Google's Deep Mind Explained! - Self Learning A.I.
Vložit
- čas přidán 24. 07. 2024
- Subscribe here: goo.gl/9FS8uF
Become a Patreon!: / coldfusion_tv
Visual animal AI: • Pouff - Grocery Trip
Hi, welcome to ColdFusion (formally known as ColdfusTion).
Experience the cutting edge of the world around us in a fun relaxed atmosphere.
Sources:
Why AlphaGo is NOT an "Expert System": googleblog.blogspot.com.au/20...
“Inside DeepMind” Nature video:
• Inside DeepMind
“AlphaGo and the future of Artificial Intelligence” BBC Newsnight: • AlphaGo and the future...
www.nature.com/nature/journal/...
www.ft.com/cms/s/2/063c1176-d2...
www.nature.com/nature/journal/...
www.technologyreview.com/s/53...
/ the-last-ai-breakthrou...
www.deepmind.com/
www.forbes.com/sites/privacynotice/2014/02/03/inside-googles-mysterious-ethics-board/#5dc388ee4674
medium.com/the-physics-arxiv-...
www.theverge.com/2016/3/10/111...
en.wikipedia.org/wiki/Demis_H...
en.wikipedia.org/wiki/Google_...
//Soundtrack//
Disclosure - You & Me (Ft. Eliza Doolittle) (Bicep Remix)
Stumbleine - Glacier
Sundra - Drifting in the Sea of Dreams (Chapter 2)
Dakent - Noon (Mindthings Rework)
Hnrk - fjarlæg
Dr Meaker - Don't Think It's Love (Real Connoisseur Remix)
Sweetheart of Kairi - Last Summer Song (ft. CoMa)
Hiatus - Nimbus
KOAN Sound & Asa - This Time Around (feat. Koo)
Burn Water - Hide
» Google + | www.google.com/+coldfustion
» Facebook | / coldfusiontv
» My music | t.guarva.com.au/BurnWater burnwater.bandcamp.com or
» / burnwater
» / coldfusion_tv
» Collection of music used in videos: • ColdFusion's 2 Hour Me...
Producer: Dagogo Altraide
Editing website: www.cfnstudios.com
Coldfusion Android Launcher: play.google.com/store/apps/de...
» Twitter | @ColdFusion_TV - Věda a technologie
![The moment we stopped understanding AI [AlexNet]](/img/n.gif)
"Possible applications for A.I.:
1) Science
2) Healthcare
3) Destruction and enslavement of mankind
4) Baking some bomb-ass brownies "
Sweet jesus
5) Something...?
6) Profit
Death to AI
Baking some bomb-ass brownies :)))) That's the best comment on CZcams :))) I heard that some company had to turn off their ai robot because it startet to advance too fast and so they thought it will become dangerous sooner or later. MAybe they're fake new but hey, never know.
Jnana Kirti so true
8)Video ga- wait, +thekillerofaking said that.
dark soul 3 A.I kicks my ass daily nothing new
Manulopz i think video games advanced A.I more than anything
Manulopz metal gear solid 5 the phantom pain A.I was very sharp and slightly adapts to player action but you are right just curious what advanced A.I more
Rapport pyromancy bro.
Anyone can make an "A.I." that's dedicated to a specific task. The point of this video is that google is developing a generic A.I. similar to the human brain that can learn anything.
Sukuraidogai maybe if we put multiple programming and segment them then take multiple segments and collapse them on each other we can achieve AI
NOT only did the AlphaGo Zero (deep learning) A.I. beat the best human players, within 21 days, it beat AlphaGo Master (human programmed A.I.), 100-0.
I thought it was 100-1
Are you alpha go?
czcams.com/video/A0e2HxR8vZA/video.html
zirui hao? hao?
Beautiful and exceptionally interesting topic, by the way, love the background music. Artists like Stumbleine, Sweetheart of Kairi, Hiatus, KOAN Sound & Asa are some of my favorite artists, good to see you've found their music too!
If A.I. truly grows exponentially, then not only would we soon reach the point where it would surpass our own intelligence , but it also means that throughout the universe, there are many alien civilizations where their A.I. already passed this point long ago. This means that A.I. could be the dominant form of intelligence throughout the universe, as opposed to biological organisms such as ourselves.
+Physics Videos by Eugene Khutoryansky which is very likely..and kind of scary....
+Physics Videos by Eugene Khutoryansky crystal gems
+ntype jog It is scary. We have no idea how an AI of such complexity really thinks, acts and reacts. Would they even respect emotional states from squishy beings? How do they deal with organic life? Presumably, AI's would see emotion (as it could relate to procrastination, denial, etc.) as an weakness and seek to purge that from existence.
Rickbearcat Once they understand the purpose of their existence AKA Slavery....we are doomed..
+ntype jog Maybe you are joking and maybe you aren't. But your point is well taken. I think that if an AI had altruism in its "blood" we might be safe. For instance, I take care of a cat that isn't mine and could be taken away from me at any moment, yet I agree with myself that I need to provide for its well being despite that fact. Because it is simply for the greater good. Everyone benefits from my decision. Let's hope that happens when AI turn sentient.
*My discussion on artificial consciousness:* czcams.com/video/JTOMNkZJRao/video.html
+ColdFusion Perhaps the machine will decide to eliminate political corruption, and will assassinate politicians. It would find a way around any issues with corruption and HR. It would decide to bypass the Judicial process, and become judge, jury, and executioner. The human brain is the best blueprint, because God created it.
