Generative AI (e.g., ChatGPT) Content Spamming—Concerns

Hi Kiron,

My concern relates to: A & B

a) Someone using a tool like ChatGPT, generating content, tweaking it and posting it?

b) Someone using a tool like ChatGPT, generating content and posting it "as is" without verification or fine tuning?

Hi Rami,

On concern A: I agree with you if someone “takes some input/knowledge provided by ChatGBT to enrich a human expert response,” but isn’t there a difference when someone “takes content from ChatGBT, makes minor content/formatting edits, and then represents it as a human expert response”?

The lines get blurry. But you can also use tools to test for content generated by a chatbot. In addition, you can look at other patterns, such as “repetitive post timing,” wherein you can see a user generating long-outlined content in short intervals against varying questions/subjects.

If we, speaking in the “general we,” dilute the “human expert opinion” value proposition. Then, we have made a significant step toward surrendering knowledge growth as we know it.

I don’t disagree with your rationale, George. Copy and Paste is not the way to go for sure!

You’ve raised a really important issue, Kiron, one which is not getting nearly enough serious attention from generative AI tech companies. In fact, I think it’s a problem that they have no incentive to find a solution for and will not be solved properly without the coercion of appropriate regulation and legal remedy.

I've been advocating for digital forensic watermarking since the release by OpenAI of DALL-E 2, in April 2022, to solve the problem of (1) provenance i.e. explicitly identify who originated the content and when? and (2) attribution i.e. explicitly identify who owns the legal right to the content? Digital sampling in the music industry is now accepted as a legitimate component of the process of creating a piece of music. This only happened because an artist who uses the sample must acquire legal permission from the copyright holder. I’ve read the arguments against attempting to do something similar to cover content created by generative AI, which has either been published verbatim as original content or embedded within other content.

If generative AI outputs could be logged in a similar way to the ledger records in blockchain transactions I think this might provide a means of implementing the kind of digital forensic watermarking that needs to be applied to generative AI content. Unfortunately, this is way beyond my technical competence, but I hope that among our AI tech overlords there is one who will solve this in a truly ethical and responsible way--although I’m not holding my breath on that happening any time soon.

I recommend reading this short article from the MIT Technology Review, “Why watermarking AI-generated content won’t guarantee trust online.” It is a very good summary of the challenges to implementing the watermarking of generative AI content, especially the six questions or criteria that it must meet in order to be effective https://www.technologyreview.com/2023/08/0...i-trust-online/

Hi Luis,

A search engine (e.g., google) renders sites/web pages/content that match your query/question. Although the pattern search and ranking algorithms are open to bias, the delivered link/abstract simply points you to content that you can review with scrutiny.

However, generative AI attempts to deliver “factual information” from your query/question based upon what? [A] Based on foundational knowledge provided by whom. [B] Based on learned knowledge retrieved from where and validated by what (e.g., non-transparent scoring algorithms, content reviewers, etc.).

Regarding your statement, “Would there be any harm to the world?”

- Take forums such as PMC, whose value proposition is derived from the expert opinion of humans who provide feedback based on the intellectual evaluation of one’s vocational and life experience.

- Now take the Generative AI content that provides facts based on an aggregation of…who knows.

Doesn’t it seem likely, over time, that expert knowledge that is based on the whole of human experience will become diluted, that is, watered down, as aggregated AI-based knowledge pronounces and propagates its truth through believed non-biased means?

Bottom Line: Our “economy of knowledge” will become bankrupt when society demands “its truth” through means absent the intellectual human experience.