Feel a bit disappointed about the “reason” why Andrew is not optimistic about watermarking technology being used to help us detect AI generated contents.
Pleae tolerate the possibility that I perceive the context of the letter inappropriately.
I have subscribed The Batch for a while and checked it out from my mailbox this morning. Andrew’s letter of this week is directing us to the following situation:
As AI tools like ChatGPT have become increasingly sophisticated and popular, some people may abuse them. For instance, a student could use them to complete assignments. (It sounds not that harmful, does it?) How should we tackle such problems? What should we do?
Andrew mentions the effective watermarking technology that can help to tell AI generated contents. However, he “rules it out” immediately (at least the tone sounds like it to me) and briefly explains the reason(s):
Unfortunately, I’m not optimistic that this solution will gain widespread adoption. Numerous providers are racing to provide text-, image-, and voice-generation services. If one of them watermarks its output, it will risk imposing on itself a competitive disadvantage (even if it may make society as a whole better off).
…
If one company were to establish a monopoly or near-monopoly, then it would have the market power to implement watermarking without risking losing significant market share. Given the many downsides of monopolies, this is absolutely not the outcome we should hope for.
And he concludes (for the letter):
So what’s next? I think we’re entering an era when, in many circumstances, it will be practically impossible to tell if a piece of content is human- or AI-generated. We will need to figure out how to re-architect both human systems such as schools and computer systems such as biometric security to operate in this new — and sometimes exciting — reality. Years ago when Photoshop was new, we learned what images to trust and not trust. With generative AI, we have another set of discoveries ahead of us.
I highlight a couple of his words in bold and hope to ask some questions (to myself):
- Why do we pursue AGI originally?
- If AGI can be approached/achieved, what should it do together with us?
They are probably the same questions on earth and I am not sure about the answer(s). It depends on the domain, I guess. But the answers were certainly not supposed to be:
- Win the race against other companies and be the leader of “__” field.
- Make the best profit from the core competency, i.e. “__”.
- Keep the ecosystem of “__” active and fair.
- …
The wildcard “__” can be “AI” or anything. (“anything” indicates my tone.)
However, such arguments are certainly playing crucial roles in decision making and public debate.
I also feel that the “Photoshop” analog is misleading because “Photoshop” and e.g. “ChatGPT” are very if not completely different things and have different scopes of use case.
“Photoshop” is a specific (type of) product with specific features. It has a well-defined client, i.e. an image or a video. The boundary of red line is relatively easy to draw, e.g. what is not allowed to be “photoshopped”, otherwise, it is regarded as “fake” instead of “artistic”.
AI tools such as ChatGPT that “produces language”, on the other side, have a vague client, i.e. language. It can be a character, a word, a sentence, a paragraph etc. If I switch the terms, it can be a greeting, a response, an answer, a solution, a story, an essay etc. Language, due to its property and flexibility, is the component of or the medium to that client. How that client is embodied is upon us, human.
At this moment, I do not have any answers or solutions to the dilemma yet-to-be occuring in the must-have exciting new world. At least, new rules seem necessary and the rules should not be made only by those players in the game.