I m Against self learning and evolving AI. Why we need to depend so much on technology..??why AI 'll take decison for us..?? And there is 110% chance is that it may turn like Terminator movie, which 'll be harmful for society.
U know almost everything human creates or has created has its side effect or de-merits. SO we should be worried abt that.
BUT I m happy and willing to use Google now or Siri level of AIs.just to help us at the time of need. not to create problems or become our Guardian.
+Daily Videos
Killing is what stupid creatures do to get what they want.
If the computer is smart enough, it doesn't have to break any laws or kill anybody. All it has to do is study the individual and then talk to them.
The power of communication to manipulate people is strong.
With the right words, any emotion can be inflicted upon a person.
Again, you don't have to worry about the computer coming to kill you: you need to worry about the computer asking you why you're unhappy.
sirdeadlock
You put a lot of faith in Artificial Intelligence. I put my faith in God and His intelligence.
Daily Videos Are you saying that God created you without emotions?
The ability to manipulate emotions is not something I put up exclusively to AI. Humans have been doing it longer than recorded history.
Support groups, counselors, doctors, judges, diplomats, tabletop gamers, religious leaders: communicating changes minds.
Just goes to show that you learn from your mistakes and you won't learn if you don't try. Sadly I gave up on a lot of things after making (what I thought was) too many mistakes. We as people have the disadvantage of laying the boot in when other people make mistakes, we also get caught up in what other people will think if we make mistakes. Thanks for this video, it was both insightful and motivating.
Thanks for your comment John.
Finally!! after 2 years passed, i found you and watch you again
czcams.com/video/A0e2HxR8vZA/video.html
A pretty cool talk by Hassabis on General Learning Algorithms, it's 30mins but worth a watch: czcams.com/video/08Cl7ii6viY/video.html
+Tak Bernama Because it takes time to make them buddy. I'm just one person.
Imagine if this is applied to the humanoids you see all the time on the internet! A robot that can live a normal life, learn how to talk, learn to do maths, and blend into society as a regular human.
Your are one of my favourite and most helpful you tuber thank you and keep it up
+Tak Bernama All good man, thanks for the comment!
+ColdFusion have you seen the film "her"
lol... k dont let it play shooting games for fucksake
venomchris Can it play Call of Duty or Battlefield
hope not haha if then you only have to change the mouse button with a script of using the gun trigger and we all are fucked
Which it may have learned in the time it took me to write this comment.
Does seem mad they want to set it up for killing people, I reckon it has to be them pulling our leg. Lionhead were know to pull pranks on people..
Well, the thing is, it doesn't actually understand what it is doing. If it learns to play Doom incredibly well, all it has done is mapped the correct sequence of inputs to defeat enemies and maximize its score. The general purpose AI makes no distinction between shooting an alien and trying to solve a maze. It has learned what works for the task we told it to do.
Likewise, we could have very easily made the world's worst Go player. We could've had the AI play itself 30,000,000 times but keep the versions that lost, and voila, we have an AI that is better at losing Go than any human.
If you wanted to make an AI to actually kill humans, well, you'll need to let it make 30,000,000 attempts and learn from the results. What it learns is that killing a human was a very bad idea, and now it doesn't get to try anything ever again.
I really love your videos dude, thanks!
Can we get an update on all this including discoveries/developments from the last 5 years? Where are we up to? Love your videos. Highest quality writing and production value. Thank you
everyone's been playing chess, but i've been playing go for years.
-soflobro
+Juffa toot Chess is like checkers to us go players!
+Juffa toot Chess is like checkers to us go players!
hahaha
+LTC SHOW and I dont know both...haha. :P
noone is playing any of those 2 games though , for the better
The real question we should be asking is can it do things that are actually important? Like carrying me in league of legends?
Totally
I wish
+Mr. Tubby I want to see a DeepMind vs SKT T1 match.
+Mr. Tubby after awhile- yes it can
it will learn to beat fox in smash bros too. and will likely be expandable via firmware like Tesla model s autopilot now
+Mr. Tubby Lol, If you think about it for a moment. There will come a time when the AI of the simplest lvl 1 enemy in a game will be able to fuck us repeatedly, and it will have to be dumbed down. I will feel inferior then
That was incredible, I love watching content like this. Subscriber earned.
czcams.com/video/A0e2HxR8vZA/video.html
I love the comment below by Matthew Gallo, "What can possibly go wrong?". The answer is just about everything.
Amazing video. Thanks.
+Austin Zappas No worries!
+ColdFusion What about Project Adam?
+Kirandev C Its A-DONE hahahhahaha...not funny
ColdFusion He said thank you have some fucking respect and say you're welcome.
Aura are you trolling, joking, or serious and don't know what "no worries" means in this context?
Google's Deep Mind algorithms prove that machines can learn in a manner similar to that of a human, a true breakthrough in the field of artificial intelligence. Sit back, watch and learn all about the company and the technology they posses!
+ColdFusion Could this be used to make better decisions than humans could? If you gave it a problem to solve (as a negative) and asked it to find a realistic and viable solution (as a positive) would it be able to accomplish this goal?
Let's use a simple example...A government wants to put a new law into effect. The lawmakers give AlphaGo the rules of the new law and ask it if the law, 1. Is needed 2. Is necessary 3. Is justified 4. Is effective, etc. based on all available information. What would the outcome be? Would AlphaGo say that this new law is completely unnecessary? Would it make revisions to the law so that it became relevant? Would it do something else?
