The New Avengers: How AI is Giving Us Superpowers
The New Avengers: How AI is Giving Us Superpowers
From J.A.R.V.I.S. to Artistic Intelligence, Examining AI's Applications
Weekly writing about how technology shapes humanity, and vice versa—always through a lens of optimism. If you haven’t subscribed, join 40,000+ weekly readers by subscribing here:
The New Avengers: How AI is Giving Us Superpowers
To kick off 2023, I wrote a deep-dive into AI and its budding application layer. Three months later, a lot has changed. Predictions and market maps from January seem quaint; things are progressing much faster than most people anticipated. In March alone, OpenAI launched GPT-4, ChatGPT got plugins, and Bill Gates declared this “The Age of AI,” arguing that the development of AI is as fundamental as the creation of the personal computer, the internet, and the mobile phone.
Because a picture is worth a thousand words, here’s a side-by-side of Midjourney’s output for the prompt “Donald Trump and Barack Obama playing basketball” in March 2022 (left) vs. March 2023 (right).
Impressive.
Hearing Gates proclaim a new tech epoch reminded me of another time he professed excitement for a budding technology—the internet. In 1995, Gates went on the The David Letterman Show to explain the web. That interview was met with ridicule (you can watch it on YouTube here), with Letterman giving wry responses to Gates’s excitement about streaming a baseball game (“Does radio ring a bell?”) or being able to experience that game online even after it’s over (“Do tape recorders ring a bell?”). He just didn’t get it.
This time is different. People are paying attention, if only because life has changed so dramatically in the last half-century. When my dad was growing up, TV had three channels and he had to look up information in a leather-bound encyclopedia. Since Gates appeared on Letterman, we’ve become accustomed to Google’s Search and Amazon’s Everything Store and Facebook’s social graph. Tech companies have become verbs that dictate our lives—we Instagram special moments, Snap our friends, Uber to work, Airbnb when we travel.
The pace of change has been rapid, but we’re only on the cusp of the digital revolution. Computers used to be really good at one thing: computing. In other words, doing calculations. Now they’re becoming good at learning (hence the term “machine learning”), operating more like the human brain. And computers are capable of moving much, much faster than our brains—an electrical signal in the brain moves at 1/100,000th the speed of the signal in a silicon chip.
This week and next, we’re revisiting AI’s application layer. It’s only fitting, given the Cambrian explosion in the startup world: about 1 in 2 startups in Y Combinator’s Winter 2023 batch is building a product that uses OpenAI’s APIs. This piece dives into 1) What’s happening now, and 2) What might happen down the road.
I’ve structured this piece to look at three themes that excite me:
Personal Assistants for Everyone
Amplifying Human Knowledge
Amplifying Human Creativity
Then next week, we’ll examine three important questions that stem from the AI revolution:
What is reality in a world of generative AI?
Will AI take our jobs?
Who will capture the value being created?
Let’s dive in.
Personal Assistants for Everyone
Back in January, I wrote about the obscure 90s film Hyperland, created by Douglas Adams of Hitchhiker’s Guide to the Galaxy fame. The premise of Hyperland (which you can watch on YouTube here) is that Adams is fed up by passive linear TV—what the film calls “the sort of television that just happens at you, that you just sit in front of like a couch potato.”
Seeking a more interactive form of media, Adams takes his TV to a dump, where he meets Tom (played by Tom Baker). Tom is a “software agent”—essentially, a digital butler capable of personalizing your life and carrying out tasks for you.
Tom reminds me of modern-day AI.
Last week, OpenAI unveiled ChatGPT plugins. Plugins essentially mean that ChatGPT users can now interact with 11 launch partners—including Instacart, Expedia, and OpenTable—all from the ChatGPT interface. Like Tom, ChatGPT plugins act like digital butlers.
How plugins work is best illustrated with an example:
Ben Thompson asked ChatGPT for a recipe that includes pork and cabbage. You can see in this screenshot that at the end of ChatGPT’s response, ChatGPT asks Ben if he’d like a shopping list for the recipe. Ben says yes, and ChatGPT spits out an Instacart shopping cart with all the requisite ingredients.
