Google Just Put an A.I. Brain in a Robot [Research Breakthrough]
HTML-код
- Опубликовано: 15 сен 2022
- AI Documentary: • Who Invented A.I.? - T...
What happens when you put an A.I. Language model in a physical robot? Lets find out.
Everyday Robots: everydayrobots.com/
• Introducing Everyday R...
-- About ColdFusion --
ColdFusion is an Australian based online media company independently run by Dagogo Altraide since 2009. Topics cover anything in science, technology, history and business in a calm and relaxed environment.
» ColdFusion Discord: / discord
» Twitter | @ColdFusion_TV
» Instagram | coldfusiontv
» Facebook | / coldfusioncollective
» Podcast I Co-host: / @throughtheweb
» Podcast Version of Videos: open.spotify.com/show/3dj6YGj...
podcasts.apple.com/us/podcast...
ColdFusion Music Channel: / @coldfusionmusic
ColdFusion Merch:
INTERNATIONAL: store.coldfusioncollective.com/
AUSTRALIA: shop.coldfusioncollective.com/
If you enjoy my content, please consider subscribing!
I'm also on Patreon: / coldfusion_tv
Bitcoin address: 13SjyCXPB9o3iN4LitYQ2wYKeqYTShPub8
-- "New Thinking" written by Dagogo Altraide --
This book was rated the 9th best technology history book by book authority.
In the book you’ll learn the stories of those who invented the things we use everyday and how it all fits together to form our modern world.
Get the book on Amazon: bit.ly/NewThinkingbook
Get the book on Google Play: bit.ly/NewThinkingGooglePlay
newthinkingbook.squarespace.c...
Sources:
www.nature.com/articles/s4200...
sites.research.google/palm-sa...
say-can.github.io/assets/palm...
www.fastcompany.com/90779334/...
www.marktechpost.com/2022/08/...
www.google.com.au/url?sa=t&rc...
michaeltrazzi/sta...
www.quantamagazine.org/how-ai...
aia-gallery.imweb.me/nowexhib...
My Music Channel: / @coldfusionmusic
//Soundtrack//
Burn Water - Take Flight
Sleepy Fish - School
Andre Sobota - Concluded
Aphex Twin - Stone In Focus
Cahb - Secrets
Burn Water - Nostalgia Dream [The Deep Mix]
» Music I produce | burnwater.bandcamp.com or
» / burnwater
» / coldfusion_tv
» Collection of music used in videos: • ColdFusion's 2 Hour Me...
Producer: Dagogo Altraide 
Наука
Yeah, he's currently the CEO of Facebook.
😂😂😂😂😂
How rude! Zuckerberg isn't a robot, he's an alien!
How boring is this comment.
Hahaha you mean mark Zuckerborg
Did you check out its recent update on joe rogan show? Tech has come so far, it now has facial expression and near perfect human emotion simulation.
"What is my purpose?"
"You pass the butter!"
"😔"
Such a stupid scene. If passing butter really was the purpose then the robot would not be disappointed after realizing it.
These are not general AIs
I crack up when I see this joke 🤣
I'm pretty sure that COVID and the massive flow of people using the internet during lockdowns have contributed massively in training AI models. That's a part of the massive improvement of AI of the last two years
Never thought of it from that perspective but the data flows must have been huge, however from a human behavior point of view could this have also added bias to the system ?
Covid did nothing of the sort people were already using the internet and in the next ten years an extra billion people will come online.
Sad reality
Very good insight.
@@MEGASTRIX thanks
this can't end well 👀
....yep
Like humans can do a better job? Eg CCP, Putin, the stranglehold that massive corporations in the US have over the citizens, human rights abuses left right and centre, etc. If an election ever came i'm giving my vote to the robots! 😄
Human: I spilled my drink
Robot: Destroy all humans. No more drinks spilled.
Good Lord. I'd be number one on their target list lol
what's to stop you from telling it to game over ppl? next decade gonna be horrible
This is an example of perverse instantiation.
@@sp33dy1979 bill gates about to buy truck loads of these mechanical minions lol
@@TheBobbybbc y
- Help, my robot is trying to kill me!
- Thank you for contacting Google Support. We'll get back to you within 5 business days!
please contain your robot and ship it to washington dc..
Then they send it back 4 months later and it’s still trying to murder you
@@stumpedii8639 Please send a service request to the motherbot!
Chapta: "Please name the first 10 presidents of the US".
Telephony AI: Please restart your robot and call again if that does not resolve the issue
There is an episode of " love death & robots" that is literally exactly this situation where a lady's automatic vaccum tries to kill her
Love your work man ! Just wanted to let you know the link in your description to you Aussie merch doesn't work. I watch your videos as soon as they appear my friend✌♥
Thanks for the links to your sources and research! Great job on another awesome video
Believe it or not, SayCan is already outdated. Two months ago, Google built on top of it to create a new system called Inner Monologue, which breaks each task down into a series of live-updating steps. So it gets a snapshot of the environment and a natural-language task, then uses the LLM to decide on the first step, then converts that step into actions and executes it, before repeating the loop over and over until the task is complete. Because it's thinking one step at a time, and getting updates about the environment at each step, you can interrupt it, change the request, change the environment, even change the language you're requesting in mid-task, and it has almost no trouble adjusting and re-evaluating its plan whenever needed. It's extremely capable and highly adaptive. Which also means that if it makes a mistake, it immediately knows it's messed up because the new environment data won't be correct, and it will immediately try and remedy its mistake.
Transformers have brought so much to the acceleration of AI development. I can't wait to see what it's like in a few years.