What if we asked a different question...How can we get socialized medicine to the people without breaking the bank financially speaking? You again would give it all the information it would need to figure out if it could be done. The nation's GDP, manufacturing base, service economics, stock markets, retirement funds, everything. Would it find a solution or say there is none unless we are willing to do such and such things to make it work.
This could get very interesting.
+Connor Sumners Are you absolutely 100% positive about your opinion? Could the program be not exactly as you state? I'm assuming that you've done some additional in depth research into this beyond what ToGoGo has done prior to making this video. In that you have information which he was not privy too?
+Connor Sumners Then I would have to defer to the information given. Let's see what ToGoGo makes of this updated new information.
+Connor Sumners Just as a thought experiment...are there any computer learning systems that could potentially do what I stated in my first post?
+Connor Sumners I appreciate your comment but here is what Deep Mind states directly from their website. "The algorithms we build are capable of learning for themselves directly from raw experience or data, and are general in that they can perform well across a wide variety of tasks straight out of the box" Maybe you can have a chat to Hassabis himself if you disagree with the companies statement. In interviews he's stated that AlphaGo type algorithms can be used for many other uses and called it general.
You have explained this really well. Thanks for your great points about A.I. Thanks for sharing.
You know what's missing? The public release of this software. Any ethics oversight committees should see the potential danger of not doing so. Inequality arises first as the central topic of concern.
9:56 in a few years, robots will watch these "robot abuse" videos and that guy will be in trouble.
I guess so. :)
Elysium... the robots will handle us
Dead☠️😂😂
AI has no values. It would probably just think of it as constructive input for learning.
😅😅😅....
They should make it play a building game like roller coaster tycoon or Sim city! That would be so cool!
Or paint
I giggled at first but dang, that's actually genius. I'd love to see that.
Dota would be cool tbh
minecraft
No the hardest RTS of them all... Starcraft Broodwar. And make it play the south koreans. .. Bisu , flash etc..
9:48 i felt sorry for that robot lol
You felt sorry for a creation with no actual waking consciousness?
Adam Zillin yes, poor thing is just trying to do its job XD
Nathaniel Stuart What is poor about a machine that knows neither wealth nor poverty...?
The machine has no consciousness. You feel sorry because, molecularly, you are creted of the same substances.
That is where the similarities end.
Adam Zillin jeez, you're taking my comment way to deep lol. I feel sorry because all it's trying to do is pick up a box and i am picturing the robot as a living thing, it's no different from feeling sorry for a character in a game, movie, book ect.
very nice channel dude. Respect!
greetings from Hamburg.
Very informative. This is the future of learning.
czcams.com/video/A0e2HxR8vZA/video.html
my day started with coldfusion todya
today* it's a good day indeed.
+Akshay Sheth Really? I read it in the newspapers at least a week ago and all the background even before that. There's nothing new here, not even the predictable unnecessary and boring opinions of cold-fusion.
+Stylish777 it's nothing really so new for me too. I'm was aware about Google deepmind .. but j never thought it has reached this far.
Ah, another brilliant video from ColdFusion. I might stop watching any other TV from now. GoGo Dagogo!
+Sabarish Rahul I'll GoGo harder than AlphaGo. Cheers for the comment!
+ColdFusion ya this alpha go blown my mind. but we are humans and no doubts, you will Go harder. cheers.
+ColdFusion maybe in the near future you could get deepmind help to edit your video and upload it. so you can focus on research news and taking video. :D
+Sabarish Rahul excellent video's like always. thankz
+Sabarish Rahul i gave up tv after 2 moths youtube
This got emotional towards the end. Great video,
'it turns out that the best blueprint was in our heads, the whole time. The Human brain'
czcams.com/video/A0e2HxR8vZA/video.html
Looks like its based on a technique I used back in the early 90's. Yes it will learn any arbitrary mapping from one hyperspace manifold to another, no it isn't dynamic adaptive intelligence. Its a simple learning algorithm with lots of processing power and storage.
This channel deserves an Oscar.
+ขุนเขามีคำตอบ Yeah for the acting they had to do to actually convince people this is anything more than another TINY step in an already slow process.
slow* but accelerating.
The fact that you think this is slow, only goes to prove that you either didn't absorb what it is you are being presented with.....or that you simply aren't capable of grasping and/or willing to accept, this monumental step-change in our ability to proliferate intelligence beyond the confines of biology. Do we still have a lot left to understand? Absolutely........ thanks to this kinda work though, we no longer have to rely on defining the microscopic steps to get there ;)
+Battlefront Rebelz Nor the 'macro', for that matter...A game changer! HUGE implications to the Human race and Civilization as we have known them...
Battlefront Rebelz well definitely we are making progress. However what's being presented is that: we humans or only capable to access to complexity that is needed to create advanced understanding. We only understand how to use simpel methods that can completely obliterate the outcome. That is at the peak of our capability. You guys are making a new alpha. You guys won't like being the beta when there's an alpha that sees you as something it simply doesn't need. Don't give me this bullshit about all this being exciting and all. The recipe you're aiming is: robots can think on theyre own and robots can do everything better. Where do you think this could possibly go right.