This, of course, uses Instacart’s plugin. But you have other options too: with plugins, ChatGPT can book you a restaurant (OpenTable), buy you a product (Shopify), or stitch together apps (Zapier). Everyone is getting their own personal assistant.
Imagine giving this prompt to ChatGPT:
“I’m going to Paris for seven nights. Please book me flights that leave in the morning on March 30th and get me home in time for dinner on April 6th. I’d like to stay by the Louvre, no more than $500 per night, and would like dinner reservations each night within walking distance to the hotel.”
No more laborious, monotonous interaction with Expedia and Google Flights and OpenTable. OTAs like Priceline and Expedia reconfigured travel in the 2000s, rendering travel agents obsolete; AI models could do something similar to today’s industry (though currently the bookings still go through OTAs like Expedia that have partnered with OpenAI).
What’s also new about plugins is that they let ChatGPT connect to the internet. Previously, ChatGPT was limited by the dataset it had been trained on, causing it to omit events past 2021. For a while, for instance, ChatGPT didn’t know that Elon Musk is CEO of Twitter. Now, ChatGPT can access the web’s information in real time and offer up-to-date responses. That’s big.
When I think about AI applications, I think of assistants and I think of amplifiers. The lines between the two are blurry, but this framework helps me think about use cases:
Amplifiers are about augmenting human abilities; they give us superpowers that make us superhuman. More on amplifiers in the next two sections.
Assistants are less about augmenting humans and more about saving us time and effort by completing tasks for us. Assistants are the Toms, the digital butlers. Or to use a more modern reference—and to continue with the superpower analogy—assistants are J.A.R.V.I.S. from Iron Man, the computer companion that serves Tony Stark. We all want to be superheroes, and AI helps us get there, but we also need our trusty digital sidekick to make our lives easier. With AI, we now all get our very own J.A.R.V.I.S. (fun fact: the acronym stands for Just A Rather Very Intelligent System).
We see personal assistants in ChatGPT plugins booking our hotels and ordering us groceries. We also see them in other exciting new startups; Harvey, for instance, acts as a personal assistant for lawyers, helping with contract analysis, due diligence, litigation, and regulatory compliance. It’s like having your own paralegal, at a fraction of the cost. This is an interesting trend, and we’ll see more verticalized assistants pop up. Next week, we’ll explore what this means for labor market disruption—paralegals, for instance, could be among the jobs at risk.
Another reason plugins are so fascinating: they signal ChatGPT becoming a platform. Soon, thousands of companies might partner with OpenAI. Packy McCormick and Ben Thompson had some good thoughts this week on what this might look like. Many smart people have compared this emergent platform to Apple’s App Store.
Some thoughts on how this might unfold:
When ChatGPT launched, I expected its growth to be gated by distribution. ChatGPT doesn’t even have an app; users are accessing the tool via the web. But this clearly hasn’t stood in the way of ChatGPT becoming arguably the fastest-growing product in history, taking two months to hit 100M users.
My TikTok feed demonstrates ChatGPT’s mainstream appeal: video after video and comment after comment feature people talking about using ChatGPT to do their homework, to sketch out their lesson plans, to write their emails. It’s a certified phenomenon.
What’s surprised me is that ChatGPT is becoming a new interface, which plugins solidifies. I’d thought that existing distribution would win—Snap, for instance, integrating ChatGPT into its My AI feature for 750M monthly active users. But instead, people are interacting with businesses through ChatGPT itself.
It will be fascinating to see how businesses adapt themselves to this new reality. When Google Search became dominant in the 2000s, businesses optimized their websites with keywords for SEO. When Instagram became dominant in the 2010s, businesses optimized their cafes and restaurants and hotels for Instagrammable photos. What happens in the 2020s when every company needs to make sure that it incorporates generative AI? How does Uber vie with Lyft for riders when a user asks ChatGPT the simple command, “Find me a ride to the airport?”
Depending how plugins take off, ChatGPT may be the interface through which users interact with sites we used to visit directly or with apps we used to download—Expedia.com, OpenTable.com, Instacart’s mobile app. We’re seeing a massive shift in the user experience, and plugins were the shot across the bow—if I were Google with Chrome or Apple with its App Store, I’d be thinking carefully about my next move.
