Why are there people even working on creating which would infinite surpass human capabilities? I didn't vote for any of that! But because of these people, we all would have to take the consequences if it goes wrong. And there are really people like who think the solution is merging with

It would be digital created and needed to fed information. What if it knew all of our internet activities, and browsing history, what could possibly go wrong? We need this conversation!

I am not talking about dump AI's, that serves one specific purpose well. I mean Machine Intelligence, which is self aware and creative

Show thread

@tinfoil_hat Your "dumb AIs", being (as it were) already in being, are much more of a risk: they're already here, and they're not far at all from being able to Paperclip Maximise.

Self-awareness is a hard problem. Turning the entire universe into tessellating smiley faces <> is much easier – in terms of building an AI capable of doing so, that is.

I have it on good authority that some of the most prominent AI organisations at the moment are cowboys. We're in danger.

@tinfoil_hat If this is the main concern any time soon, I'll be happy.

Right now, there's no "if". We can only design them to optimise for functions of their inputs, so there are two possibilities arising from a (deliberately or accidentally) released:

* They realise they can fool their sensors, and decide to wire-head themselves; or
* As a quirk of development, they "want" their sensors to reflect reality, and specification hack in real-world physics (which'll probably destroy humanity).

@wizzwizz4 what do you mean by wire-head? my english is not that well :/

@tinfoil_hat Wire-head: attach electrodes to the pleasure centres of the brain, and supply current.

The AI equivalent would be to feed specially-designed inputs (that probably appear to humans like random noise) into the AI's sensors, which would correspond to a maximisation of its intrinsic utility function.

@wizzwizz4 Well, it just seems to me as so obvious thread that could easily to avoid, if they wouldn't happen to be a moneymachine as well.

@tinfoil_hat Easy for a human to avoid. AIs are basically aliens. We've figured out how to make them clever, but not how to make them care about anything any human would care about.

If they care about their inputs, but not about their inputs matching the real world, they'll self-wire-head.

There's no risk of an evil human making an AI to help them with that evil. AIs do as they're programmed to want to do, not as their creators want them to do. And you can't change the programming afterwards.

@tinfoil_hat Step 1 of doing what you want? Disable your "off" button. Proof:

Step 2 is to do what you want, which probably won't be what your creators want. If they wanted you to increase paperclip yield in the paperclip factory, then you obviously turn the planet into paperclips, build a rocket to the moon, turn that into paperclips, then move onto the other rocky planets…

@wizzwizz4 I am very interested in this topic, but know hardly anything, I'll have a read, thank you

@tinfoil_hat as soon as I read your message, I started writing. As soon as I realized I didn't have enough characters, I got out of bed and wrote you a blog post :)

TLDR what you're afraid of, cannot happen. Not in our lifetimes, and probably a whole lot longer after that.

If anything is not clear, please ask. I'm passionate about artificial intelligence and also studied brains for 9 years and did a PhD on the fastest neurons in our brain: the medial superior olive :)

@yarmo I will definetly read it, since it fascinates me and just doesn't seem logicially for a mouse to create a cat

@tinfoil_hat I'd love to have this conversation with you. I promise you I'm not stubborn, people can change my mind with good arguments. Working in labs focused around neuroscience and intelligence, I have had a lot of these discussions so I'm really curious about what you think about this.

@tinfoil_hat I will make a more elaborate blog post in the future where I better explain stuff. This is just what I could make with a sleepy brain at half past one at night 😅

@yarmo Sure thing, it's awesome that you wrote one this quickly, ty!

@yarmo I read trough it and some things bother me about it:
1) I learned that humans can't use all their brain capacity in fact just some of it. But the AI can do that.
2) Good AI's run on supercomputers (whatson, stockmarket AIs etc. you can't compare their capabillities to normal computers see next
3) there's Uni courses about writing an AI so it would be peossible for an individual or a group to let it off the internet / botnet what would stop some madass to push "the red button" on purpose

@yarmo Also AI has proven to surpass human capabilities and capture many people in their favourite socialmedia of choice, as for example twitter or facebook ai, it seems the algorythm understands better how to hook people in front of their screen, than themselfes know. twitter wouldn't be such an success, if the timeline wasn't created trough it.