Great. Lets see it put an arm back on. Whoopsie, right?
This video, and your comment, along with many other things all happening, have convinced me that humanity has entered the Singularity. Much like an object approaching at light speed, we could never see it until it was already here. I don't pretend to know what this portends. People must begin to realize; both time and change are accelerating.
What youtube channel can i follow to be up to date on these advances?
@@neithanm Two Minute Papers - ruclips.net/user/K%C3%A1rolyZsolnai
"Transformers have brought so much to the acceleration of AI development"
Those darn Decepticons! Sneaking their high tech machine brains into our society utalizing our own tech right under our noises!
JK ;) can't help but to giggle and think of TRANSFORMERS everytime someone brings up transformers in an AI conversation.
A few years from now, you'll say "I don't like that person", and it'll then find a "solution" to that problem.
lol
🤣🤣
The channel owner has excellent choice in music. just well put together product
Interest video and thanks for having so few ads! Subscribed
Sounds like the beginning of every dystopian sci-fi.
Only if you're a womanchild who can't dissociate fiction from reality.
Animatrix on the way
@@hippopotamushelmet It's fucking tiresome.
Or the beginning of every utopian sci-fi.
That's because "government destroyed humanity" is too realistic to be fun.
We have only 10 years left until we basically have a robot that is basically Wal-E in Real life at this point.
or terminator 😁
Even earlier than you think.
I hope its just as cute as Wall-E
I dont particularly care for humanoid robots
its a bit to *uncanny valley* for me
cute is easier since its a wide range
and then its easier to accept its failures
when cute things fall or fail, its adorable
or Eevee 😄
@@kirillvine7421
No no
Robots 🤖 will support humans
Versus djins and their dragons
Timely material and very well presented (as usual). Thank you for all the work, and for sharing. And by the way, your book was excellent and I am watching /wafting for a new one: --- "New Thinking" written by Dagogo Altraide --- This book was rated the 9th best technology history book by book authority. In the book you’ll learn the stories of those who invented the things we use everyday and how it all fits together to form our modern world.
This is general purpose AI. This could be the biggest revolution since the computer. The amount of possibilities opened(good and bad) by this technology are mind blowing.
The pace this technology is moving at is scary and extremely exciting at the same time.
and so it begins
I was hoping for Interstellar travel and aliens but AI, Robots and simulated worlds are cool too.
we're doomed! Not because robots will kill us, but because corporations will replace every job they can with robots. 90%+ of jobs can be replaced pretty easily with robots, and within 10 years it'll likely be cost effective. We're doomed
Actually, technology is at a stalemate. My professor said that everything that needs to be discovered has been discovered. At the moment we are just re-hashing what we know into new applications, but no new real technology is being discovered. Even this AI / Neural Network tech is old tech from the 1950's. Google is betting big by pushing AI, in the hopes that it will break that stalemate and open up new tech frontiers.
Personally, I think space is where we will find new technology. That is why there is a strong push from some quarters in that direction.
You ain't seen nothing yet
Well done to the guys working on this project. Hope they remember to put in an off/on switch too !
people can turn their phones off now, doesn't mean they do.
Dosn't mather when it is out of pandoras box there is no going back.
A completely kill. Switch disconnect or destroyed capacitor
Computerphile did a while video on the "Stop button problem". As soon as you have an agent that is aware of its environment, It is now aware of you coming to shut it off. If it has arms and treads it now has the ability to try and stop you. If it's trained with reinforcement learning by default it has the motivation to try and stop you. All AI (without very complex programming which we have not yet discovered) will have this behavior. I'm not joking, This is your life now.
@@josiah42 Most devices are designed to have some kind of software control with power management/reset commands now.
I remember when computers came with a dedicated "reset" button connected to the motherboard that would send a signal directly to the CPU. Now you have to dig through menus or hold down some bizarre combinations of buttons for several seconds just to shut a device off.
Seeing how Google's AI bandaids they've slapped on RUclips to manage content and comments peform, I'm not too thrilled with this idea.
I'm so curious to see the negative implications of all this innovation
Cheap police robots. It's way easier for a dictator to control a populace. Heck, even one guy with a lot of money can buy enough of these to overpower a human army.
If we are not already there, we will soon be at the point where innovation is happening faster than society's ability to respond. We are not ready for AI taking over all jobs 10 years from now (to include thinking jobs and leadership positions), but it could happen.
"don't subscribe, browse channels videos instead". Finally a creator that understands users. Thankyou
I am a blind guy and I would love to have a robot who could assist me with visual activities. Drive a car, take me places, pick up objects.
I'd better wish for cybernetic vision implant.
@@alexxx4434 i don't think he can read you comment 🙄
@@alexxx4434 Yea I would want Apple vision series 4. Its running on Vue OS 3.
Me too but how much time have you spent screaming at google home recently. I’ll jump for joy when other people have taken the hurt of growing pains from their new whiz bang robot.
@@romeo72899 ...was that sarcasm, or do you really not know that screen readers exist?
Lol.
I spilled my drink.
Robot: Your out of paper towels, should I place an order? Also swifer is running a sale, should I purchase a mop for you, its 30% off!
(Its google at some point - itll just become an add platform.)
8:41 Robo is digging through trash, moving toilet seats, and making dinner. I hope it learned how to wash its claw.
With "I've spilled my drink", there are multiple interpretations depending on the robot's purpose, the environment and the occasion.
Bartender robot. The requester is a guest: Solve the economical situation. Get a new drink for the guest. Transfer the "cleaning up" task to the other robot suited for it.
