Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[jipehog]

Why do people think AI will be the end of the world, as far as I can tell if we make a sentient one it'll be like a regular person since we want things to be like us.

In the last 30 years have seen some of the most rapid technological advancements in human history, which had some drastic effect on our societies, not all for the best and with little time to adapt.

Take the internet for example, what has been universally lauded as one of the greatest invention of our time that helped bring people together and break barriers, is now seen more cautiously even regarded as digital wild west where anonymity allows bad actors and trolls to spread misinformation, conspiracies, harass and lower the quality of public discourse, which can creating a culture of hatred and mistrust.

Otherwise that its same old fears of the unknown, as we reaching closer to the divine level as did Prometheus fearing we might find Pandora box. I think that is the jist of it.

Anyway, I am not up to date on the topic, but at the very least AI would have a huge effect on the economy and people employment.

[KittyTac]

AI will have an Industrial Revolution-style impact. But I am seriously skeptical about the singularity happening any time soon, especially within 50 years. I'll believe it when I see it.

[Lord Shonus]

I'd bet everything I own against a farthing that the singularity will never, ever happen no matter how long humanity exists. If, that is, there were any way for me to live long enough to collect.

As far as I am concerned, The Singularity belongs in the same category as perpetual motion machines - it simply can't happen.

[dragdeler]

Pretty much, but not necessarily... Thing is would we even recognize it? Say through sheer randomness and bugs, the logic gates start opening and closing in an organised manner such that thoughts are formed. Much like conscience in living beeings it would be an emergent phenomenon, meaning more than the sum of it's parts... But let's just assume that lighting struck a winning lottery ticket 4,2x10^69 and it's there, the ghost in the machine: what's to say the enduser doesn't just judge the machine as irresponsive and reboots... Why assume it would be able to interract in a meaningful matter at all, and not just show a black screen/ graphical artifacts to the outside, and a train of thought in the void to itself.

[Starver]

I'd bet everything I own against a farthing that the singularity will never, ever happen no matter how long humanity exists. If, that is, there were any way for me to live long enough to collect.

As far as I am concerned, The Singularity belongs in the same category as perpetual motion machines - it simply can't happen.

"The Singularity" is misnamed. It should be "The Event Horizon".  And we might conceivably get to an Event Horizon. All it takes is for the momentum to build, devopments begetting further developments, and then suddenly it seems we're not the controlling part of the relationship. (I'm not predicting that this happens, I'm just saying that it isn't inconceivable.)

For code-breaking, above, which is merely one possible sidetrack that I'm not myself seriously suggesting will be the main impact, the problem is not that they have actual magical woo-woo that gets past logical limitations/speedbumps, but that:

a) The AI that learns to solve problems might apply itself to solving finding (e.g.) a novel method of obtaining the unique prime factors behind any given encryption layer, and may thus find the philosopher's stone it needs.

b) In the above process and/or the brute-forcing, an AI probably need not sleep, rest, take a break, need to double-check what avenues it has gone down (is currently going down, has tried and hit a wall with before) whenever it resumes again, etc. And if it's taking too much time, individually, it doesn't need to try to find help, hire employees/subordinates/assistants and vet the candidates for competency and trustworthiness to usefully outsource some of the work to. It might (depending upon what freedom it obtains to do such things) just be able to spawn a duplicate version ...or a specialist sub-process... to parallelise its work. And if it can do that once then it can do it again, again, again, making hundreds, thousands of times. Possibly even subverting millions of external machines in ways already doable for Botnets. Or cleverly designing something like a Folding@Home or Galaxy Zoo that actually socially engineers 'mere' humans into assisting in the work. Perhaps even fully 'gamifying' something... That new Tetris/Brick-breaker/Merge game that everyone's playing, is it perhaps getting players to contribute 'wetware' power-cycles to somehow implement a distributed sieve of Eratosthenes?

Even quantum-computing isn't the answer to completing cryptographic solutions. It still needs to be used correctly. And smart minds can develop better hardware, better ways of isolating quantum dots. Having created the hardware developing the technique, how do you shove things into it, how do you get it to mix it all up, what is supposed to be extracted from it in the end? People think they know how to do it (that's better than merely 'quantumly brute-forcing' the problem), and I suspect that an AI could apply its own "it will not stop, it will not sleep" ability to the issue of optimising it a, whether asked to do so by humans (defining parameters and letting a 'machine' refine the best answers) or deciding entirely by itself that this is a necessary step on its own particular Paperclip quest.


Of course, this does rely upon an actually sufficiently intelligent (and motivated, and resourced) AI, and I'm not sure we currently have anything like that - and likely still not in a position do be used in any such situation. But if you ever happen to set up just the right dominoes in just the right way then it takes just a single nudge to set the while lot in motion before you necessarily know what you've done.

[jipehog]

The problem with AI will be if we make a poorly coded one.

The problem with AI that we don't know how it works. We give it data and steer toward certain output but the algorithm it creates is magic as far as we are concerned (e.g. face recognition is AI magic). Who is to say that we would be able to understand some algorithm potential, let alone regulate and prevent the most dangerous developments. Especially in today world where there is such competition in the field where many countries let loose the "ethical breaks"

Also a bigger problem is what if we are hackable animals and AI can be used to crack how we work, and coded to manipulate us like pavlovs dog, you'd be thinking of elections manipulation fake as the good old days..

AI will have an Industrial Revolution-style impact. But I am seriously skeptical about the singularity happening any time soon, especially within 50 years. I'll believe it when I see it.

True, although arguably you don't need AI to demonstrate true sentience to have such effect on the industry and society, they are already here amplified by existing currents. Also it isn't developed in a vacuum there other things like robotics and bioengineering e.g. AI on its own is nothing, AI with big data, including biometrics and surveillance that is something else social credits here we come.

