Generative AI tools such as ChatGPT, Google Bard, Midjourney, and Stable Diffusion make it fast, cheap, and dead simple to make images, text, music, and other content without innate talent, learned skills, or focused effort.
The ability to generate creative content without the need to know how to make it is an amazing gift from the gods of technology for the broad public. For those who are inspired and driven to craft things with their own imagination and hands, the rise of AI is a catastrophe.
As the world becomes awash with cheap and easy AI-generated knockoffs, many forms of art and expression a put on a collision course with commoditization. And while there remain noticeable quality gaps between what talented humans and the latest AI large language models produce today, the quality of AI output will very soon reach the point where it’s indistinguishable from human-crafted.
There’s a scene in the second episode of the HBO series Westworld, where the character William asks Angela, the concierge at the AI theme park, whether she is human or not.
William: "Are you real?"
Angela: "If you can't tell, does it matter?"
This question from 2016 sci-fi has even greater resonance and scope today. If you are entertained by a novel, does it matter to you if the story came from:
The mind of an author who imagined a story and considered the selection and placement of every word, phrase, sentence, description, and line of dialog to express their vision.
A large language model AI that examined thousands of stories in its training data and selected each character of text based on the probability that will result in a string of words, sentences, and pages that resemble what people consider to be a good story.
Why It Matters
Affordable, high-quality, machine-woven rugs with beautiful, intricate designs and flawless construction have been available for a long time. Yet people still pay thousands of dollars for hand-knotted Persian and Oriental rugs—even preowned ones—with all of their imperfections. Why?
There’s a deep human instinct to create things and an innate sense of value we put on things people create. Human-made things are vessels that contain the imagination, craft, and effort someone put into making them. Whether it’s a rug or a film or song or an oil painting or a GIF, there will always be people who appreciate and value things imagined by human minds and crafted by human hands.
Fighting Fugazis
So, in a world where AI-generated becomes indistinguishable from artisan-crafted, how can we tell the difference?
This is not a new problem. There’s a long history of counterfeits, fakes, forgeries, knockoffs, derivatives, reproductions, and unauthorized copies of original creative works. We’ve developed an assortment of systems and methods for countering and mitigating them over the centuries—albeit imperfect ones.
When the gods of technology close one door, they also open a wormhole to another dimension. There are a number of emerging technologies that together may hold the key to saving human creativity.
Digital watermarking: Various methods exist for embedding markers into digital files that establish their authenticity or creator. In the case of AI-generated content, digital watermark information could be automatically embedded by default as it’s created. Some propose that this becomes a requirement in AI regulations. Creators can proactively watermark their works to combat copying and label it as something LLMs should exclude from their training data.
Proof-of-personhood: When bots can convincingly impersonate humans, how can you prove that you are not an AI? Several approaches are in the works that aim to address this by providing means verifying that we each are, in fact, each a unique human person. Some approaches, like Worldcoin, capture scans of an individual’s biometric markers to prove unique humanness. Others, such as BrightID and Proof of Humanity take a social-graph approach. Individuals participate in some activity whereby other pre-vetted humans vouch for the newbies’ homo sapiens bona fides to earn them a verified digital ID (DID). Vitalik Buterin provides his insights into the pros and cons and risks of each approach in an informative blog post.
Cryptographic provenance: NFTs are for far more than monkey jpegs. Unique nonfungible tokens can be used to provide immutable, tamper-resistant authentication of original works recorded on a public blockchain. Think of this as a global decentralized property title system for all types of creative work. When you buy a book from an author or a song from a band, you will also get an NFT that proves it’s authentic and that you own it. If you give it to someone else or sell it, the NFT goes with it. This preserves the ownership record from the creator through the entire chain of owners. The only way to demonstrate you have the genuine article is to also hold its NFT.
The Three-Body Solution
The three techs can work in concert to offer human creators a kind of digital, on-chain makers mark for permanently branding creative works as made by a particular human person. (Plus, the creator could remain anonymous or pseudonymous.) It works like this:
An author publishes a novel that includes digital watermarking.
The watermark integrates the author’s proof-of-personhood DID as a signature.
Each copy of the book is paired with an NFT minted by the publisher that includes metadata about the book and the author’s DID.
There are still many technical hurdles and open questions to address to get to this kind of solution. And the main question remains: In the end, for most people, does it matter?
I believe it does. Preserving ways for people to be recognized and valued for the things they create matters very much. Fortunately, there are also many of us who will always appreciate the many things that originate from human inspiration and not machine generation.
A world filled with art made without inspiration will be uninspiring indeed.