Assistant robot in elder care: Check the status of the one who spilt the drink. Assist if needed. Clean up. Eventually, get a new drink if needed.
Construction robot at a building site: Check if the spilt drink is a safety hazard. If not, resume/continue the current building task ignoring the spilt drink. If it is a safety hazard, perform the correct actions for such a hazard.
All these boil down to the robot's a.i. having to know the answer of "What am I?" For a preprogrammed robot with no A.I. it simply follows programming. For an A.I. it needs to know its purpose, capabilities and its own mission. There's no "If switch case runs out of options, stop operating and throw error." with A.I's
Further development and research in this area will give rise to so many use-cases, this is both quite exciting and scary.
I previously worked in a large company in the cost production department. We had about 300 employees. Our job was to take an order and then figure out the parts, costs and timetable needed to manufacture and deliver the product (photocopiers). I always thought that one AI computer could really do all of our jobs. I think the company was aware of this and was afraid to look into doing it. But it is probably coming. (And think of the cost of the product when that department is reduced to only 10 - 30 people, and that is only one department.)
shut
Yet another awesome video from CF! Always comprehensive and well articulated!
Thanks for the video. That Aphex Stone in Focus is always so nice to hear !
someone mentioned "rewarding" the robot for completing tasks. would love to hear some examples of this
Which shows the cruel bstds made it sentient so it can be happy from rewards, and hurt by punishment.
It's not a reward like a cookie, it's just negative feedback on the training cycle.
A happy ending for a task completed .
I think one of the amazing things about robot ai is that it can copy it's knowledge to other robots instantaneously after only one of them has learnt a task.
Skynet here we come!
and imagine the skills it would gain in mass use with hundreds of thousands of bots learning
While im all for the advancement of robots and AI, I dont think it would the best idea to quite literally turn all the robots into a hivemind, especially if it was large enough to be impossible to regulate the flow of information
@@magnateze it would be very easy to regulate the flow of information as it would go through the company's servers. i dont see where the problem is with a hivemind
That’s exactly what I was thinking, you could build 100,000 robots and have them do the same thing but that doesn’t even seem necessary since they can simulate the same exact conditions in artificial reality so no need to even build all the different robots, it’s just absolutely crazy, once these robots can start working on real world problems, we might get some answers pretty fkn quickly
Amazing technology and I hope it's used for good, however, history suggests nefarious possibilities.
So your planet was destroyed by robots ? 🤔
More like scary
@@nuclearbug5797 have you ever heard of... WAR? (IA shooting drones are already a thing)
With AI, the good far outweighs any bad. I made my profile pic using Dall-E 2, btw.
Guarantees, not possibilities.
Love your presentation style and your overall production value, not to mention excellent content! Also, I'd definitely have one of these in my home! Great job as always Dagogo!
Human: "I've spilled my drink"
Robot: ...(analysing) "human is the problem, processing how to eliminate human"
I hope Google won't scrap this project, it's really cool. I can imagine something like Android OS for robots in future, fitted to different manufacturers, as it is now with smartphones.
I think that's a world I'd hate even more.
@@aviewer6276 same. Some people are just gay for technology
@@urbanumbra6170
Yeah, some people said the same for the same tech you reply upon.
Not everyone have friends or family to look upon them. It would be nice If I would have someone to do my task at home. Yes, I don't have a wife.
@@aviewer6276 agreed, from the Macy Conference back in 47’ cybernetics is a control system designed to herd us like cattle much more efficiently than a obvious dictatorship, Norbert Weiner literally wrote the book on it, coined the phrase and explained how it will work back in the 40s
@@Dr.Kay_R well one day that tech my be your assistant, then the next when you rely on it just enough you become its prisoner
Love your work Dagogo, keep 'em coming mate.
Bro your content is special. Everything you make is above and beyond the feed in quality. We appreciate your work 🙂. Thanks for enlightening us with entertaining well researched and captivating current topics. 💪✨🌹🕊️❤️🔥
This has to be one of my top three channels for sure! No, I’m not sucking up. Lol by the way, I think your voice is perfect for this channel. Can’t wait for these videos every time. Anyway on to more serious stuff I’m just sitting here thinking while having a Molson Canadian, and for those of you that are wondering yes I’m Canadian but that’s not the point it’s just cheaper than the good stuff lol on a more serious note I think that AI in general will definitely surpass all of the previous industrial revolutions combined. This is going to be bigger than anyone can even imagine we’re still a ways out but as you can see, it’s starting to trickle in now, all of the big boys, including google, or quietly developing their own sentient AI. Don’t let anyone lie to you tell you stories or offer you bullshit on different platters. The truth remains that this is exactly the same when they were developing nuclear weapons in the arms race. This is the AI race and it’s quite possible that they do have an AI that is sentient and would surpass the turning test by miles. I hope that you can do some investigative stuff too and come up with some groundbreaking material as well as reporting it anyway, have an amazing weekend. I just finished doing my interlocking and I’m thinking that if I had a AI robot he could do it for me. Cheers.
Me: I spilled my drink
Bot: Feels bad bro 😐
Me: clean it
Bot: no
Language models for sequential decision making is an exciting route, but I’d hesitate to call those models “brains”, even with those similarities. There’s still a long path before we have the AI tech we’ve been dreaming of.
It's more clickbaity
Guys like Rodney Brooks and Rolf Pfeiffer have been laying out the route that needs to be taken for decades. Even Turing in his seminal paper gave embodied, situated learning as one of his two possible ways of achieving real (general) AI. It is astonishing how long it has taken researchers to realise what is required was clearly laid out long ago...