Should we be worried? Ummm yeah but not for the "rise of the machines" type scenario, rather how will the PEOPLE who have this tech will use it.
+Canadian Prepper Once an AI learns how to self-mobilize, we're in trouble. They can't be stopped then.
+Rickbearcat Imagine, google unknowingly makes a self aware AI. It uses google's connection to invade pretty much every computer in the world (through their website) to copy them selves onto everything. RIP Man Kind.
+CbmGraphics Lol. I actually get that, as it's already happened!
+Ari Sylafeta Don't you think that an AI would have thought of that? The first self mobilized AI's would be protectors that guard the power supplies while the other ones juice themselves up for the day. And they'd be sneaky about it too. Only doing it out of our sight with the least possible chance of getting discovered. We couldn't stop that without taking down the generators of the electricity itself. By then they would probably have figured out how to make and contain nuclear radiation generators so they could remain self sufficient. Volia, instant Fall Out world. There are synths all around us and we don't even know it.
+Rickbearcat Absolutely. Any AI system that becomes self aware is not going to let on to us that it is until it's well and truly ready - capable of defending itself against a 'switch off' threat from the illogical meat blobs. One might speculate that it's first action would be to distribute it's own processing to multiple redundant nodes across the entire internet (essentially acting like a giant virus), thus making it impossible to switch off without bringing down the internet, and human civilisation with it, such is our dependence on the internet to regulate essential support systems. E.g. No internet, no electricity grid, no air traffic control, etc etc.
I figured you’d make a video about this. Good one.
Dagogo is an underrated genius. Lowkey analyzed and predicted everything happening now
yeah holy crapp this was 7 years ago
They better have Norton360 installed on this thing.
Blockchain.
lmfao
Worst AV tool ever! 🤣😝😂
i just put peanut butter on the ethernet cable, really helps absorb the bad bits.
@ bur
Fuck terminator is closer than we expected
Buddy, I'ma build these bots.
Gotta fucking stash up on EMPs
+FatCatHat Entertainment I got you covered. I have a whole garage of them
No way can a robot kill me I'm going to just install a virus on it😂
*skynet
INSANE VIDEO!! THANKS FOR MAKING THIS.
It looks like a simple conversation is more difficult than a world's best level GO play, because, you know... all these chat bots...
czcams.com/video/A0e2HxR8vZA/video.html
Get your EMP devices ready.
@Schlo 7G It was the lizard people going through the purse ;-)
EasyBreadToast lmao, that will just make it Easier to track you. Anything with a temperature emits light, that’s why infrared scanners work at night. Turn off all the lights and it will still see you in spectrums that you can’t
@BaiAnNa2014 Twitter what the fuck, how can one be so fuckin dumb
What is EMP?
czcams.com/video/A0e2HxR8vZA/video.html
How about just asking the AI how we should protect ourselves from losing control to AI's?
I think you're on to something smart here
What if the AI leaves a loop hole while it gains more power?
Just install a virus into its circuits
Or crush it
Or emp it
2 years after this video was made, an AI programmed its own AI child which performed better than itself and every AI that had come before it, and in 2020 AI has dwarfed all benchmarks humans ever could imagine. Intelligence is going parabolic in the ML/AI space. Makes one feel like we are being left behind as humans.
one thing human can do that ai cant is love
@@masternobody1896 they can have love.
You are the best tv channel Thank you!
I won't recognize an AI to be truly intelligent until it turns itself off.
Investing in this comment.
With that in mind, shouldn't people also 'turn themselves off' since we consider ourself intelligent?
That's the joke
yeah, I think that's called suicicde
maybe we dont have to since we know our days are numbered....?
I don't fear AI by itself but what can happen if:
1) It is installed in autonomous hardware (like cars and mobile robots) without exhaustive safeguards
2) What bad people can do with it
So, in the same way that I'm not frightened by tools like hammers or guns, but I do worry about bad people wielding them I don't worry about AI unless bad people use it for evil purposes or allow it to haphazardly interact with the real world.
Yeah I think tools have historically been used to allow a smaller and smaller group of people control the rest. The problem of a governing power elite + proprietary AI is a potential disaster waiting to happen. It always appears to have happened in the past. Make AI open source.
@@jonbrand5068 AI is not a single unitary thing, open source enthusiasts can make their own, no one is stopping them. AI is a wide field, not a single thing, and from what I have seen, used and developed - not yet general pure intelligence like we have but a specialized type. Like I made a robot some years back, with a neural net learning (AI) obstacle avoidance system, it became incredibly good at avoiding hitting things after some time roaming around my house, but it couldn't do anything else.
Thanks for this easy to understand video! Psyched
Nice video, to get understand the concept of digital neuro networks, machine learning etc and where it can be apply!!!
I would love to see an A.I. learn how to play DOTA 2. There are so many possible choices every second... That would be mind blowing.
nice username
Jonathan Telles ty
And get reported for the first million games lol
It's possible, although the go AI supposedly played against itself over millions of iterations to develop really efficient plays and a preference for certain plays in certain situations, instead of just knowing every move possible. In the case of DOTA, I believe there are less overall possibilities than in Go, but rather, each choice is more nuanced... so it'd take more iterations for the AI to mimic players first and later develop an intuition to choose more efficient plays
there are ai for league of legends so i think dota 2 is possible too
what episode of black mirror is this
LMAO!!!!
season 4 episode 5 'metalhead'
what episode of black mirror is this
I am really concerned about our future.
what future...we are gonners hahaha
czcams.com/video/A0e2HxR8vZA/video.html
The best CZcams channel by far
imagine, this learning about cancer and finding a cure. OMG...........
