Transcriber: Hubert Stasik
Reviewer: Zsuzsa Viola We are at an inflection point
for human creativity. These happen every so often
in our history, usually caused by some sort of
technology innovation. The innovation this time is the artificial intelligence,
or generative AI, and it’s going to change the way
we tell stories. I know this personally
because it already changed me. In the last six months, I’ve made over 2000 images
and videos from AI. I was not a creator before this. Perhaps I’m a little addicted to it. My family wonders sometimes
when I share with them the next image. You know, I kind of think
they want to see the next one, but... but along the way, I learned a lot
about how to make things with AI tools and also a little bit about
what it might mean for creativity. We have such strong imaginations
when we are young. When I was a child, I loved to read.
I loved when my parents read to me. I loved playing with Legos,
all the good stuff. And I remember when my parents
brought home a home computer, 16-color graphics. I was in heaven, and I could play this game
where I was Gandalf the Wizard. Later in my life,
I never really learned to paint. You know, video photography
was a little too expensive for me to do, so I didn’t become a creator. But I think there was something
inside me during that time. And I think it’s inside all of us.
It’s been there since the beginning. Humanity has been telling stories
for 100,000 years. And about 70,000 years ago, we learned
how to record them, cave paintings. Eons later, we invented
the printing press, learned how to distribute these ideas. That was one of the inflection points. The camera. The camera allowed us
to record ideas instantly. And then, broadcast technology,
one to many became one to millions. The next big inflection point
has been recent. That was mobile and the internet. Where all of us,
it democratized creativity, it democratized us as creators, we were able to use our camera phones
and capture photos, capture videos, and share them with everybody. And we did, trillions of photos,
trillions of videos have been created, more media than ever
in the history of our species. The next inflection point is now. By the way, it can be scary. Hopefully, after the end
of this presentation, you’ll see that it can also be uplifting. Generative AI takes the camera
out of the picture. And allows you to start with an idea. An idea that you put into text and create a text prompt
into an AI engine. And that comes back
with a turn paper, or a tweet thread, or programming code,
or now images and video. Image creation has really rapidly
and radically advanced in the last six months. About six months ago,
if you would have made something, maybe it had an extra finger
or an extra arm or something, and now it’s becoming
more photorealistic. The way the technology works
is fascinating. In a diffusion model,
which is a machine learning model, we’ll start with a latent space, which is essentially the image, but use random.
It makes random noise there. It starts with a single point
randomly each time and builds out mathematically
based on your prompt as a guide, along with the training data
to generate an image in seconds. Each time that you regenerate, it will start with a different point,
a different seed. And that is why our images look different
every time we create something with AI. And that’s why it’s not copying.
It’s generating something new. So the technology
is the most advanced technology we’ve ever seen as humans. But it’s actually pretty easy to use. Let's do a text prompt together
and see what happens. So, there are some building blocks that a good text prompt
for imaging will have. Specifically, first, a medium. It’s like a photograph or a painting. Second, a subject - a man talking. An action on a setting, a Ted stage, and then, finally, the style. This would be the creative direction. These words here represent
a sort of photorealism. So, let’s see what happens. Normally, this would take
about 15 to 20 seconds. And here we are,
a Ted guy on a stage talking. So, it’s pretty cool,
but it’s really fast. So, let’s iterate a bit and have fun. Let’s put our favorite Gandalf
into the image and see what happens. So, let’s give them some
grotesque clothing as well, just to see what that looks like. So, here we go. And here’s Gandalf, kind of, right? He’s a little funky; I like him, but he is on a TEDx stage,
so that’s pretty cool. Let's do one more thing with Gandalf. Let’s give him some special
effects behind him. He’s a cool guy. And let’s use different styles. So, the style would be 80s sci-fi
illustration, and see what happens. (Laughter) Okay, now that’s a wizard, right?
Okay. It’s pretty good. Now, you might be an artist,
and you might say, I could make that. I couldn’t make that. And if I could, or if you could,
it would take you days because of the detail in this image. It took us seconds. Not every prompt has to be
elaborate to be powerful. This is a prompt study that was done
by user researcher, Lilly Godbole. And she took this prompt of a girl
and just incrementally aged her, just changing the age in each prompt
to see what would happen. Five years old. Gradually getting older. 15 years old. A 20-year-old woman.
A 30-year-old woman. A 40-year-old woman. A 50-year-old woman and a 60-year-old woman. So, the prompt study quickly for me moved from a prompt state to a study
on aging and a study on bias on how the training data
perceives 40 years old. Also notice the fact that the old woman
defaulted to Caucasian. So, there is likely a whole Ted talk on bias in training data
and bias on engine. But it does show you that
by changing the words a little bit, you can get a very different outcome. So, as I did more creations,
that addiction I have. There’s some hacks I learned. This is a little one I’ll share with you. Take an iconic photo
or just a cool photo that you like, and you can put it in an AI engine
and use the describe function. This Elvis picture,
it’s iconic of Elvis on stage. This is what the engine thinks
the prompt would be to create that. These are four prompts
that it comes back with. Now, if you do a close reading
of some of the words up there, you’ll notice that the syntax
is not that syntax that any of us would create
to make this image. [Piles/stacks] -
I do not see those in this image. I wouldn’t come up with that myself
to come up with a prompt to make that. “Lovecore”. I’m not sure that’s a real word,
so AI is creating words. Actually, the style, it’s in the engine.