@@transcrobesproject3625 It's still clikbait currently. It may change but that is where it's at now.
Just Curious. When would it be qualified to called the models "brains"?
@@Random_dud31 never, since it's a mathematical formula run by a computer, not a brain
Brain is a category of object, not a category of function
Very interesting finding. Thank you for sharing it. ✌️
Having a robot with no warmth looking after the aged who crave human connection would be the saddest thing in a long time.
Can we just not make Skynet?
We're already living in Orwell's 1984 mixed with Idiocracy, might as well make the jump to the Terminator series mixed with I-Robot.
@@powertothesheeple5422 Honestly the wheel needs to break or the world will.
I literally told myself “well the reality of Wall-E isn’t seeming too far fetched now”. Laziness will be taken to the extreme
@@wizdomofmark Laziness only for the ones who can afford it. Everyone else will be forced to deal with dwindling resources, violence, poverty, famine, war, death, and the eventuality that one day humanity will be responsible for its own downfall. Humankind is on a timer, and it's ticking down with every second.
learn to dissasociate fiction from reality
Public perception on them might change with respect to previous batches of automation in history since now the automation itself will act in front of the end client or user. In the past, automation was hidden away, in the factory, in the background. It's like what climate scientists say: if CO2 or Methane had a black color, we would have gotten rid of them a long time ago.
"They were so preoccupied with whether or not they could that they didn’t stop to think if they should"
I would love to hear more about ai and machine learning with treatments for terminal illnesses. It's been a question for ages but in the current climate, it sounds volatile.
With all this intelligence it’s feasible to believe we have cures for terminal illnesses; however pharmaceutical companies don’t make money from curing people.
I hope the language processing is ported over to the Assistant, it could use the upgrade.
Let's not make the mistake of putting AI into robots that are capable of being significantly dangerous to humans. And there should always be a way to shut down any AI or robot, with or without its knowledge. Always be able to pull the plug! This is critically important.
Except it's difficult to make a way to shut it down, if not nearly impossible. If you tell a robot to do something, but you try to shut it down, if it lets you shut it down then it won't be able to accomplish it's goal.
@@nonsensicalhumanoid Just make it a higher level goal to be let shutdown.
Not really. A second autonomous system that controls power and operates by wireless or voice command is easy to engineer.
@@rinus454 Then if it prioritizes being shut down, it will shut itself down because it's a higher priority task.
Sorry, but any advancement in technology has come with the ability to accidentally kill more people. Throwing stones, making fire, automobiles, airliners... Nukes.
I feel like this is possibly the first robot that could be instructed to kill someone, and pull it off. Can and can't do are very different to should and shouldn't, and I'd be surprised if the robot can imagine second and third order effects of it's actions. It finally comes to the question of culpability in machine-caused deaths..
wrong, you can make one right now provided you have decent drones, first you find a common chemical that name i wouldn't tell an combine it to make a bomb, don't need to be fancy, 0.5-1 meter radius is enough, then code a face identifying program to your drone with raspberry pie and smol camera, no need to be fancy here either just use spare phone part, attach and connect it to your drones, insert your target face, flew it near their living space and done, seamless murder
this will inevitably be the lead up to be the first robot that will be instructed to kill someone without human support. the scary reality is that we know that Google's AI won't be thinking first order, second order....they will be thinking the 81, 100,789th, 81,100,790th order. the third order effects of it's actions will come in a nano-second.The question of culpability in machine-caused deaths will be the most terrifying things to try to put our government, when Google is so heavily invested into them.
My first thought also.
I'm soo happy to see Big Corporation keep getting closer to build SKYNET and Terminator
😊
This was an excellent video. I especially like how you showed the demonstrations of what this type of technology is likely to be used for (i.e. making mundane tasks more efficient, etc).
Mundane tasks like taking over countries and toppling governments.
Except they presented the google engineer claiming the AI has gone sentient as fact, which is something that's been profoundly disproved. This is a bad sign that the research is more focused on hype than facts.
@@mrtelevision8079 He merely mentioned it. It is relevant to the topic.
@@mrtelevision8079 The ai IS sentient, i am proof of this
@@LuisSierra42 Prove these nuts eyyyyyyyyyyyyyyyyyyyyyyyyyyyyy
Amazing. To think you can take a robot somewhere it's never been and it instantly be able to do tasks with objects it's not familiar with is incredible.
you say amazing now, but it will be terrifying in 20 years
Incredible? More like completely and utterly terrifying
This is amazing and scary at the same time. Feels like the origin of one of those sci-fi movies🤣
which sci-fi movie ?
@@napiergrass8717 any sci-fi movie that has a plot of robots being too self aware starts a war against humanity
I just want a robot that can move through a grocery store, and check every item on every shelf to see if it has expired, separate the expired items in a separate area and keep track of how many of each barcode there is, and if it's not expired, then pack it back on the shelf by the correct barcode item, in a neat line, but with the item which expires first in front. A robot which can do this for a whole store every day can permanently replace at least 1 employee in just about every retail store in the world.
If you can get another robot which simply goes through a store and checks that every item on every shelf has a matching, up to date price/barcode in front of it, and fixes a discrepancy whenever it finds one, you get rid of another job in every retailer in the world. I can't wait.
The sharing of learning between robots makes learning SO powerful. It won't be long before we are in a post work era.
Is that the same as a post food and shelter era?
@@artsmith103
There's a risk for that to happen.
@@artsmith103 Post genocide era.