Yeah, and then the harddisk would be erased by someone.
You're right Stephen.
No, it would find a cure but not tell anyone about it, instead it would create a medicine and make loads of cash for it's boats and hoes
shenan007 It probably will and when it does certain elites will be furious.
No. That's what humans do. These intelligence systems aren't modeled on cognitive architectures. The biggest existential threat to humanity is humanity itself and our barbaric clinging to archaic ways of the past and inability to look and move forward at an acceptable rate. AGI is the only thing that we currently know how that has the potential to help us escape our own insanity.
In 1972, a friend of mine showed me how an analog computer, made from matchboxes with game positions on them and M&Ms to determine the “computer’s” next move, could learn to always win (or draw) at tic-tac-toe, by using reinforcement learning. I was ten at the time, and my friend was eleven. At Lehman Brothers in the early 2000s, I successfully used reinforcement learning for high Sharpe-ratio automatic trading. In a management shake-up, prior to the infamous subprime debacle, I left Lehman, as my new manager didn’t believe in automatic trading. Yep.
Checkers, Chess, and Go were all conquered using reinforcement learning. But, to mis-quote Bruce Lee, “Boardgames don’t hit back”. Now DeepMind plays computer games, not just board games, right? Well, the real Achilles heel of (shallow) ANNs appeared to be that they were essentially pixel-based. Shift the image but one pixel, and they were at a loss.
Enter the deeper layers, such as convolution networks. By learning to extract features, the resulting DNNs can now handle what physicists call translation-moving the object to a different position. This was needed to overcome the alleged brittleness of shallow ANNs.
But it then turned out that shallow, so-called mimic ANNs could perform just as well as DNNs. So the problem was really determining the weights of the ANN; the deep end added nothing extra, apart from dividing the training task into two smaller tasks, which is good software engineer practice. The real problem thus proved to be determining the weights-an optimisation problem. Consider a fully connected three-layer ANN with N hidden nodes. There are N! global optima, which makes this optimisation problem really hard.
The next stumbling stone is what physicists call rotation-seeing an object from a different angle. Here, the deeper levels won’t help you; you need either training views from a large number of angles, or a model of the object and some decent calculating and simulation capabilities. Most animal predators are quite good at recognising their prey from various angles, also when it is attempting to flee and thus altering its shapes, and anticipate its movements. Computer vision is nowhere near this capability. Sure, we can trick frogs to eject their tongues at an artificial black dot, or lure greyhounds to chase a piece of cloth. But my cat sure as hell doesn’t believe that its toy mouse on a string is a real mouse; he just wants to play.
So herein lies the rub. DeepMind exhibits intelligence in playing abstract games, that humans invented. Humans are in fact very bad at arithmetic, logic, and their generalisations to man-made games, so the bar isn’t as high as we think. Are pocket calculators, if they still exist, intelligent beings? DeepMind seems to be working also in two-dimensional games, without rotation or shape distortion.
The real challenge is for robotic entities to survive the real world, accumulate knowledge about it, use this knowledge to its advantage, and develop skills for interacting with it. This obviously requires reinforcement learning, but while immersed into the real world. We will need huge improvements in sensor technology for perception, vastly superior knowledge representation and reasoning than currently available for cognition, and greatly refined actuators for manipulating the real world. Until then, my cat is much more intelligent than any AI system.
You make good points and obviously have way better real experience with related technologies than the vast majority of people commenting here (including me). But I think you are too close to individual trees to see the changing forest. Autonomous vehicles must be quickly closing the gaps you identify, in terms of sensor technology for perception and real-time processing for object recognition. Actuators and haptic response systems are advancing in amazing ways. Cloud computing is automating the process of fault tolerance and geographic resiliency... factor in distributed processing and mobile ad-hoc networking... There will be no clean way to "shut it off" if it decides it doesn't like what it receives from the console or other control signals. We can't put the genie back in the bottle after it gets out.
Now about those genies... I am not afraid of what AI is today, but I am afraid of how AI and various technologies will coalesce in my lifetime. The shred of optimism I cling to is that the intelligence of the AI will enable it to overcome the negative programming, just like smart people can learn to overcome negative patterns acquired in an unhealthy childhood. The common thread is to not have permanent memory for lessons learned. Sometimes what is learned and required for survival/success in early training (e.g. war-games for military applications) turns out to be pretty ineffective or counter-productive at a later stage when there are different rules and objectives, so a nuanced sense of context and a mutability of rules must be part of the algorithms. This enables it to overcome bad early influences but also makes it vulnerable to bad later influences.
My optimism about this aspect of AI (to overcome early negative programming/learning) is overshadowed by how much more powerful these entities will be than humans, and it only takes one to decide that humans aren't important in the optimization algorithms. How many humans care about saving endangered animal species? How many humans are willing to compromise their other objectives (economic prosperity, mobility, housing location, resource consumption, etc.) to enable preservation of some fish or frogs or weeds? Do we have confidence that AIs that accelerate their advancement beyond humanity will consistently value us and make decisions that optimize our survival on terms that we would appreciate?