Amplifying Human Knowledge
Ruminating on the rise of AI, Bill Gates explains that there have been two defining technology experiences of his life:
The first came in 1980, when Gates first used a graphical user interface. A GUI is how you and I interact with computers—the pointers, icons, windows, scroll bars, and menus that we’re all accustomed to on our screens. It’s hard to imagine now, but before GUIs you had to input a C:> prompt to interact with a computer.
The second defining experience for Gates came just last year, when the OpenAI team demonstrated how ChatGPT could ace the AP Bio exam. Gates had challenged the team to the bio exam for a specific reason: “I picked AP Bio because the test is more than a simple regurgitation of scientific facts—it asks you to think critically about biology.” But the AI model aced the test anyway, getting 59 of 60 multiple-choice questions correct and producing top-tier responses to the essay questions. An outside expert scored the test, giving it a 5, the highest-possible score.
AI will make us smarter. Technology has made us better at math for decades—calculators, Excel spreadsheets, computer programs. We get computational superpowers. Think of the same analogy, but applied to all human knowledge. We see this in emergent startups like Hebbia (neural search for the enterprise), Rewind (“The Search Engine for Your Life”—essentially a better memory), and Kumo (predictive intelligence) that expand our brainpower and our human abilities. We see this in established productivity tools like Notion and Google Workspace that are integrating generative AI, letting you summarize meeting notes and write thoughtful emails.
ChatGPT’s laurels continue to pile up: ChatGPT passed an MBA exam at Wharton; ChatGPT took the bar and passed; ChatGPT passed medical licensing exams.
And AI model abilities will only improve. One fascinating chart to end on—how GPT-4 compares to GPT-3.5 on exam performance.
In just a short time period, we’ve leapt from 40th-percentile to 88th-percentile on the LSAT and 20th-percentile to 70th-percentile on AP Chemistry.
Amplifying Human Creativity
There a few laws that govern the universe. One such law: as people approach age 30, their interest in interior design grows exponentially. I’m no exception: my Instagram Explore Feed is littered with home decor content, and my YouTube homepage is almost exclusively Architectural Digest home tours.
Recently, I turned to Midjourney to get design inspiration for our apartment. As someone with precisely zero artistic talent, I was actually fairly successful. Here’s the output for the prompt:
Architectural Digest Style Photo, New York apartment, Contemporary, Primary Bedroom, Fireplace, Dark-Patterned Wallpaper, Animal Print, Leather, Black Rug, Restoration Hardware, Yellow Lighting, Nighttime, Cozy Vibe, Unique --ar 16:9
This is what’s so exciting about generative AI; it can make anyone (even me!) more creative. Artificial intelligence 🤝 artistic intelligence.
When you see the fidelity of generated images, your mind starts to think about the possibilities. In a talk with Sequoia last week, NVIDIA’s CEO Jensen Huang said: “Every single pixel will be generated soon. Not rendered: generated.” It’s interesting to think of this applied to rich media formats like film and gaming. We can see the improvement in Midjourney’s creations already (think back to the Obama v. Trump basketball image earlier), and the steep slope will only continue: soon, we’ll generate rich video and immersive 3D worlds.
An example of Midjourney v5 from Nick St. Pierre reminds me of HBO’s White Lotus. It’s the output for the prompt:
1960s street style photo of a crowd of young women standing on a sailboat, wearing dior dresses made of silk, pearl necklaces, sunset over the ocean, shot on Agfa Vista 200, 4k --ar 16:9
Imagine having an idea for a story, and being able to generate an entire TV show off of your script. One of the broad themes in Digital Native is that the arc of creative tools bends toward more affordable, accessible, and high-quality creation. Generative AI could make it possible for anyone to become a showrunner. This could, of course, have a major impact on jobs; we’ll unpack that next week.
We’re already seeing flavors of user-generated generated content (UGGC?). Last week, Runway presented the 10 finalists in its festival for AI-generated short films. You can watch the 10 videos here. The films are impressive and engaging (and you’ll see in the credits that they still employ plenty of people). It’s interesting to think about next-gen content platforms, social networks, and games that incorporate generative AI—your 2023 versions of Minecraft or Roblox or Rec Room.
