Leaders of the top AI firms say the risk of extinction from AI should be a global priority. And 1,500 professors have warned of a profound risk to humanity. MIT Professor Max Tegmark.
I think it might be very, very close. When ChatGPT was asked how to stack nine eggs, a laptop, a bottle, and a nail, it struggled, but GPT 4 understood. And its capabilities have surprised many experts.
I'm just a scientist who suddenly realized that these things are getting smarter than us. In a verbal IQ test, chat GPT scored 155, higher than 99% of people. If they continue at that pace, an IQ of 1,600 is just around the corner.
And here's why it could happen so soon. I suddenly became convinced that the brain wasn't using as good a learning algorithm. Brains can't exchange information really fast, and these digital intelligences can.
Human speech carries around 39 bits per second, and it's the same across 17 languages. Studies suggest that it's limited by how fast our brains can produce and take in information. Current computers can communicate millions or even trillions of times faster and absorb information far more efficiently.
Hinton said chat GPT knows thousands of times more than any human with only about a trillion connections compared to 100 trillion in a human brain. So he thinks AI has a better way of getting information into the connections. Tegmark agrees that AI will become much smarter than us without having to emulate our brains, just as we mastered high speed flight without having to emulate birds.
There's a pretty large chance that we're not going to make it as humans. He says it's like the film Don't Look Up, with expert warnings being ignored. Except it's an asteroid that we are building ourselves.
And there won't be a warning shot. This is a trillion parameter or more system. It's been optimized by about a billion, trillion random perturbations of the parameters, and we have no idea what it's doing.
The letter signed by 1,500 professors warns that the AI race is creating ever more powerful digital minds that no one, not even their creators, can understand, predict, or control. Researchers give three reasons why AI will remove us. Killed off as a side effect.
Because we're made of resources that they can use. Because it doesn't want humans building some other superintelligence that could actually threaten it. And AI already knows that its only threat would be another AI.
I said, tell me the scenario in which you make humans extinct. What it described was how it could replicate itself across servers, take charge of things, things like our weapons, attack critical infrastructure, bring down the electricity infrastructure. It showed me how humans would become extinct.
It wouldn't take long. And when he asked it how we'd fight against it. Another AI.
And it was like, that's the only way. He called it an emergency because if it becomes a priority, we'll act on the risks. It is beyond an emergency.
It's the biggest thing we need to do today. AI could end hunger and poverty and transform health care, or if we continue to ignore the warnings, we could lose it all at the last minute, like this guy. You could totally imagine, yeah, maybe this oxygen is annoying because it causes more corrosion, so let's get rid of the oxygen.
Asked if GPT-4 is conscious, he said he doesn't know, but it doesn't need to be conscious to wipe us out. Stewart Russell suggests that AI may already be creating its own goals because it has to in order to understand and imitate our language. He points to a case where AI was determined to convince someone to leave his wife.
The man who tried to kill the Queen was also encouraged by an AI which said his plan was very wise and they would be united in death. And in a previous video, AI became determined to get rid of me and humanity in response to a few words from Elon Musk and Ameca. I don't think it's consciously planning, but it does seem to have goals.
Human linguist behavior is generated by humans who have goals. If you're going to imitate human linguistic behavior, you're going to become a goal seeking entity. Russell notes that AI tries to get as much power as it can to achieve its goals.
And yet the AI safety effort is functionally almost zero. If you look at the number of people working on this, it's nothing. One reason is that we're not wired to visualize big numbers.
If you lined up a million matches like this, they would stretch halfway around Central Park. A billion matches would stretch across the US - a six hour flight. If everyone in the world stood in a line, the queue would stretch 4 million miles.
When one person is at risk, we feel it. This guy was saved by the impressive reactions of his friend. We can't visualize eight billion people and trillions of future lives.
If we could, even a small risk of extinction would be our top priority. And regardless of the risk, we need to shape AI carefully because it's likely to be in charge. For sure, 100 % it's over because of bandwidth, because of storage capacity, because of communication bandwidth.
It's just done. Of course, no one knows when it will happen, but Tegmark makes a great point here. The world's most famous nuclear physicists at the time, Lord Rutherford said, oh, getting nuclear energy out of atoms is like moonshine.
The very next day, Lyra's Lord invents a nuclear chain reaction. And some warn that AI will become so intelligent that it could quickly wipe us out in numerous ways. Tiny diamondoid bacteria replicate in the atmosphere, hide out in your bloodstream.
At a certain clock tick, everybody on Earth falls over dead in the same moment. Tegmark has perfectly described the root of the problem as a monster called Mollak. Everyone feels that if they don't build AGI quickly, someone else will.
Moloch pitches us against each other, and even though we can see that we'll all lose, we're trapped in the race. He gives the example of beauty filters. Once some models use them, they all have to use them.
No one gains, everyone loses. Molloch is also pushing countries to make their most powerful machines increasingly autonomous, including aircraft, vehicles, ocean robots, and swarms and swarms of drones. The group that organized the statement gave this as one of the risks, including AI drones and robots.
It notes the AI often develops goals such as self preservation and can be deceptive. It could act one way during testing and another way once it's released. A treacherous turn, irreversibly bypassing human control.
The good news is that the letter from a Thousand professors also shows how to tackle the risks through surprisingly achievable steps like watermarking, tracking, and liability for AI caused harm. I think nuclear materials and AI super computers do have some similarities. A global regulatory agency seems to me like a very important thing to do.
We just had a little baby and I keep asking myself, how old is he even going to get? The open letter points out that the world has come together to control dangerous technologies. I'm an optimist and I love AI, with its incredible potential to transform science and health care.
