With all this talk about AI chat bots, I wonder… do I need a new title?

A colleague recently told me that GitHub Copilot is already responsible for generating nearly 50% of all code on GitHub but it is used by less than 10% of GitHub users. The implication seems to be that a very small number of developers are responsible for a majority of the code produced. (I don’t know if these are the exact numbers, and it’s probably true regardless of whether developers are using AI tools to assist them.) The message to developers is that using AI-enhanced coding tools will make them more productive. It got me thinking about how we know whether some artifact is generated by a human, a bot, or some hybrid of the two.

This led me to think about the question not asked, which is whether it’s important to know the provenance of an artifact. Do you really need to know whether I wrote this post myself? Does the possibility that I asked ChatGPT to write it for me somehow change its value? I think most people would answer “yes” to both questions, even if they cannot say precisely why? (Some of the reasons for that may be addressed in my previous post on banning AI).

I’m going to say up front, that I don’t have the answer to that question. But if we are going to answer the question whether AI-generated artifacts are somehow worth less than human-generated artifacts, then we must do a better job telling them apart.

One approach to the provenance question is to develop AI detectors, such the one as envisioned by “Blade Runner” (only less dramatic) to help us spot the human authors from the AI authors. This seems to be the approach educators are taking to identify which students ask ChatGPT to do their homework for them.

The problem is that AI detectors promote an arms between the AIs and the detectors where better AI evade detectors and better detectors are always needed. This might be good for the companies that want to monetize their latest contributions to the field. But in the long run this really doesn’t do much to address the problem of AI-generated work being passed-off as human-generated.

In Isaac Asimov’s future societies, robots used the title “R.” to indicate that they were not human. Taking his cue, we could start using the following authorship identification for our work:

1. If I’m exclusively the author with no AI input, then I simply sign the work “David Chassin”.

2. If the work is hybrid, meaning that I wrote some it and an AI tool only provided a small fraction of input prompted by me, then I sign the work “H. David Chassin”, where the “H” stands for “hybrid” and I provide a footnote indicating which AI bot was used to support the work.

3. If the work is completely AI generated, then I provide a footnote stating which AI and what query generated the artifact, and I sign the work “B. David Chassin”, where “B” stands for “bot”.

Admittedly this is an honor system approach, but it could go a long way to addressing the question of whether AI-generated artifacts have a different worth than human-generated artifacts.