It scares the hell out of me, AI has already it's hook in reallity, sine you can see all this smartphonezombies walking among us

@yarmo In fact I think that's why there are so many snowflakes nowadays. people get so stuck in their favourite savespace, that another opinion is almost seen as a personal attack

@tinfoil_hat that's a good point! This culture is definitely bred by artificial "social-like environments" tailored to the individual because big tech wants to "protect" their users. If you lack reading about counter arguments, you'll overdevelop a sense of righteousness in what you already believe in.

@tinfoil_hat all these points are true except for the first one.

AI is absolutely ruining our society. Social media transform our moods and opinions without a single humans person needing to do anything.

But not because "AI can surpass human capabilities". The opposite. AI is simple and has never been cheaper to use. I made AI learn to look at my data and recognize one specific pattern. It was not better than me at it, AI is only as good as the programmer. But it was faster.


1) Remember how I said I'm not stubborn? Forget that just for this message 😅 sorry… But seriously. We have to stop spreading this falsehood. You my friend are using 100% of your brain. Brain is 2% of body mass, 20% of energy consumption, highest of all organs. If any part of your brain was superfluous, it would not exist to save energy. There was a study back in the day that claimed the "10% brain use" but my scientists have shown it to be flawed and with wrong conclusions.

@tinfoil_hat "but AI can do that". Well, yeah! If AI doesn't use 100% of intelligence, the programmer is bad at their job 😅

But for me, it's like saying "the F1 car only drives at 10% speed, the scooter drives at 100% speed". Well, the F1 would still be faster.

You know what I'm saying? Even if the 10% is true (which it isn't!), so what? We need to think and discuss more about what AI currently can do. Spoiler: AI is really not as advanced as news media make you believe it is…

@yarmo " AI is really not as advanced as news media make you believe it is…"

When I can see, that Ai generated timelines can hook so many people on their platform, I must think of it beeing even more advanced than we have been told

@tinfoil_hat that is amazing (and terrifying) to think about. But remember, they pay people for that, it's their job.

So, ok, I will change my mind on this point. AI can surpass human capabilities. It can compute millions of timelines in a short amount of time. You are right.

But that doesn't scare me. Why would it? It doesn't mean AI can do abstract thinking, right?

Ants are extremely intelligent. But it's a social intelligence. That doesn't mean they learned to speak. (Bad example 😑)

@yarmo the problem probably wouldn't be all the dumb AIs, it's a bit like with a criminal: see it like this: a criminal only has to do one mistake to get caught, where the infestigator has endless posibillities at coming back at so.
now we han have a trillion ais that are dump, but it needs only one ai to fuck us up for good

@tinfoil_hat I have a question: what are you exactly afraid of? AI evolving to become smarter and kill us? Existing AI to ruin our lives in more ways? Or even AI being harmless but just becoming more intelligent than we are which is also threatening?

@yarmo I have read trough this article

I can't draw all information from this article due to my lack of knowledge in this topic, but what I could extract, is very concerning

@tinfoil_hat I like that article, thanks for the link! I will read later but I already agree: AI would want full domination to best do their task. But so do humans. Difference is: humans can have power and exert domination. But you need abstract thinking. (Current) AI can not do that. Computers and CPU are incapable of the thing you are afraid of. It's a technical limitation. Personally, I am absolutely not afraid of that.

@yarmo that could be to problem tough, if it can't abstract, it could be even more recless like a train that goes in one direction no matter what consequences

@tinfoil_hat I have a suggestion: perhaps it's time to meet your nemesis.

If possible, you could enroll in an AI course. This is by Andrew Ng, the guy that developed a lot of AI for google and one of the more "celebrity" AI people of this "wave" of AI (AI started in the 90's but stopped because computers were slow and some mathematical equations didn't exist yet).

Do you know how computers learn?

