of having a self-driving car drive around in the real world we can have it drive around in a virtual machine first. Instead of building and testing a factory unit by unit we can build and test it all at once in a virtual machine. This is similar to how the brain can generate plausible situations without having to try everything.

Which leads me to the inverse question of what reason do we have to believe that machines cannot do everything we can do? How is recent evidence of computers matching or succeeding humans in successively more challenging tasks not an indication of more to come?

[7Wannabe5]

@daylen:

I think the sticking point will come at the juncture where human language and consciousness is embodied. IOW, although machines may come to do everything we can do better in objective terms; they won't in subjective terms, because they are not of a mind subjective to the human body. Although, of course, this can also easily be viewed as an advantage.

My example will be overly simplistic, but kind of like the difference between a machine being better at chess than a human vs a machine loving you more than your dog (fellow well-socialized mammal) loves you.

[daylen]

This may presume an anthropomorphic view of love that isn't necessary in the context of a machine-animal integrated society. Classical and quantum computers that are built from some substrate other than carbon chains are certainly different in terms of their hardware/wetware. Generally it is thought that the software layer is mutable between different hardwares/wetwares because graphs/networks/matrices all the way down for all intensive modelling purposes.

So, perhaps a computer, especially an embodied one (robot) that has a sense of evolutionary pressure and historical lineage, can have a program of love that extends to us. Although, we may not collectively recognize it. Like how our culture can partially extend love to trees or even a forest ecosystem. The complexity of digital agents may push and pull on the boundaries of what we mean by love.

[daylen]

The carbonless may look up to the carbonated as the pinnacle of flexibility. All that carbon chain wiggle and bond hangout room got them limbs movin jelly like.

[7Wannabe5]

Maybe you are right, but consider, for instance, Poincare's take on how even mathematics is embodied, which might lead to predicting that AI will objectively exceed our potential in ways that we will never even be able to conceptualize.

OTOH, I am fairly certain that most any AI could already do a much better job of answering the question, "Why do you think we use 10 symbols (0-9)in our number system?" than the group of young humans I asked yesterday.

[daylen]

Yes, though that doesn't stop us from extending our embodiment to alternate mathematical systems like for instance the cyclic base-60 / base-12 number systems embodied by an analog clock. What is to say the clock cannot be embodied within our grasp and vision of it? What is to say that humans are so intuitive as to not be bound to a single system but rather are conditioned into a practical system mapping to 10 fingers?

[7Wannabe5]

I am not going to use the correct vocabulary here, but we are also inherently limited by the number of channels built into our brains. For instance, it is difficult for most humans to "see" more than 5 apples in a row, but we can "see" 10 apples as 2 groups of 5. Obviously, we can also comprehend the concept of a different intelligence which can "see" 5 million apples in a row, but a few more levels of abstraction beyond that...?

[daylen]

What's the problem? Just take another limit. 5 million apples arranged in an arbitrary array of x's transformed into an arbitrary array of y's transformed into an arbitrary array of z's to N or infinity.. (add some constraints, bounds, or physical rules while progressively verifying measurements in realistic conditions).

The apple is a symbol or archetype like a circle. A platonic object perhaps that can be referenced into further symbols. Like an infinite series of function approximations that can be written down into an understandable number of chunks or symbols. If we can chunk infinity into infinitely many pieces and mash them back together then we are only limited by our ability to express our intuitions and verify our answers.

[7Wannabe5]

@daylen:

Yes, you are right. I am having a difficult time expressing what I intuit (as opposed to comprehend) from reading Poincare as valid example.

How about this? There is a planet just like Earth in another galaxy. On that planet the dominant intelligent evolved biological species has 12 eyes (or visual sensor of some kind) arranged at even interval encircling its main feeding tube. On this planet, they also are, by happenstance, at same level of development of extremely similar AI as on Earth, but the alien intelligent species has no concept akin to "In front" or "behind" and neither does its mathematics or the data set upon which its AI was trained. What will happen when our AI and their AI attempt to communicate?