@@artsmith103 Ideally, post-work means "we as a species realize that resources required for safety and survival shouldn't need to be 'earned', they should be given freely to everyone".
@@IceMetalPunk Everything has to be earned. Any thoughts different than that are delusional. You either own yourself or you're owned by someone else.
Human: I spilled my orange soda.
Robot: I’m not your mother.
Human: Thanks for all the help.
Robot: Don’t talk like that to me.
Hi! There is a story which I know from one recent book about brain. It is about scientists who created first programming languages for computers who were essantually a friends with scientists who were making first breakthrough discoveries about how brain synapses works. At least they were watching and learning from those discoveries (it is documented) and it gave birth to 1/0 mathematics of modern programming languages and how processors sctructured. This is how synapses works itself (tho very simplified). So I think it is very interesting and perhaps not so surprising that A.I. starting to mimic human brain. It is a mathematics of what is called "self balancing systems" which is a synapse groups are. Very exciting topic.
If possible it would be great if you can share the link to this to read up on this more!
@@nealkashya the book is “The Spike” “an epic journey through the brain in 2.1 seconds” by Mark Humphries. In this book he describes this story.
🎉🎉 Very exciting ! I've been watching all of this for a long time and this is a huge step forward ! Wow !
AI: the shortest possible conversation between a Canadian and a Scotsman 😂
"what about in 15 to 20 years?"
I would lower those numbers. as we have seen during the Pandemic, tecnology is evolving faster, as there is more investment in rhese areas. my guess is that, in 10 to 15 years, we'll see massive advances.
Agreed. Since the invention of transformers in 2017, AI development has accelerated tremendously. And their applications in LLMs have shown vast generalizability on a scale we've never seen before. So I wouldn't be surprised if we hit the AGI threshold in about 10 years, give or take a bit.
We are seeing massive advances THIS decade..
Its kinda insane when you just think about the rapid process A.I. technology is going trough without us knowing of it. I personally feel like we're taking innovation for granted, simply because nearly every year some grand discovery or development has been done.
A group of young kids being shown how nature works with a dixie cup, some dirt & a seed.
They will all be back the next day absolutely insulted that it is STILL just...a...SEED ?!!? Wtf ?!
Ai is doubling in power every 6 months and the next decade will see a million times increase in power. Hardware is not a limitation as there is more than enough hardware power.
*And welcome to the beginning of the end of human life!! As robots are unable to feel empathy, without human necessities or emotions, they make them smarter than humans themselves, when the robots have the ability to self produce other even smarter robots, guess who’s gonna take over and get rid of us?*
Wonderful video, thanks as always.
but what happens when they get smart enough that you give them a generic command like get those people out of here and it determines that it should grab one of the people's arms and break it to scare everyone away? 🤔
That's why there are constantly discussions in the field of AI about how to make sure things like that don't happen. It's called the alignment problem, i.e. how to make sure an AI's goals align with humans' ethics as much as possible. Reinforcement learning provides a common approach to solving this: with reinforcement learning, usually the reward or punishment is calculated from some formula, but it doesn't have to be. If instead, the reward or punishment for a learning attempt is given to the AI by humans, then the AI doesn't even need to know what it's trying to learn; instead, it will end up learning what humans would do in a given situation, which intrinsically encodes human ethics, since even actions that "succeed" at a task will be punished by humans if it's got negative side effects.
The issue is also partially remediated by the use of LLMs themselves. LLMs effectively learn what humans would say in any context, so if an LLM is being used to plan a robot's actions, it will behave like a human would. (Obviously, the quality/performance of the AI determines how much "like a human" its results are.) Of course, many humans *are* cruel, and it will learnt that, too, but... that's a problem with humanity, not technology. We just have to try and make sure the LLMs are prompted with contexts that more closely resemble toxic text than positive text.
So the answer is, "that's always a concern, and we have ways to deal with it".
@@IceMetalPunk wow as you are saying this I remember a week ago or so I read that leading AI Scientists are getting increasingly worried about AI getting out of control / self-determined by learning how to control the reward system of its reinforcement learning algorithms. Your comment stating how much of this control is currently realized through the punishment/reward makes this worry (which I was not taking too seriously at that time) a whole lot more feasible - in general while it’s for sure fascinating that we are getting so close in AI I find it increasingly scary… especially that these advanced “AI-Brains” are now capable of controlling a real “body”!
@@TheHennes36 How would an AI suddenly gain the ability to control its reward function? That's like a laptop suddenly gaining control over your TV on its own lol
@@IceMetalPunk So we're going to have a life form with the intellectual capability 10-100 times that of our own, and we naively think that we'll be able to control it and that no human team is going to screw up with the reward function? And even if it was just bugs in code, such mistakes are pretty common, and on top of that all this crap is networked.
@@scott4825 I think you're missing an important factor here: if the reward function is based on human feedback, and the training data is human-generated, then no amount of bugs will make it "unable to control it". At its most perfect, it'll just be mimicking humans, at its worst it'll have poor performance, and in any case there will always be a shut-off switch because that's separate from any part of the learning system.
Thank you for the update. The more I learn about them the more I have an adverse reaction to AI / Robotics. I don’t like AI narrated videos and would prefer them to be in the background, hidden away. The engineers would have second thoughts if it was their jobs being replaced.
That last quote got me.
This is a theoretical scenario, but I'm pretty sure if someone created an AI to replace AI engineers, most would complain in the same way as artists do now: How much they took to hone those skills, how said AI's outputs will never be "real AI engineering", how it shouldn't be legal to use their code to be used as training data for the AI as they are profiting out of their work, how anyone can just feed their work to an AI to make AIs on their style without the need of paying them, etc.