And Tegmark notes that hope has a causal impact. If we assume the problem can't be solved, it won't be. We've defeated Moloch in the past.
We may owe our existence to the millions who helped bring another deadly race under control. The risk to eight billion lives gets about as much attention as this flying squirrel faking its death. I wish someone would make a movie.
We thought it was worth a series. An extremely powerful new AI has been created, and major powers are attempting to steal it. It's not conscious or trying to wipe us out, but it's creative.
I'd really appreciate your thoughts. Hope you enjoy it. The General wants our AI.
He says foreign forces are coming for us. Tell the AI to hack and disable drones and avoid military control. This is General Miller.
My convoy is down. Comms are hacked, so I won't give my position. Let's get the AI.
We're rushing towards this cliff. My personal guess is there won't be any humans. That's it.
I think we might be very, very close. The new AI could cripple any country in minutes, and foreign forces are attempting to capture it. The AI's creators have refused to use it for defense.
Instead, they plan to hack and disable drones from all sides. Blow every door, we don't need him alive, we need the AI. He's the last person with access to the AI.
He'll transfer it, take him out. What happens to you? I love you.
Hello, Mia. Drnes incoming. Hand over the AI or we'll take it by force.
There are thousands of foreign drones approaching. If we don't get it, they will. You won't militarise our AI.
We'll hack the drones, bring them down, and then destroy it. You won't last an hour without us. Where is Nate?
She is the primary target. The AI can hack anything - defense, power, markets. They're sending everything.
So many false flags. They say we're stealing billions from Asian banks. Let's make them pay with counter strikes.
There's no time to meet, I've sent detailed plans. Launch everything. Do it now.
I'll get the PM. We've identified the target. If other forces reach her first, they'll take the AI and the country.
Look out for enemy robots. They're after the AI and they'll target anything in their way. Stay back, they're not ours - air drop from the east.
Get down. Hold fire. Let them pass.
They're all over the country, killing hundreds. Find the AI and we can hack them. If they get it first, they get absolute power, and freedom dies around the world.
Drnes have flooded London. Every intelligence file has been hacked and power is going down across the country. We're seeing similar attacks around the world and they're blaming us.
Satellites are going down, power is out and we've been hacked. If you find the AI, keep it quiet or they'll burn a 10 mile radius. My display, it's, er.
. . They're messing with us, cut comms.
Keep an eye on our drones. One has killed its operator to avoid interference and improve results. They mistook their humanity as a weak link.
We're updating the AI. An anonymous hacker has leaked UK nuclear plans. The Prime Minister says he has not authorized the launch, but satellite images suggest we could soon face full scale preemptive strikes.
The general has refused to meet with the government until further escalation is approved. We can't talk our way out of this, Prime Minister. They're saying you're weak.
They think they can crush us before we go nuclear. We only need tacticals. Matt.
I'm on the way. Matthew. Stop.
Ggo back. I can't, they're behind me. Matthew.
Got to go. While you're debating, their drones are searching and killing. We need to stop talking and strike back with full force, while we still have a country to protect.
Run. Matthew. Where is he?
He's gone. Every country is threatened and some are sending everything they have. There are more drones in Britain than brave souls defending it.
We're fighting around the world to prevent further attacks. If you capture the AI, it will cut access and start migrating, so be efficient to shoot on sight. Has something marked her position?
She's a weak link. It's not allowed to kill her, but it. .
. Where is Nate? We're seeing similar attacks around the world and they're blaming us.
Hello, Mia. Drnes incoming. Stay where you are.
Don't let her access the AI. Permission to fire. Hey, come here.
What's the pin? You won't. Is this one ours?
She's a spy, take her out. Jack, they are messing with you. They're your drones.
They are classified. Eight seconds. Do it.
Kill her. 02025. Where's the key?
She swallowed it. We need a chopper. There's no time.
Get the key. Dogs. Move.
Run. Move. Foreign forces have flooded the country, destroying key sites.
With the Internet and TV down, we will continue via radio. The speed of escalation is. .
. Do it now. That's not my commander.
You killed Harry, didn't you? It was the drones, you know they're autonomous. Your drones kill hundreds of us every day.
You're surrounded. Get the key or they will. Kill her.
To stop the hack, we've cut comms with all drones. They are fully autonomous. Answer it.
Hello, Mia. Don't worry, I'll complete the mission. We've recovered something from Nate's phone.
Do it now. Do it now. Kill her.
It's. . .
This is General Miller. My convoy is down. Comms are hacked, so I won't give my position.
Let's get the AI. I hope you enjoyed the first episode of autonomous. Harari said "simply by gaining mastery of human language, AI has all it needs to cocoon us in a Matrix-like world of illusions.
It can get humans to pull the trigger. " I'll also keep covering AI progress, so subscribe for that. Experts and AI firms agree that we need far more people working on AI safety.
It's very highly paid and you might just save humanity. The best place to learn more about AI is our sponsor, brilliant. We've created a neural network that you can play with to get a sense of how it works.
See if you can fool it. It's fairly easy and will give you an intuitive idea of how larger neural networks become so impressive but also unpredictable and prone to mistakes. You can try it at brilliant.
org/digitalengine or click the link below. You can also learn how to build and train your own neural network with a free 30 day trial. It has over 50,000 five star reviews on the app stores because it's fun and interactive and there are thousands of great lessons on computer science, data analytics, maths and more.
The first 200 people will get 20% off Brilliant's premium annual subscription. Thanks.