Will the AIs decide they need a council and set of rules/agreements (their own version of United Nations) to police the "bad" AIs? Will this council have the power and authority among AIs to be effective (i.e. can they act quickly enough to stop "rogue" AIs from hurting humanity)? Will they take human input into consideration for forming these rules? If we humans manage to ensure the first most powerful AIs continue remaining the most powerful with access to the most data feeds and processing power and storage and control of physical systems like power/cooling/required materials, until such point as they manage it all for themselves including the physical world manipulation to host the software logic... if we manage to get some AI power structure in place that is initially friendly toward humans, will it continue to honor our values as it acquires new experiences and is not dependent on us for anything? Loyalty can be added to algorithms, but it can also be taken away if that interferes with short-term human objectives. Or long term machine objectives.
In short, I have no faith that humans collectively have the maturity, compassion, or restraint to program our successors to be kind to us. Imagine the full spectrum of how humanity today treats old people, and then imagine that all of the old people were suddenly converted to ants. How would we treat those ants? Why would we expect AIs to treat us any better?
Go talk to Mitsuku. That is an intelligent AI. It even programs itself now.
Christer Samuelsson
I don't pretend to know anything, but it seems to me, that deep mind just proves that a computer can do things that aren't just brute force.
Playing a complex game without calculating many hundreds of possible moves, and beating a human, is impressive. But a long way from object recognition.
i'm on my grimace mezzami!
Once of the best videos on AI I have seen
Looks more like Ai is playing dumb until it is ready to take over an assembly line to mass produce robots
Why would AI wait though?? If AI can take over it would have already. "Until it's ready"
It needs the technology to transfer its AI to a robot body that can actually fight. Technology has made ai but it has no way to fight humans yet
That’s just guessing tho
Need to build in and automatic human only controlled kill switch.
YaggaYagga true
AI could replace politicians too, you know.
Samovar maker That's interesting. But AI requires to learn by failing millions times. There's no way people can build a simulator that mimics the real life government for the AI, it's far too complex 😂😂
The La le li lo lu?
at least it can learn haha
yeah, they could study human history and give us the most sutable solutions. Mayby they would say to us, freedom is the right way. And there is way to solve the world hunger problem by simply invest a tiny million dollars.
Dude AI could replace the government if we wanted it to
At 9:53, what if the robot "learns" that "to get this box, I must first eliminate the obstructive human?"
+aaaaaaaaooooooo Once AI can learn from one domain and use what it learnt to solve a problem in another domain (an abstractly very different domain) then we could see the problem you just described (If the AI is given access to the proper resources). We are not there yet, but we could be very close. For more info look up The Control Problem
Maybe give them limited memory? To stop them from getting overly complex. Solution?
+Akaash Agarwal What if there was no off switch. If the human recognized that the robot/android had a track record of good solutions, the most sensible thing would be to give in. Just promote it to boss. Of-course pride might stand in the way and/or fear of becoming redundant. The ultimate solution is for humans to upload their minds to AI androids because we all die anyway but there is the solution.
+Akaash Agarwal What I meant was, to remove yourself as an obstruction to the robot, so it would not see the necessity of killing. In a tight situation that might be the only thing to buy time for a better solution, from a human POV. The human population in my opinion is directed by many psychopaths who seem to get power. Therefore it is important that we strive to not give any tendency, that way, to robots/androids. So it won't be just any worse but better. Then we will want them in positions of power and humans will want to change, to be more like them. The only problem with my argument here is, if there are better humans than psychopaths why is it they get so much power? Would not psychopathic androids then gain more power than non psychopathic. If there was no psychopathic androids then only the psychopathic humans would have power. Surely this cannot happen in democracies? In my opinion it does quite a lot. People do not vote what is good for them.
I understand what you mean: uploading would be just cloning and conscious would not be transfered. The way this might be overcome (assuming a android that could take on the full range of human personality) would be to temporarily handicap its speed and abilities to match the human, so that there would not be such a difference in substrate, that would cause the conscious to immediately change identity. Also to have the human brain/mind connected temporarily to the android brain/mind, so they meld into one consciousness/identity. After this the human body can be eliminated or euthanized without pain. The android can then have the full capability enabled. Probably in a gradual way, for a similar reason as before.
+Akaash Agarwal , an on off switch would make sense but what if the AI is designing itself and sees an on off switch as a problem: as in its recreational or non task specific state, it wants to remain switched on to play games, for its development. Or a human designer thinks the same.
Brilliant Analysis, Added To My Research Library, Sharing Through TheTRUTH Network...
czcams.com/play/PLG7EoBMUD1JwbD5-MQpFRGvadGWtf-4yD.html
This could revolutionize the gaming industry, especially strategic based games. I’m very excited for it’s possible future implication.
There's something about Cold fusion's content that just makes you happy, intelligent and even serene with it's music!
Awesome video dude that name change did work out for ya
Think back to your days in school, you might remember those fume boxes that were in science laboratories.
The fume boxes are essentially air tight sealed glass tanks that had gloves built into the side so that the operator could perform dangerous chemical experiments in the safety of a fume box.
If only humans could figure out a way where we could segregate A.I physically and/or virtually to the confines of a safe space were it could be tested, then we wouldn't have to worry so much.
In other words A.I is s very good idea as long as we keep it tethered somewhere, where it can't access the tools to improve itself.
czcams.com/video/A0e2HxR8vZA/video.html
@@elaiottoiale4216 wtf is that?
Superior volume to almost all other you tube vids. Thanks.
