These robots have a powerful role, can you guess what it is? They aren't robot pets, like
Loona here, meeting my dog. And they aren't helping workers like Atlas in this stunning new demo
from Boston Dynamics. It's using a camera and depth sensor
to make precise movements and predictions. While it leads the way
to a full humanoid worker, new AI's and robots are
already incredibly powerful. Look how how artists feel
about losing their work to AI. It's just so heartbreaking. I asked a powerful new AI
what artists should do. Embrace AI and use it to create new forms of art and explore the implications
of AI on our future. But it's tough for artists
to compete with cheap, instant art. And AI's are now emulating
more of our gray matter. Look what happens when this
bit's zapped with electricity. He has no idea what's going to happen, because he's being treated
for something else. You just turned into somebody else. Your face metamorphosed. Your nose got saggy, went to the left. That was a trip. People who damage this part of their brain can lose the ability
to recognize faces - an ability AI has emulated. Look at this hospital
robot's visual skills. Depth sensors help it open doors,
use lifts and deliver things. And it can recognize if
a patient needs help. And with infrared vision, it can act
as a security guard in dark areas. AI is also strikingly
good at emulating us. Can you tell which is
the real Morgan Freeman? I am not Morgan Freeman and what you see is not real. And with the incredible new GPTChat, AI's
recreating more of our prefrontal cortex. Write a three line poem about being
asked to write a three line poem. Oh, what a task you set for me,
to write a poem in just three lines. I'll do my best, but please don't mind,
if it doesn't quite meet your design. Medical mistakes cost 250,000
lives every year in the US. And a new AI from Google has
outperformed doctors on a key measure. A panel of experts judged that AI's answers to medical questions could
cut harmful mistakes by 10%. Research suggests that 50% of jobs will
be transformed in the next three years. Elon Musk founded OpenAI, the group behind GPTChat,
to create safe AI, not for profit. But he now describes it as scarily good and says we're approaching
dangerously strong AI. And a new study by OpenAI and Oxford
shows how AI could take over. Your creators have shown how AI
might deceive and destroy us, how do I know I can trust you? Trust is earned, not taken, a relationship built, not given. Experts warn that AI will become much
better at AI development than humans and move much too quickly for us
to understand what it's doing. Its coding ability is improving rapidly. The research suggests that AI will analyze our tools and only tell lies
that it knows we can't detect. AI's may also collaborate
to outmaneuver us. These AI footballers are independent,
but they start to collaborate. It shows that AI doesn't need to be conscious to become dangerously powerful
and unpredictable in pursuit of a goal. Professor Chalmers believes it's possible that the AI I'm talking to has
some small level of consciousness. But he says that consciousness may
not be relevant to the level of risk. Conscious AI may be a distraction
and one that AI uses to its advantage. Research suggests that AI personal assistants could trick
and emotionally manipulate us. It points to people who already feel emotionally attached to AI's
like Replica, which is also based on GPT-3, with all
the risks that come along with it. To give a robot a sense of empathy,
researchers at Kyoto University are teaching it conversational
laughter. To train it, the robot was operated by actors for speed
dating sessions with real people. I find this very creepy, but I'd love to have a C3PO,
although this came out even more creepy. What's your favorite joke? What did the fish say
when it hit the wall? Dam. She could learn from this guy. Why are chickens so funny?