One final example from St. Pierre of Midjourney v5 being used creatively—product design. Here’s the output for the mock-up of a Nike Air Force 1 and Slytherin collab. The prompt:
Street style photo, Closeup shot, Nike Air Force 1 slytherin collab, unique Colorway, snake skin, hogwarts, natural lighting, original, unique, 4k --ar 16:9
Pretty slick. Nike’s current customization tools are rudimentary, but that’s bound to change; you can imagine generative AI expanding who can be a designer. Just in the past few weeks, both Adobe and Canva launched AI products: Adobe unveiled Firefly, which includes a text-to-image generator, and Canva launched Magic Design, which creates design templates based on your inputs. One open question is who will capture the value being created here: incumbents like Adobe and Canva, or the startups being founded today. We’ll explore that more next week.
People are finding innovative ways to use AI in their creative work. The YouTuber MKBHD (17M subs) has been using ChatGPT to write his video scripts; Isaiah Photo (9M subs) let AI run his channel for a week; and people are even using ChatGPT for DJ sets:
There are some dystopian elements to AI’s creative side, of course; at what point does AI stop augmenting our creativity and start negating it? But AI is certainly proving that it’s capable of human-level creativity. In one study, researchers compared human-generated ideas with AI-generated ideas across six chatbots, reporting: “We found no qualitative difference between AI and human-generated creativity.” (!)
When it comes to creative work, it’s interesting to ponder how new technologies shape art. My friend Trung Phan recently wrote about how photography changed painting. In short, before cameras came along, most painting was hyper-realistic; artists sought to capture reality. But when photography became possible, artists no longer felt the need to mimic the real world, instead feeling free to paint abstract and creative works. Monet, for instance, used colors in his famous water lilies that didn’t exist in nature. And Van Gogh’s Starry Night was groundbreaking in how it didn’t represent the realistic night sky, but instead captured Van Gogh’s own interpretation of it.
It will be fascinating to see how generative AI expands our definition of art. Just as cameras changed painting, generative AI will change writing and music and film and design. The technology is bound to be a controversial, transformative tool in the arsenal of creatives. It will influence the art we make and the stories we tell in ways good and bad, all while crowding in more creation by making the act of creativity more accessible to more people.
Final Thoughts: AI Is Electricity
In The Atlantic last week, Jacob Stern debated which technology AI most resembles in terms of its impact on society. The best analogy, in my mind, is electricity. As Stern points out, many “technologies” are fairly limited tools: a saw for cutting, a pen for writing, a hammer for pounding nails. Electricity, meanwhile, has no specific function, acting less as a tool than as a force that pervades all aspects of life.
We’ll see AI bleed in unexpected ways into every part of life. We’ll all get digital butlers, personal assistants that make mundane life tasks easier; we’ll have our knowledge meaningfully expanded (again, a silicon chip’s signal moves 100,000 times faster than our brains); and we’ll become more creative, unlocking new artistic ability. Put more simply: we’ll get superpowers, and we’ll get a trusty sidekick too.
Of course, AI’s rapid adoption and exponential improvement curve raise important questions. Next week’s piece will tackle three of them:
What is reality in a world of generative AI?
We’ll talk deepfakes, Tom Cruise, and Hollywood de-aging.
Will AI take our jobs?
OpenAI put out a study that AI may impact 80% of U.S. workers; we’ll unpack whether that’s true, and which jobs will be first to change.
Who will capture the value being created?
There are good arguments for why incumbents with massive distribution will win, and there are good arguments for why startups have the edge. We’ll look at what moats exist in the world of AI.
Until then 👋
Sources & Additional Reading:
There have been some great recent pieces about AI in mainstream publications like The Atlantic (by Jacob Stern) and The New York Times (by Ezra Klein)
I’ve enjoyed pieces on AI from some of my favorite writers, including recent ones from Ben Thompson, Packy McCormick, and Mario Gabriele
You can read OpenAI’s announcement of plugins here
Bill Gates’s piece The Age of AI is here
Nick St. Pierre has many of the great visuals above on Midjourney v5 in a Twitter thread here
Watch the 10 finalists in Runway’s generative AI film festival here
Related Digital Native Pieces
Here is January’s deep-dive into AI: AI in 2023: The Application Layer Has Arrived
Thanks for reading! Subscribe here to receive Digital Native in your inbox each week: