The below is a masterful oil painting of the Sears Tower at dawn. You can't find it at the Art Institute, or the Met, or even through a Google search. It sits as copy 1 of 1 on my hard drive (and now in this post). The painting is a wholly original creation by a Machine Learning model called DALL-E, built by OpenAI, crafted from the prompt I gave it: "oil painting of downtown Chicago at sunrise."
DALL-E and similar AI models (Midjourney is another popular one) have burst onto the scene, to much amazement and consternation. They are winning art contests. They are redefining what it means to be an artist or designer. And they are just getting started.
None of this should be surprising. To paraphrase Ben Thompson: content on the internet has generally flowed along three phases. Text yields to imagery yields to video. Here are two, very broad, examples:
Blog sites like Tumblr helped my generation catalogue their daily lives until the rise of Facebook and Instagram, which are now in an existential battle with long-form and short-form video content from YouTube, Tik Tok, etc....
AI-driven services have been generating original content for several years now. Whether writing sales copy or auto-completing your sentences on Google Docs or Gmail, most people in the developed world have interacted with AI generated original, text-based content, whether they were aware of it or not.
With that context in mind, the rise of DALL-E is the natural evolution of AI technologies into the imagery phase. While it currently struggles to create photo-realistic imagery (it's still stuck in the uncanny valley), the nature of the technology means that it is getting better ("learning") every day. And, if DALL-E is an evolution from text to still imagery, it stands to reason that AI models will one day be able to generate wholly original video content.
Companies are taking notice. Tik Tok is successful because its proprietary algorithm learns what a user wants, and feeds that person a stream of related content. But the content Tik Tok can feed you is bounded by the availability of human-generated material about, say, your newfound baking obsession. But what if Tik Tok weren't constrained by what human content creators filmed? What if the algorithm could learn what you wanted to see, and then create wholly original content about baking that was uniquely designed to suit your tastes? The added benefit would be no more human content creators to share revenue with.
I'll reserve judgement on whether feeding the next generation ("Generation Alpha" I'm told - I guess that's what happens when get to the end of the alphabet) a constant stream of AI-generated content specifically designed to monetize them is a good idea. Maybe on the next post. This is where I bring things to an area I care deeply about - National Security, and preserving the fabric of liberal, democratic societies.
Imagine a world in which an adversary, a rogue actor, or even a rival political organization can create realistic, video-quality AI-generated content to sway public opinion. Here's what DALL-E created when I prompted it to create a "photo of damage to a hospital":
It's not hard to imagine the above picture going alongside a fake, breaking news article about Ukraine bombing a Russian children's hospital. At a moment where trust in institutions are at an all-time low, where truth and facts are already up for debate, the slow, seemingly inexorable move towards AI-generated video content feels like a disaster waiting to happen. Another tool in the deepfake arsenal.
This will be a major problem that requires the cooperation of governments and the tech industry to solve. We need smart public policy and technical tools that are just as smart on detecting AI-generated content as they are at creating it. We all know that government policy is reactive, so the onus will be on the private sector to find ways to mitigate the problem before any damage is done. Perhaps there is a way to incentivize the big players in this space to police each other.
For the time being, it's important to stay aware of what's out there, and to recognize that there are always risks alongside the benefits of technological change. When it comes to AI, the stakes are high enough that we need to neutralize the risks before we can fully embrace the conveniences.