Artificial Intelligence Gets Real
Does my photo look a bit different this month? This is the new me, thanks to the website Generated Photos (generated.photos).
This person does not exist. His photo and millions of others you can view on the website have been generated by artificial intelligence. You can purchase a photo for $2.99 or use photos for free if you link back to the site. Bulk orders cost less, such as 1,000 images for $1,000.
Why would you want to? Some people use them online instead of their real face. Digital designers use them in projects such as computer games, web and mobile applications, and landing pages.
They seem astonishingly real. Take a look for yourself.
More important than realistic faces, though, is a new development: the artificial intelligence tools to create faces, text, conversations, computer programs, and more are now being made available for commercial use. The tools aren't difficult to use. The New York Times used a computer graphics tool from Nvidia (called a Generative Adversarial Network) to create amazing examples of artificial faces. Not only do they look real, as you can see (nyti.ms/3fP7138), but you can tweak the faces with a slider in order to change things such as age, mood, gender, and ethnicity.
Here's another example of an amazing advance in artificial intelligence. The following paragraph on creativity was completely generated by a computer in response to the prompt, "How do we become more creative?"
"I think creative expression is a natural byproduct of growing up in a diverse world. The more diverse the world is, the more you get exposed to different people, to different opportunities, to different places and to different challenges. And the more diverse that is, the more likely you’ll be to be able to put the dots together to form something new. And in many ways, I think if you want to be creative, you have to go for it. If you want to be a writer, you have to write, if you want to be a musician, you have to create music, if you want to be a comedian, you have to create comedy, and the more you create, the more likely it is that you’ll stumble onto some interesting stuff."
The tool in this instance was GPT-3, a new system unveiled this past summer by an artificial intelligence lab in San Francisco called Open-air. GPT-3 spent months learning natural language by analyzing thousands of digital books, the entire Wikipedia, and nearly a trillion words posted to blogs, social media, and other internet websites.
Last fall Open-air made it available for testing, with the intention of making it commercially available early this year.
The software has worked so well that even those who created it have been surprised. For example, they didn't imagine it could also be used to program smartphone apps. A developer who builds apps was given access to GPT-3, and he was curious whether the system could replicate what he does.
So he fed the system a number of descriptions of apps (in plain English) as well as the code that was used to create the apps. He then tested GPT-3 to see if it could code apps on its own. It worked. When he described an app for posting and viewing photos on Instagram, for example, GPT-3 created the code to build the app. He would sometimes need to slightly tweak the code it created for an app, but overall it was clear that it greatly sped up app development.
How do these intelligent systems work? They use so-called neural networks, modeled on how our brains function, to assimilate huge amounts of data and identify patterns. Feed a neural network thousands of images of cats, and it learns to identify cats.
In the instance above, the system learned to identify human faces, then created mathematical models of those faces. It then used those models to generate new faces that look human but belong to no one.
In the second instance above, the system identified over 175 billion patterns in the way words are put together, then was able to use what it learned to create novel text.
It's not always perfect, occasionally creating text that's nonsense. But GPT-3 works often enough that it's clear that it's going to have a huge variety of commercial uses. Without any technical knowledge, one can feed the system some examples in a matter of minutes, and then get something useful.
Applications might include chatbots, which are automated conversations used by companies to provide tech support. Or the system could be used to summarize news articles, write blog posts, create recipes, translate languages, answer math questions -- all of which it has shown it can do.
And one day soon it might even write this column.
© 2021 by Jim Karpen, Ph.D.