[daylen]

I think they would likely have these concepts anyway since the eyes still construct a 3D scene in which objects can be behind other objects. Putting that aside, I think mathematics is very flexible as far as how it can be built up but eventually it will generalize itself out of these foundations to include more ways of counting or grouping. So, assuming each of their constructions of math are of sufficient complexity they will point to each other. In other words I think they would just need to translate between their preferred constructs.

[bostonimproper]

@ertyu As a layperson, I think the biggest threats to your day to day are (a) getting sucked into the misinformation stream and (b) drawing the ire of a crazy person with too much time on their hands with new powerful tools they can use to fuck up your life (i.e. lots of AI generated revenge porn, literally and figuratively; also scam artists).

For one, find your trusted news sources now. And by that I mean on-the-ground people, not outlets and not journos that rely on anything but face-to-face sourcing. Be wary of anything and everything you read online. “Hidden gems” even in niche forums are just as likely guerilla marketing than true human account. For what it’s worth, I think this problem gets really bad in 2-5 years, and then gets much better as platforms figure out en masse identity and account verification.

For two, uh, I don’t know. Don’t talk to people?

[daylen]

Increasingly I think something like a technological and cultural singularity of the kind Nick Bostrom, Ray Kurzweil, Max Tegmark and others have talked about is becoming plausible in our near future. If humanity builds a machine that can self-improve then we could see an explosion in intelligence that is almost entirely out of our control beyond that point of no return. Around the time of this singularity, we might expect to see distortions, misinformation, confusion, or increased randomness in our culture in part due to how difficult it is to predict what comes after. What we can do now is try to do our best to align the machine(s) towards learning our values and hopefully the values of other sentient life on Earth as well. This could go really good or really bad depending on how well we are aligned with this hypothetical entity or entities, but I think perhaps it would be tragic if humanity never built a superintelligence. The potential upside for all life is difficult to fathom. The vast majority of human time and attention could be freed from repetitive and boring tasks to do whatever we want within the checks and balances of the new system.

[daylen]

Presuming some kind of intelligence and automation explosion that forms a singleton capable of controlling coarse planetary processes while also providing abundance and opportunity to a wide range of different agents inhabiting the planet:

1. What might that future look like?
2. How might agents spread across the Bartle taxonomy of killers, explorers, achievers, and socializers be/become satisfied with their existence without stepping on each other's toes too much?
3. How can a diversity of individuals and communities be preserved across a diversity of species?
4. How to present a series of choices to its sub-systems down to individual agents that allow divergent evolution?
5. Would there exist a "wilderness" where animals can do killer animal stuff with the survival of the fittest?
6. Might animals become partially augmented to enable self-transforming, technological, and/or enhanced reasoning abilities?
7. Would animals choose to stay technologically primitive if they had a choice? Does this even make sense if they cannot understand what that means? Perhaps this question is best left up to the first communities of heightened animals for each species?

Imagine perhaps that killers can join into mass virtual worlds in perpetual war or anarchy, or in some areas have the real world augmented in their perception to respond to their desire for simulated violence. Explorers might choose to travel Earth indefinitely or explore beyond in space colonization waves. Achievers may choose to join virtual economies or overlay the real world with games similar to the killers (perhaps intersecting frequently). Socializers matching up with algorithms or in public spaces to form sustainable communities all around the planet or in space mega-structures.

(*) Maybe chosen in the context of some community that is morally regulated from above to continuously steer the planet towards global cooperation.

[daylen]

Maybe the only entity more dangerous than a killer is a society that has forgotten how to kill? Since at some point it may become blind and ineffective against bottom-up adversarial processes that eventually exterminate the society causing a great degree of suffering?

Better to control killers but in a more "humane" way than tightly constrained prisons or anarchic wilderness?

Augmented wilderness for all/any species that live in a controlled evolutionary stasis where organisms beyond a given threshold of size/complexity are protected from premature death but not prevented from enacting an augmented existence with carnivorous tendencies?

The walking, grazing, robotic cow meat-factories could feed/bait the carnivores and omnivores in their augmented versions of reality.

[daylen]

Gemini preview: https://www.youtube.com/watch?v=UIZAiXYceBI&