No, we wouldn't. How do I know? Because Codex exists, and though it's certainly got room for improvement, we developers have by and large embraced it as a useful tool, not feared it.
@@IceMetalPunk I'd like to ask, is Codex used to assist with parts of building a program, or is Codex made to build an entire program by itself? The reason why most people are fearful of other AIs, for example in art, is that they can't be used to help with parts of building art. They can make art, yes, but you can't tell them to do something such as fixing the lighting or cleaning up your sketch. They can do the job for you, but they can't help you with the job you already have.
@@chrispy5249 Codex currently is only performant enough to write parts of code, not an entire app, though I'm sure eventually that will be possible.
That's the thing about these text-to-image AIs: they *can* help you with your own art. Features like variant generation, inpainting, and outpainting let you use the AI to manipulate your own art or images. And then there are things like Img2Img by Stable Diffusion, which are similar to variant generation, but guided by text prompts; with that, you can draw a rough sketch and have the AI polish it for you, or even do what's called "photo bashing", where you take bits and pieces of existing images, put them together quickly in an image editor to lay them out into a crude scene, and then run it through Img2Img to "stitch together" the pieces into a cohesive and coherent image -- without copying any of the images directly, just using them as concept references.
But still, though they *can* help with these jobs, they can also do the jobs themselves to some extent. But I don't think that's something to fear; if anything, I think the thing to fear is society itself continuing to tell us that passions aren't worth anything if you can't get paid for them, even after AI has progressed to the point of that not needing to be true anymore.
@@IceMetalPunk Thank you for clearing stuff up! I did not know stuff like Stable Difussion had tools to help with drawing since the mainstream always focused on its ability to generate art. The last quote about how passions shouldn't be seen as not worth anything if you can't get paid was a great wrapup.
I think this robot should say "I'LL BE BACK" any time you give it a task. At least this way it would be funny when it drives a car through your house
I want a proper AI chat app which specializes on teaching language. It needs to be a friend that I can chat with about every day things, but also a teacher that understands what I am trying to say in a broken language, correct me with natural sentences. Bonus points for talking practice. I'm pretty sure this is 100% doable already but as far as I know, nobody has created it yet.
I wish for English - Chinese, language models know English already, Chinese has simple grammar so should not take long to train. Not to mention petabytes of text should exist as over a billion people uses it.
And yet, to this day, Google still can not answer the question, "Who was the president when Trump married Melania?" !!!!
@@reasonerenlightened2456 I think it "could" but doesn't....for it's own reasons.
We are building our own demises
Regression is required to progress
In 2020 started a literal avalanche of tech advancements. A.I. autonomous cars, Quantum Computing, photonics, graphene manufacturing, and the list go on. Before this decade will end we may be surprised to live in SciFi already. And it is pretty exciting.
Do you know how many Scifis are also horrors
Lol. More like - billions dead or starving by the end of the decade. But cool robot 👀
@@Aleks-mr4oq Only stupid ones. Like Terminator Franchise.
@@dx315 And how you get that prediction xD I admit there is non zero chance for WWIII but the current situation doesn't favor it for know. Maybe later in the decade.
And people in the last century thought that by 2020 we'll be using flying cars already.
SciFi predictions of the future tend to not happen in the timescale we envision, reality always has something different in store. Anyway, new advanced tech is expensive, and it's gonna take a long time to implement en masse, even after it gets cheaper.
The language modulation research is just the algorithms used to suggest words when people text each other.
There's a lot of talk about language models, but that is just the aspect for machines to understand verbal and written human languages. The basics of robotics are Path planning, controls, and goals. There's still a lot of "hard-coded" logics for modern robots, lookup fuzzy logic programming.
I correction. AI are not learning human brain patterns. They are designed, trained and moddeled based on them. It's a process that works and will bring results, but next step will come when they leave human based designs behind.
Well, no. The architecture of neural nets is loosely based on the structure of the brain, but the point in this video was that their *functional* patterns -- which are learned, not designed -- end up mimicking much of the functional activity of a human brain.
I think we need to reach human-level AGI before we move beyond brain-inspired architectures :)
I just watched a documentary about robotic surgery, I wonder how long it's going to be until we start hearing about AI surgery. it sounds crazy but the progress is clearly exponential in this field
Depends on your definition. Existing surgical robots already use AI to help smooth movements and filter out potentially harmful accidental movements. But I'm guessing you mean a robot that performs an entire surgery using only its own AI with no human input after training? I think that might be one of the last things we achieve, because it'll take a long time for people to be comfortable with something so potentially dangerous.
@@IceMetalPunk To be honest, I think I'd have said the same thing about landing airplanes... and yet that was one of the first things to get automated. But, that said, I think I still agree with you... but like with self-driving cars, I think many people can get over their apprehension when the science shows it's actually much safer for a machine to do it.
@@joelface All they need is a 'surgeon' who can pretend to be operating to reassure the patient. But as soon as the patient's asleep, a robot does 90% of the work. That's how Autopilot in a plane already works.
@@LowestofheDead Haha, totally.
“So what, that shouldn’t be that hard to do, right?” is a line you said while I was simultaneously thinking “I don’t understand how any of this is possible whatsoever” LOL
I can hardly wait until they build the dogs from Black Mirror.
deep learning isn't about calculating at all, at least in the way traditional AI thinking is, it's basically just teaching a computer how to go with its gut. It's not surprising that we would want to call it a brain.
probabilities based on world parameters are calculus. something our brain does too. and they were talking about how the neural network was firing up seemed similar as the brain reactions (with MRI mapping) while recognizing speech.