The style is lovecore. But the point is, it’s not a word
a person would ever come up with to describe this image. My favorite on there is kombucha punk. (Laughter) Seriousl, well, it’s awesome. I was like: “This is cool.
I can work with this.” So, I said: “Look, if our goal is
to get the coolest image to come out of AI, it’s like, work with the AI engines
and give it what it wants. Let’s give it back
these sorts of phrases.” So let’s try one right now. Let’s take the post-70s ego generation, which is a style that came up a lot
with some of the describes I did. Turns out that’s actually
an artistic movement of Chinese artists. Notice the images are all Caucasian again. So again, not to make this the bias talk,
but it's interesting. Let’s combine that post...
[but their cool]. Let’s combine that post-70s ego generation
with one of those funny words that we saw. This was one of the funniest
that I ever found in the six months. Mommy’s on the phonecore. (Laughter) I literally do not know what that means, but I was like,
I got to see what it makes. So, let’s mash the two together, post-70s ego generation
and mommy’s on the phonecore, and see what happens. Boom. There you go.
(Laughter) So, I got the 70s vibe -
a Caucasian woman again. And the telephones are not something mommy’s going to want to spend
too much time on. (Laughter) But her hair is looking great today,
and she’s feeling good about that. (Laughter) You can kind of have fun with these.
As you can tell, these are your Legos. Put in some kombucha punk next time, put
in a brooding mood. See what happens. Video is next. So six months ago, this was
the first video I made with a prompt. And you can see how it's warping. And its video is rapidly
and radically advancing as well. In fact, this week OpenAI announced
some amazing things with video. I’m guessing that this talk
is going to be out of date very quickly, giving how fast AI is moving. But I think the point is
that it’s moving forward. A realism is coming to video,
a wide range of styles and our ability to tell stories in seconds
with a prompt, with an idea is coming, is now. I think what’s going to happen
with this innovation again, I hope in two years it's a lot more. Those dance videos that we all see
on TikTok that we’re sharing right now, the performances are going to become
more videos of ideas, more videos of creation. So, the next thing you can try,
if you’d like, is music. So, there’s this music engine that you can put a prompt
in a text prompt again. It will create a music bed,
it will create lyrics, and it’ll have someone sing it,
all artificial in seconds. So let’s put our favorite post-70s
ego generation mommy on the phonecore, prompt into this track
and see what happens. (AI music text) “Mommy’s sitting on the couch, Scrolling through her phone,
she’s always in her own world. Will never want to let go.
Snapchat, Instagram and Facebook. So she’s gotta get that perfect angle
for the likes. It’s true. Mommie’s on the phone.
She’s always on the phone. Will never get her attention.
She’s always in the zone.” So, that song. (Laughter) That song. (Applause) That’s not a good song, everybody knows. I like, honestly,
it was going through my head, and I had to practice, and it’s like,
I can’t get the song out of my head. (Laughter) What's weird about that song, though, is that that little mommy on the phonecore
created a whole story about a child trying to get
the attention of the mom who’s addicted to social media. I mean, like, so this stuff is amazing. And it’s fun, and it’s uplifting
when you get to use it a little bit. Okay. So, storytelling
is the next big thing. And this is a masterpiece. It's a painting by Hieronymus Bosch,
The Garden of Earthly Delights. You’re probably familiar with it.
There’s a lot going on in this painting. And I said: “Well, what happens if we could ask the AI
to kind of give us more of the story, kind of get in there
and tell us what’s going on?” So I asked for an extreme close-up
of this painting, and this is what it came back with. These are not close-ups. These are not blown up-pictures.
These were generated. This is what the AI thought
would be happening in that painting. So that, to me, was fascinating. It was sort of like fan fiction
for art geeks, you know? And so where researchers
and everyone are going now is this notion of character consistency, this idea that to tell stories,
we need to have the character, not warp on you by the end of the video. With celebrities, this is a lot easier because they’re embedded
in the training data. But for you and I, it’s a little harder.
But that’s changing. Now we can be part of these stories too. This is me. Not me now. This is me from college,
because I’m not stupid. I used the good photo from that. Right? (Laughter) But I never played football. So, this is me from college
with a weird prompt where I’m playing football
with these uniforms that would never be on the football field. This is me in a 19th-century painting. (Laughter) This is me in a movie. This is me in a music video. (Laughter) As you can tell, I had way too much fun
with all of these places to put my favorite character,
myself, in things. And so, it’s really interesting
to think about where this is going. This is me asking the engine,
what do I look like older as a video. In 18 to 24 months, I believe that we’re going to be able
to make videos of ourselves, and what if our lives
and all sorts of things. AI has been trained, as we just discussed,
on hundreds of millions of images, more than any one of us can see
in our entire lifetimes. But each of us has a brain,
the original AGI. And we've been trained on our lives
deeply for a long time. I believe there's an opportunity for us
to take the influence that we have, have the influences,
and the influence that AI has and combine them to create new things, new ways to create
new ideas, new artistic directions. And it’s certainly the right time. AI is hitting an inflection point. The early adopters are now
moving into mainstream. There are 7 billion smartphones. Billions of computers. Everyone’s going to have access
to this technology. We are at an inflection point for human
creativity. The “aha moment” that I had.
I can be a creator. Maybe you’re having one now, too.
You can be a creator. Maybe you will be the next Scorsese.
Maybe you'll be the next Mozart. Maybe the next Picasso. I think we can all create
new things together, and I can’t wait to see what you create. Thank you. (Applause)