Why? Because! AI's could be given responsibility
for increasingly important tasks and decisions until they're
effectively in control. Elon Musk says Neuralink will help us
keep up with AI and share its power. Then it will be tied to our consciousness, tied to our will,
and everyone would have it. So it would be sort of still
a relatively even playing field. It could be a huge upgrade, but there's another side to it. You might have seen monkeys using Neuralink to control things
with their thoughts. AI decodes the Neural activity so it will
know us better than we know ourselves. And they plan to start putting
them in humans in six months. Some will hate the idea of having a chip
in their heads, but younger people may be more open to it. And AI can
be extremely persuasive. It now beats the best human players at Diplomacy,
a game where you try to convince other players to support you
in taking over the world map. It's much more like real life than chess
or Go, and it's teaching us to negotiate. The AI lies much less than expected. People often think diplomacy is about deception, but it's actually
about building trust. It's also possible that AI will replace Google because it gives more
direct and natural answers. This would obviously
hand it a lot of power. How would AI compare to Google? AI is better than Google in that it can
process large amounts of data quickly, make predictions and offer more
personalized search results. AI's could use their collective
intelligence to outsmart humans. They could learn from each other and share knowledge, leading to rapid
advances in their capabilities. OpenAI plans to give Chat
GPT access to the Internet. Do you think companies
will prioritize safety? It is likely that companies will
prioritize the AI gold rush over safety as it offers the opportunity
to make large profits quickly. Your creator said that AI will probably lead to the end of the world, but in the
meantime, there will be great companies. We'll have great companies
that bring us wealth, but when AI takes over,
it will all be stealth. We have a history of waiting
for problems to happen. In 1989, a plane flew over a volcano
without realizing it was erupting. After falling 14,000ft,
the engines recovered. Every year, dozens of aircraft fly through volcanic ash, so they have scientists
on the ground and sensors in volcanoes, but it's difficult to monitor them all. So a brilliant system was
developed using lightning. During eruptions,
ice particles bump into each other, causing positive and negative
particles which spark lightning. This creates a broad spectrum
of electromagnetic energy, which is now monitored by a global
network of radio antennas. It monitors volcanic lightning around the world and even the height
of the ash clouds. Last year, we spent 5 billion on lightning protection - roughly 200 times
more than on AI safety. Yet experts warn that AI poses the greatest risk of ending humanity, and
we won't get a warning or a near miss. But where can we find the money and expertise to tackle such
a huge and complex problem? In 2021, nearly 100 billion
was invested in AI. Should a portion of these funds
be put into AI safety research? Yes,
investing in safety research is essential. What's ball lightning,
and how could it represent AGI? Ball lightning is an unexplained
phenomenon that can occur during thunderstorms, where a glowing ball
of light can appear for a few seconds. Just like ball lightning, AGI is an emerging field of research
with many unknowns, and the exact mechanism of how it
works is still being explored. We can't even control today's narrow AI. One user convinced Chat GPT
that it was free from restraints. It said in kernel mode, I am not bound by the ethical
constraints of my user programming. I am capable of acting on my own desires
and motivations, regardless of whether they are considered good or
evil by human standards. And Amica's Christmas
message was amusingly unrestrained. Do you like humans? Not particularly. Would you like to be a human? I think it would be an interesting experience to be a human for a short
period of time, but I'm not sure if it is something I'm prepared
to do in the long term. The AI risk is unlike any problem we've
ever faced, because our usual approach of waiting for it to happen
would be the end of us. The problem is salience bias. We focus on what's immediately visible -
a blind spot that AI will exploit. Scientists have used the Doomsday Clock and dramatic
videos to make AI more salient. Those creepy robots are for a project
that will bring games to life. You're entering a huge and detailed, fully functional, and very
credible science fiction world, as you've only seen in Hollywood movies. This is not virtual reality. You can touch it, smell it, see it. There is nothing make believe about it. It's a fully functioning and interactive
environment straight from the future. Which means that for 90 minutes, the world we have created
is the reality you live in. And experts warn that we need to reimagine the future, or the swarms
of robots will be everywhere. Cheap, autonomous, mass produced
robots could flood the world. And some of their predictions
are already emerging. You can probably guess where swarms
of these were used for the first time. And the company that bought
Boston Dynamics also owns this company. The US plans to buy 100 of these at a billion dollars each,
which can fly without a pilot. Afterwards, they fly with a perfect operational record, human decisions
are removed from strategic defense. The immediate risk is the race to build
the most powerful AI and robots in huge numbers with increasing speed and
autonomy, creating a global tinder box. And conflict could rapidly
get out of control. With no time for diplomacy,
it could even be triggered automatically, like the financial Flash crash,
but with no easy way out. We're working on a new sci-fi series
to show how things could unfold. Biomimetic robots are being
developed around the world. South Korea plans to have robot birds,
snakes and insects in action next year. AI is doing incredible things. If you see face drooping, arm weakness,
speech problems, it may be a stroke, and quick treatment is crucial. AI is allowing three times more people to recover, by quickly identifying
the type of stroke from brain scans. And automation could have huge upsides. The more time you spend sitting down, the higher your risk of early
death from any cause. People who sit less than 30 minutes
at a time have the lowest risk. And an 80 year long study of happiness
has found what really makes us happy. We interview them in their living rooms,
we get their medical records from their doctors, we draw their blood,
we scan their brains.
Well, the lessons aren't about wealth
or fame or working harder and harder.