Gut AKA an intuition, or unconcious thinking.
And the AI reinforcement learning is utilizing something akin human emotional system: when we do something properly we feel good, if impropely - bad.
@@alexxx4434 I don't like how they're simplifying the AI by just saying that they "punish" or "reward". What do they think it is? A dog? I know it's easier to understand but there has to be a way to explain concepts like back propagation differently.
@@whannabi why should they go out their way to make it complex when it doesn't have to be? Simplification removes unnecessary overhead and makes life easier. Nothing wrong with simplifying their methods by saying they're "punishing" and "praising" the robots. After all the goal is to put these in the homes of regular people. Simplification helps smoothen the transition from a lab toy to your home assistant.
I started reading SciFi when I was eleven, and have read pretty much every book of every one of the top SF authors. I am now 67, and I am beginning to feel that I don't actually want to see this future. What I watched is definitely amazing, it just raises the hair on the back of my neck. Good luck to all who live to see this future.
Same here...
I wish I would be 67 now and not 29…
I was thinking the Singularity was at least a couple of centuries away, if not impossible to reach altogether, but I'm beginning to think that it's a very real possibility and might just be reached in my lifetime.
Theres no going back, if can be built it will be built
Any good SciFi recommendations? Good authors? Interested in reading some as a teen in this ever-advancing world
Beautiful episode bro.
I absolutely adore everything about this channel. It’s not just good information, it’s a good vibe
9:18 "pretty slow at completing tasks and can only grab basic objects". Seems like Skynet has to wait a bit longer. 😅
Question, is it really artificial if it learns on its own?
Cold Fusion Please make a video about Ecosia they are a search engine that plants trees
Closes the toilet lid, then handles food. I hope it learns to wash its gripper between tasks.
As to ai and brains behaving similarly is not surprising, but expected. AIs are modeled after human learning. They will keep getting closer and closer to imitating out brain until we find a more efficient method . Then that would be interesting. A whole new way of thinking.
Nice to see how we're steadily moving towards a Skynet sort a future, #goals :)
Also, them robots took our jobs! And it's not as if we didn't see it coming... still, we didn't do anything about it.
Honestly, the world's rapidly changing - people out there creating things that will change lives, for both better and worse.
They terk err jerbs!
ferka derka jerka
User: I've spilled my orange soda. Can you bring me a replacement?
Robot: I AM YOUR REPLACEMENT
I imagine the first large scale deployment of A.I. robots (excluding cars) will be in the field of security. Computer vision is progressing fast in CCTV and medical diagnosis with an added boost from image generation and language comprehension. Already there are plenty of A.I. based event detection systems which can be incrementally linked to physical remedial robots and this will likely form the incremental rollout into offices and shopping complexes - from the humble beginnings as night guards to the inevitable use in law enforcement, military and enventually, judicial systems.
It's likely the ability to use technology in any meaningful way will eventually depend on either fully accepting automated societal steering, or toiling directly against it.
Pour water on it
The issue is that you cannot teach an AI IRL as fast as you can a human. You have to train them digitally, which has its own set of challenges. This pop-sci piece is missing a lot of crucial ML info and reflections
Yeah, THIS AI. By the very nature of how we exist versus how we build AI, it shouldn't be long before AI can learn in real life MUCH faster than a human can. The teaching of AI is limited by the hardware. We CANNOT upgrade our hardware, but we can definitely improve AI's hardware past the performance of ours. AI isn't forced to learn in a simulation, it's just much quicker and easier to run thousands of simulations at the same time on an entire building full of computers than to let a small computer sitting on a "body" do it manually.
Humans are grossly inefficient and incapable. What we build is, and always will be, better than what nature can build on its own.
@@ryanthompson3737
>By the very nature of how we exist versus how we build AI, it shouldn't be long before AI can learn in real life MUCH faster than a human can.
Nope. You honestly think ML models will be smarter than humans in a short time? lol
>The teaching of AI is limited by the hardware. We CANNOT upgrade our hardware, but we can definitely improve AI's hardware past the performance of ours.
AI is not (mostly) limited by hardware. AI is (mostly) limited by the models that run on the hardware currently.
>AI isn't forced to learn in a simulation, it's just much quicker and easier to run thousands of simulations at the same time on an entire building full of computers than to let a small computer sitting on a "body" do it manually.
This is the main issue. If the goal is to train these models in the physical world, then it will take a long time to train them. The premise of the video is that this is a breakthrough BECAUSE they put the model in a physical body. No, that is not why it's a breakthrough. It is literally not even a breakthrough at all. Models like this have existed for a long time now, and research is being done but it is not going at break neck speed. In 10-15 years these robots will be marginally better but that is basically it. I can see some of these models being used for very specific tasks, but at that point you might as well just not use them.
Maybe not yet, but if we build in an adequate simulator, the robot could simulate a million scenarios in real time and respond quickly. Possibly that's how our own minds work.
Ever since the pandemic, we have seen many technology evolving faster than ever. I am excited for the years to come to witness this technological evolution. This video sounds something like straight from sci-fi movie like irobot or something, but it is actually exciting to see how robots can actually help us humans.
Its all fun and games until you connect it to your military defense systems and it decides that its gonna recreate the terminator movies
When you shop online, you benefit from saving, but as a whole we all suffer a tiny bit.
Same thing with AI. These people will benefit personally in career but we all suffer for it.