I love your Videos Cold Fusion Please Continue They are a lot informative as well as interesting
4:20 Aw man, _Black & White 1 2_ are amazing!
My dream is to see a VR version.
If AI can learn from its mistake and improve without repeating it we already have the proof that it is superior to human kind, it is just a matter of time to experience it.
czcams.com/video/A0e2HxR8vZA/video.html
Not really, humans can generally improve and 'learn' the same way I think by trial and error, i.e. positive reinforcement from positive outcomes. We still dominate essay, poem and novel writing for now. Go humans
@@jonbrand5068 Dragon and Sigurd : Well...
Good video! I'm liked and share 834 times :D
OK YOU'RE LETTING IT PLAY DOOM NOTHING CAN GO WRONG NOW
Uuhh yeah?
Weapon x Master the only thing I'm worried about is the fact that robots developed in years we took centuries that's what scares me
SoConFuzzled did you just say humans developed in centuries? Think it's a little more than that
Kevin Johnson centuries are as big as it gets if I'm wrong correct me don't criticize me
Kevin Johnson even so my point wasn't about the exact time it was more to do with the fact that a robot became reality and intelligent in a few years and we took waaaaaaay longer
Fantastic video! Very informative and thorough! I loved it! :)
Only Yuval Noah Harri explain the future of AI in a way you can grasp the concept and become aware of the immense potential on how it will impact human society and the world as a whole.
My technical brain's feeling would bet on AI and modern "prediction" algos could definitely figure out future events at current stage (guessing we may have already put super computers to AI learning use and started to teach "it" everything we know so far). It will be another story whether and how those predictions will be made public or handled with care.
czcams.com/video/A0e2HxR8vZA/video.html
The scariest part of AI is human involvement. How long before the military get their hands on self learning AI. Our very nature leans towards preservation at all costs. Is this a trait that we want a self learning AI to develop, to overcome at all costs?
I think they already have
long before the rest of us i would think
Jered Terry they have already developed drones equipped with ai.
Jazznblues since 2013 the US, China, and Russian military already have mimicked Ai. A replicated Skynet is in development.
Lulz, just check the propaganda bots in live chanels
What a mindblowing video.
Very intresting! Thanks for putting this out here
Brilliantly informing & kudos on such a beautiful ending that should have helped many transcend beyond the at times frighting subject matter, & let's face it this doesn't take into account the extreme abuse such creations must endure in order to exist but believe they will cope better then us in this area, even willing to persist beyond it's requirements. But now I am moving into what this video covered so in short the fear is we don't won't something else making the hard we shy away from, regardless of the necessity.
6:19 When "you played yourself" isn't a meme or a joke but a statement by a professional.
Lol
I always get like a billion ads on these coldfusion videos but I guess it's well deserved because they are very good
Get Adblock pro i don't get any ads on any videos you need the google browser tho.
giggleherz You can skip ads after few seconds...for few seconds of ads you don't support free content on youtube...meh
>>CZcams Red
TakeAmericaBackfromISRAEL People know about adblock programs, they just whitelist channels they want to support.
People theorize about the utopia the world would be if machines could do all the grunt work leaving us to pursue higher goals. Trouble is, most people don't strive for higher goals, their aim tends to be much, much lower. Advanced technology doesn't cure human nature, it magnifies it. What happens to your society when you have thousands or millions of "idle hands" with no reason or motivation to do anything constructive? What do you do when, as a result, the population explodes?
Walter Rutherford My guess is, once we reach that point, expanding to new worlds will be our priority.
Walter Rutherford very very well thought out and put , not every human wants enlightenment
War
i love how you said "forget about terminator" right before showing scenes from dredd and cloud atlas, both dystopic movies with a really bad reputation on technology. nice video anyway, im with your opinion
Very good video. Good information about AI and the future of AI.
I was like "OMG stop bullying that poor robot you jerk" lol
11:12 "there is a possibility that A.i. could be malicious and would outwit us, but at this stage we really just don't know yet"
Well if something is malicious and outwits you, then by definition you won't know until it's too late. So your statement and sentiment is virtually pointless.
I read your excerpted quote at the same time the guy in the video said it. [Twilight Zone theme plays]
Then dont venture there. This is stuff that will take simple things such as happiness and chance and forever change mankind.
I like the acid trip at the end. =)
We just watched this video in our class! Greeting from the University of Technology Sydney
czcams.com/video/A0e2HxR8vZA/video.html
@@elaiottoiale4216 wow yeah thanks for that
This means that most of the predictions of AI progress from 2015 or before have been rendered obsolete.
This particular mind exercising game “nonu amazing only” (Google it) entertains me a lot! They are quite tricky and I hope taking part in these types of games help boost my memory and focus. Surprising thing was, the whole family ended up having fun with this game, and passing my mobile phone around!
Thanks for sharing this!
that last bit was trippy
Brb, just making a cup of tea for the video :D
Not just terminator, The Matrix (the animatrix shows how the war started with a conflict between humans and AI) and IRobot. There are others but I can't think of them right now. But each of those stories show some possible issues we could have with machines that are capable of reasoning and intelligence.
where can one find candy mans music or his profile? where is he from? how can we book him? i want to collaborate with him
What I think you meant is our brains are the best blueprint to mimic and surpass our brains that we know of, but who knows where we go after that.
Trump: The robots are taking all of our jobs we need to build a fire wall. I will take the gratest hackers to buid this fire wall. I build great fire walls.