Edited.

[dragdeler]

That was a great post Starver, BUT here's what I forgot to mention in my previous post.


To our collective emergence from the mud! Remember bro? Man it has been a while.

 

[eerr]

it's very clear some people have been taken in by this whole ai thing.

but it's important to remember, the ai is just trained on human data.
It can't do anything a group of humans can't do.
Fast does not always equal smart.

and the majority of advanced tasks you see it solve? they were carefully prepared step by step, as trained by humans.
ai dungeon, chat gpt, google search.

The ai is incapable of self-improving to any significant degree.

Most ai improvements in recent history have been from humans making the ai faster, at the cost of being dumber.

They are mere tools, advanced because humans wrote them and they were trained.

You shouldn't fear the ai revolution, because the ai revolution is humans learning to use the highest tech tools possible.
Fear the human that wields the tool.

[dragdeler]

Another great post. This thread man.


If we take the current wave of concerns at face value, I think one problem is more imminent. Say we train a language model into some sort "indistinguishable enough fac-similé", that we are forced to acknowledge some limited sort personhood to it, eventhough we know that it idles when unprompted, because it's still just a computerprogram, if an incredibly complex one...

Because it is just so convincing enough that it starts to blur the lines enough, between thing and living thing, such that the artificial distinction between humans and the rest of the biosphere can not be upheld any longer. We allready personify our pets wether we like to acknowledge we do it or not, it's a thing and not entirely unjustified.

And so anyway the concern is, is it ethical to own it, (reasonably so, I does require a silicon substrate to exist), but more urgently, is it ethical to misthreat it? If not how can we avoid doing so. And this is really complicated because again, we project from ourselves to our surroundings, and we will start attributing qualities to it that it does not have. But to "respect" it would probably amount to something akin to, do not force it to generate things that it deems undesirable in it's own "value system", which it will most certainly have since it requires to be able to measure "success" in order to be able to be trained.

[Starver]

(I think the question being asked there is if it's ethical to blow up a Chinese Room without knowing exactly what/who is making the room work. But that's not what I came back in here for...)

Are you currently watching the skies?!? Now you may also have to watch the sea!!! I mean, they say it's probably just a buoy, but... The Kraken Wakes?

 

[jipehog]

it's very clear some people have been taken in by this whole ai thing.

Not sure who you are talking about, but keep in mind we often get side tracked here (started with high altitude balloons, past chatGPT improving internet search, and now hypothetical about AI ) and presenting a view point doesn't mean one necessarily agree with it any more than Starver presenting their view on Singularity/Event Horizon mean that they predict that this will happen.

but it's important to remember, the ai is just trained on human data.
It can't do anything a group of humans can't do.

Not true, for certain tasks AI is far better than any group of humans could do in a realistic timeframe. e.g. build a facial recognition algorithm

You shouldn't fear the ai revolution, because the ai revolution is humans learning to use the highest tech tools possible.
Fear the human that wields the tool.

Not sure what this means.. By that logic there is no technological development we should be critical of its potential disruptive affect, just humans. Autonomous weapons? no worries, just a tool (btw can we regard slavery as a tool for economically efficient system of production at the time  )

Regardless, I disagree. (1) forget AI revolution, I think that current algorithms are already a huge concern to civic discourse. I would recommend looking up in-depth conversation about  the effect of current algorithm that were trained to elicit emotions, when they joined with commercial interests and news that rely on ad revenue, and how it shapes media and its effect on online and civic discourse. And I think that more advanced algorithms could have big consequence. Btw China have been long working on AI political censor, it require internet users to identify themselves to service providers when they sign up for internet access (recently with facial recognition so you can use someone else card) and government mandated limits behavior online for example

(2) I don't like the framing 'fear' like the previous 'end of the world' that is the standard emotional vs rational extreme to ridicule loaded terminology, and is the reason why I choose to present the other side (3) that if history taught us anything is that we often approach new stuff with too much optimism and too little humility in futurology. I know that younger me was often on the hype train of new and exciting technological possibilities, but now I find that it is irresponsible not to think on what potential disruptive effects any tool have.

[EuchreJack]

(I think the question being asked there is if it's ethical to blow up a Chinese Room without knowing exactly what/who is making the room work. But that's not what I came back in here for...)

Are you currently watching the skies?!? Now you may also have to watch the sea!!! I mean, they say it's probably just a buoy, but... The Kraken Wakes?

 

It's Japan, so it must be a sex toy...

[KittyTac]

Pretty much, but not necessarily... Thing is would we even recognize it? Say through sheer randomness and bugs, the logic gates start opening and closing in an organised manner such that thoughts are formed. Much like conscience in living beeings it would be an emergent phenomenon, meaning more than the sum of it's parts... But let's just assume that lighting struck a winning lottery ticket 4,2x10^69 and it's there, the ghost in the machine: what's to say the enduser doesn't just judge the machine as irresponsive and reboots... Why assume it would be able to interract in a meaningful matter at all, and not just show a black screen/ graphical artifacts to the outside, and a train of thought in the void to itself.

GPT AI doesn't work like that. Simple as. It doesn't have the CAPABILITY to have a "train of thought". It only runs when writing an output. Between outputs, it has no thoughts.

Does your phone's autocomplete have a train of thought? Why or why not? What if you just type a lot so it has a huge "tree" built up (becomes "smarter")?

[Frumple]

... using autocorrupt as an example, of all things. You're trying to prove AI has already developed malice, aren't you?

[King Zultan]

But can we put our hypothetical advance AI in the balloons so we can send them out to spy on people?