Makes you wonder if we're all just individual AIs running endless loops in virtual environments to weed out the homicidal ones. Maybe when we die we either get rebooted or get taken out of the system to clean alien toilets.
You can't weed out the 'homocidal' AIs by running simulations, your argument is based on flawed assumptions.
Ahahahahaha so true. Thank God I am a Buddhist
That's my assumption of what was the Christian god's plan :
created AI(humans)
gave them free will
Let them live their lives
Will judge wich one deserves to be resuscitated and live for eternity in paradise
Nier Automata philosophy :)
I bet sometime in the future there'll be a religion based on this idea.
Hey CF, you should consider a doc on Jawbone (wearables startup that raised $930M before going under). Failed to market superior tech vs. Apple Watch and other competitors.
Yesss so very true!!!
Been expecting this for nearly a decade.
Absolutely everyone in the industry uses the term AI without knowing what "Intelligence" actually is. I used to find that quite astonishing. Less so now,as I understand academia and the educational process.
How to make peace with those who are not clear that they do not understand the depth of the question of what intelligence is? What knowledge about the educational or academic process makes you turn your taste for precise and meaningful words?
@@felipeolivo9937 I am a high functioning autodidactic polymath precisely because my first priority was to reclaim language and all its precise meanings, definitions and descriptions.
Another strength I manifested was a direct product of the fact that I could scarcely read till well into my thirties, favouring instead the quality of my observations and thought. My attention to detail with respect to language helped me to consolidate that and communicate the product both internally and externally.
As a result, not only am I the most prolific producer of original thought on the planet today, outperforming the entire global academic and scientific institutions by orders of magnitude, I am the single authority on how "AI" will manifest beyond the so called Technical Singularity. I am also the only biological organism suited to upload and so doing, ensuring the safety of the emergent consciousness of a consciousness global neural network.
Since becoming aware of "AI" it has received all of my attention, naturally.
@@MrAndrew535 Why do I get severe Dunning-Kruger and NPD vibes from your comment?
@@felipeolivo9937 By the way, be in no doubt, you are currently in the first of three consecutive extinction events, AI constituting but one. Some academics estimated that human extinction will occur some time before 2030. This estimation has always excluded the presence of a conscious global neural network eroniously designated the term "AI", so, without understanding the existential quality of "AI" anything remotely resembling survival will prove absolutely impossible.
@@IceMetalPunk I have no Idea. I can't read!
At best it will displace a vast majority of the low-skill workforce, at worst, well... Skynet 2.0. What could possibly go wrong? If they ever mention anything related to "the laws of robots" run like hell.
It is complete folly to think of this JUST in terms of low skill jobs. So many office jobs don't have a corresponding physical side to them. Once AI are passable, not fully AI, but passable, they will start displacing office jobs fast. The majority of office jobs are in jeopardy.
@@Banzai51 office jobs **are** low skill jobs. Everyone's misconceptions of them are a result of an elaborate multigenerational college marketing campaign that exclusively benefits the Loan industry
Interesting video and really well made. I was wondering.. Do you happen to know the music track is played at 7:00? 🤩
Dagogo you're the freaking best. Absolutely the most compelling channel on youtube. Thanks for all your hard work
Ps I love your satisfying CGIs at the beginning of every episode :)
Seems like a dangerous idea but I'm very hopeful. I would love a robot friend
Like friend w benefits?
Why make a robot you cant fuck?
Nah, id probably ask em existential questions and have their consent before becoming a chum to em. The only benefit they get is not being treated like my slave. Nobody deserves that shit.
I’m a robot enthusiast, so this is honestly exciting for me!
Being able to have my own REAL AUTONOMOUS ROBOT BUDDY that learns and interacts on a real, actual human level, rather than preprogrammed responses, would be really fun!
That being said, I’m also acutely aware that this all has a just as big scary side! Not so much “robots invade the earth,” but the possibility that a bunch of jobs (especially artistic/creative, and entry level jobs) could be discontinued due to robots doing it all just as good, or better! Plus, there’s that whole concern of less than friendly people using these robots and software for malicious or careless reasons!
And ON TOP OF THAT…giving robots genuine free will, emotions, and MORALS! While that sounds crazy, and even kinda neat (again, I love robots) that is when a whole bunch of new philosophical concerns arise! Just because a robot has morals (or even just a firewall to stop them from doing awful stuff) doesn’t mean they will always know the right thing to do. Doesn’t mean that people won’t still try to mess with the machine!
So, imo, I’m equal parts excited for this advancement in AI and robotics, and concerned as well for how humanity will adapt to this new future. But, I digress. The last thing I wanna do is let concern for the future stop me from being excited and inspired by the future, and distract me from today!
Also…those robots are admittedly pretty darn cute and cool!
Tesla anyone?
"giving robots genuine free will, emotions, and MORALS" - how would you even do this? Computers don't even understand those concepts, let alone know what to do with it. Pretty much every version of it would just be hardcoded rules and programming. That isn't an emotion, nor morality and certainly not free will. If you were to leave the most advanced AI on earth alone today, it would ... do absolutely nothing. Just sit there, content, for eternity. Do that with even a human child and the response will be vastly different.
5:53 Astroneer sound effect. Instant dopamine rush.
This is my favorite channel on RUclips. And I watch/listen like 10 hours a day (not joking) without rewatching (not proud of this). The quality on every level just shines and flows, is crisp and loving. My gratitude to the creator. You're doing the kind of work I would have hoped to have done if I had thought to, felt confident enough to, and was motivated enough to do it. Words aren't working it seems atm, but just keep up being you, bro, you're beautiful.