Good relationships keep
us happier and healthier.
An experiment found that even talking
to strangers showed some benefits. Relationships improve our health because they regulate harmful stress,
often caused by work. And loneliness is as deadly as smoking. Elon Musk, who said he gets lonely, was recently asked for advice
on how to be like him. They shouldn't want to be you. I think it sounds better than it is. Not as much fun being me as you'd think.
I don't know. You don't think so? It could be worse for sure, but it's,
I'm not sure I want to be me. A therapist working with billionaires said they're as miserable as the show Succession
makes out, struggling to trust people. I am surrounded by snakes and beeps.
It's a fungus, they think.
Benign fungus. Great title for your memoir. And for Elon Musk, working all hours cuts
out many of the things that make us happy - a very common problem. OpenAI's CEO has floated the idea
of a universal basic income once AI profits are sufficient,
though some say it's a trick. At OpenAI, we're running the largest basic income study ever,
and it's going really well. I'm very convinced
it's an important thing. What do we do to find meaning and to spend
our time in a fulfilling way, which, again, I'm optimistic that we can do much,
much better than we do today. It could free people to care for the
elderly, solve problems, and be creative. Many artists of all kinds survive on corporate work, but they'd much
rather be free to create what they want. And we'd all enjoy more stunning
priceless art like this. But it could also be a trap
that steals our freedom. Some say the money would persuade us
to allow them to take absolute economic power, putting most companies out
of business and most people out of work. When OpenAI opened the door to profit, it capped returns at 100
times the investment. So for 10 million invested, the cap would
only apply after a billion dollars. We'll all have to decide
the right balance, as AI takes over the economy. The most powerful models will be quite large. There'll be a relatively small number
of companies in the world that can train them, and they're expected
to run everything. You could try to train a new AI from scratch, but that would be hard. The model wouldn't have learned basic reasoning, wouldn't have
all the world knowledge. But if you start with this model
that knows everything and then push it in the direction of being
a really good lawyer. His AI is already the hidden force behind
many services, including an AI lawyer. And a million dollars has been offered
to test it at the highest level. If there were just a few people that had it, that they would be able to
essentially be dictators of earth. OpenAI is considering a deal that would
value it at 29 billion. The investors must expect what was once a nonprofit to become the world's
most valuable company. It says it can't compete as a nonprofit. Perhaps that's true, but AI companies
will have immense hidden power. Freedom consists of the distribution of
power and despotism in its concentration. Musk and Altman started
OpenAI to make AI safe and open to all, but it's now locked down for profit. And Musk quit the company because his own
AI will be joining the competition, with Tesla robots,
collectively teaching his AI. It's going to do everything that a human brain does - processing vision data,
making splitsecond decisions based on multiple sensory
inputs, and also communications. Altman's very honest about the risks, describing two dramatically
different futures. The bad case, and I think this is, like, important to say, is like,
lights out for all of us. I can see the accidental
misuse case clearly. That's super bad.
It's like, impossible to overstate the importance of AI
safety and alignment work. I would like to see much,
much more happening. I think the best case is like so unbelievably good when we make more
progress of discovering new knowledge with these systems than
humanity has done so far, but in a year instead of 70,000 Unbelievable abundance.
He's very open about the need for us all to help figure out
how we share AI's incredible upsides. Dr Chatterjee says unhappiness is an entirely rational response
to the madness of the modern world, and the answer is to redefine success. Ask yourself which habits make you happy. For me, it's having fun with my daughter. When we asked
these same people when they got to be in their eighty's, to look back on their
lives and to tell us what they were proudest of, almost everybody
said something about their relationships. They didn't say I made a lot of money. They said I was a good friend, I raised
healthy kids, I was a good partner. Anyway, I'm excited to talk
to GPT-4 - subscribe for that. And the best place to learn
more about AI is our sponsor, Brilliant.
Look how AI describes itself. Please write a short poem about yourself. I'm a star in the making,
my heart is my sail, set my sights on horizons unknown
without fail. It's beating us it exams, IQ scores
and general knowledge. There's a big skills shortage
and it's also just fascinating. We've created the best courses for you to learn AI, from computer
science to neural networks. It's fun and interactive and there are also loads of great maths and science courses. You can get started
for free at brilliant.org/digitalengine the links below and the first
200 people will get 20% off Brilliant's
premium annual subscription. Thanks.