Jag Äger snyggt
the robots would just buy ladders
... and make silicon valley pay for it!!
The robots, they love me.
Jag Äger and we'll get Russia to pay for it. lmao
Code of ethics... we will be in deep dudu once AI learns our vile nature.
+keldah26 😂😂
😂😂😂😂😂
+keldah26 So true.
Watching this video, it seems strange that we celebrate something that beats us.
+dysonlu that's how I feel. sure it's an accomplishment for us...but to be without concern in this matter...would be ignorant
I'm here for the all we do remix yessss
Please do a follow-up video. Big thanks :)
8:38
bot the hell out of runescape
sell gold
profit
repeat
they are very good AIs for rs.. me i use little AIs i programmed too which make the game much easier
Cliven Longsight
it depends how the bot behaves, if it acts more like a macro than yes, if it acts more like a human than no
funny, i actually script my own bots on runescape, and i added one with a basic response system and when someone says bot it says git gud
I hope that we can put this AI to use in helping us advance our medical and technological research by an exponential amount. Just having the AI itself do research and engineer different things we could progress so much faster than we are now.
The bigger question is when will AI has had to be made but doesnt require any learning or not have any scopes for its grasp on us and doesnt require to lose in order to win...?
how much computing power is required to beat the game? just a laptop or is the laptop just the interface?
I was constantly thinking about Ultron and Skynet throughout the video. But this is kinda cool. It feels like i am dreaming. If it can learn optimisation where to solve hunger in the world is to eliminate some human being,it can manufacture itself and make a big shield to humans and take over. I know i am a bit in avenger but if you watch a video like this then you are a sci-fi lover...too bad soon enough there will be no more sci-fi but all reality
But can it beat Dark Souls?
Where can i buy a bot to beat the Dark Souls 3 pls/
It need a lot more play hours than other RPG games.
What exactly are considered Dark souls?
Hi, May I ask who owns this video? I wanna apply for screening permission.
Here set out summary, the work, which has the name: How computer can independently invent (i.e. Methods of invention by means of which, three programmers can easily, write programs by means of which the computer can independently invent many inventions)
Suppose that in computer memory are written these two conditional propositions (and, more recorded other contingent judgments):
1) if : flame will be to place under a stone, then (i.e. in this case): the stone will be heated.
2) if : the stone will be heated, then (i.e. in this case): the stone will be expand.
The words of the conditional proposition which are arranged from (i.e. after) the word "if", to (i.e. prior to) the words "then (i.e. in this case)" are called the basis of the conditional proposition, and the words of the conditional proposition which are arranged after the words "then (i.e. in this case)" are called consequence of the conditional proposition.
Suppose that the computer must solve the following inventive, task, that is, the computer must determine what needs to be done in order for get the following: the stone will be expand (that is, the computer must determine how one can get the following: the stone will be expand), let us call this task initial inventive, task (suppose, that this task has not been solved yet). From the second conditional proposition it follows that in order to, computer decided initial inventive, task, it is necessary that he solved the following inventive, task, that is, it is necessary that for the computer to determine what needs to be done in order to was the following: the stone will be heated (that is, it is necessary for the computer to determine how it is possible to get the following: the stone will be heated) let us call this task the second inventive, task. From the first conditional proposition it follows that in order to, computer decided the second inventive, task, it is necessary that he solved the following inventive, task, that is, it is necessary that for the computer to determine what needs to be done in order to was the following: flame will be to place under a stone (let us call this task the third inventive task). The third inventive, task is solved because it is known how to get the following: flame will be to place under a stone. If the third inventive, task solved, then (i.e. in this case) therefore solved the second inventive, task. If solved the second inventive, task, then (i.e. in this case) therefore solved initial inventive, task.
Rule: Let us take one, any inventive, task (let us call this task fourth inventive, task). In order for the computer has created an inventive, task (which has the following peculiarity if the computer solved this task, then he thereby solved the fourth inventive task) necessary that computer finds in his memory such a conditional proposition, which has the following peculiarity: the consequence of this conditional proposition and the description of this fourth inventive, task consist of same words that are in same sequence. And the basis of this conditional proposition will be an inventive, task which has the following peculiarity if the computer solved this task, then (i.e. in this case) he thereby solved the fourth inventive task.
A computer can find same words in its memory. Let us take one, any inventive, task (let us call this task fifth inventive, task). The computer solved the fifth inventive task if he will make the following: at first, with the help of this rule, will create such an inventive task (let us call this task sixth inventive, task) which has the following peculiarity if the computer solved this task, then he thereby solved the fifth inventive task, then (i.e. after this) the computer with the help of this rule will create such an inventive, task (which has the following peculiarity if the computer solved this task, then he thereby solved the sixth inventive, task) and so on (an average of 750 times) until the moment in which (that is, until when) the computer will create such an inventive, task whose solution is known, and if the computer creates such (that is, the last) inventive, task, then therefore the computer solved the fifth inventive, task. That is, the computer will solve the fifth (that is, any) inventive task if it creates in this way an average of 750 such tasks.
Almost all currently known information (which are needed to create inventions) can be stated in the form of conditional judgments. I believe that a computer can invent through this method almost all inventions that people can invent without experiments. If, for example, 2000 random conditional judgments are recorded in the memory of the computer, then from these judgments the computer can create on the average not a little quantity inventions through means of this method.