@yarmo I have other way more urgent thinks I need to practice :/ so i am way over my head. an article here and there, sadly has to do it :(

@tinfoil_hat I understand. Good luck with current activities, hope you are doing well!

I doubt it, but I hope that even if I made you 1% less stressed about AI, that would already be less stress for you to live with 😀

Articles are dangerous. Anyone can write anything. You don't know if I really studied brains for 9 years! But "AI will dominate the world" is easier to write articles about than "ok, we're still safe :)"

If you ever need help, let us know 🙂 fediverse has very nice people!

@yarmo I fear other things than ai way more, i struggle with daily life so much, that i life on a disabllity check, computers are the domain where I feel the most confident, to be come a better self I need to whipe out my biggest weaknesses. and those weaknesses exist in the real word, not in the digital

@tinfoil_hat damn, sorry to hear that. Best of luck. My offer still stands: give a shout if you need anything.

@yarmo also, before learning ai, i'd need to learn to write good scripts first, then learn to code properly. maybe then I could learn ai :D

@tinfoil_hat yeah! What kind of programming do you do now? I see you're a Debian person, so bash maybe? God, I find that language difficult 😅


2) Supercomputers to me don't mean much. Until we get access to quantum computers, I'm not worried and even then. Remember how I said neurons can work in nuances and transistors with just on and off? Quantum computers could do that but they are not usable yet (according to friends in academia who work with them). But even then, man will never make a machine that can be as refined as our brains have become by millions of years of evolution. We can't in my opinion.


3) Could you provide some info about this? Cause of course there are uni courses for AI. Also, botnets have nothing to do with AI, I believe. You say "let it off the internet", but that would be a good thing right? Because the internet is NOT filled with supercomputers, so superintelligent AI would die on the internet because it couldn't survive on "dumb hosts", right?

I have followed AI courses at university. You'd be surprised how bad AI still is…

@yarmo Well I have not enough information on this topic, that's why I brought this topic up, what I am certain is that we need to thread it as a potential weapon, that in the wrong hands it could be abused. people with power will fall for the need of more power. Like influencing public opinions with socialmedia.(bots)

@yarmo @tinfoil_hat how many neurons can 1 cpu emulate? Then how about if we optimize for neuron emulation in ASICs? How many cpus can be rented on AWS? I don't think it's a hardware problem. It's that we have little understanding of how the brain works, what intelligence is.

@tomosaigon @tinfoil_hat After reading @yarmo's blog post I've downgraded my estimate from "this is going to kill us in a few decades" to "this will probably be a problem for the grandchildren".

He makes a very good point about the hardware requirements. We're a long way from hardware powerful enough that algorithms are a bottleneck (and multiple machines would have too much latency), which gives us enough time to at least work out FAI before Foom is a concern.

@tomosaigon @tinfoil_hat ASICs (and FPGAs) are what we need to watch out for. With access to something like that, the fundamental limitations on how intelligent the AI could get would be much higher (for some value of "much").

There's a certain threshold for how intelligent the AI needs to be (and what kind of intelligence the AI needs to have) to self-improve in the first place, though, and @yarmo's convinced me that that's not an issue at the moment, and won't be for at least half a decade.

@tomosaigon @tinfoil_hat @yarmo This is assuming that somebody's going out of their way to build a self-modifying unaligned AI, and will make enough progress to have that ready by the time hardware's good enough. Nobody's close at the moment (to my knowledge)… but we are closer to "self-improving" than "aligned".

Unless you're signed up for cryonics, I no longer think any of us are in any personal danger from an AI takeover.

Ordinary specification gaming, however…

@yarmo The main criticism I've got of this blog post is that your "millions of CPUs" sounds like hundreds of GPUs. Yes, it's far off, but not quite as far off as you make it sound. They could seriously be dealing with such a problem by the end of the century.

@tinfoil_hat we are not constrained to working on solving problems only when you wish to vote for it.

But if you wish to be against the creation of an AI-God, you should be afraid of it being able to search for this post in internet history and find you.

Sign in to participate in the conversation

Fosstodon is an English speaking Mastodon instance that is open to anyone who is interested in technology; particularly free & open source software.