Accessibility and ableism concerns regarding the AI participation policy

I recently considered contributing to a technical discussion on this forum. I had spent quite some time developing my arguments, researching relevant approaches to programming language design, and thinking about how I could share something useful with the Python community.

However, after reading the community guidelines concerning AI-generated content, I felt genuinely hurt and disappointed, to the point that I reconsidered participating altogether.

I have motor difficulties and chronic pain that significantly affect my ability to write and perform many everyday tasks. I rely on assistive tools, including generative AI, to help me communicate my thoughts, organize arguments, and produce written material.

This isn’t a temporary inconvenience, nor is it something I can simply overcome by trying harder. I need ongoing assistance because of my disabilities.

With that context, I would like to explain why two statements in the guidelines were particularly upsetting to read.

Write your posts in your own words and verify any information yourself that you share with others.

I understand that the intention may be to encourage intellectual responsibility, originality, and independent verification. I agree with those principles.

But the phrase “write your posts in your own words” is particularly painful to encounter when your disability makes producing written text without assistance physically difficult or impossible.

I can develop my own arguments. I can critically evaluate evidence, identify mistakes, challenge suggestions, and take responsibility for what I communicate.

What I cannot always do is physically produce and repeatedly revise the written text needed to express those ideas without assistance.

My ability to write independently is not a measure of my intellectual independence.

When a community makes unaided writing appear to be a prerequisite for participation, the result can be ableist, regardless of whether exclusion was the intention.

The second statement concerns me even more:

Content that is generated by AI tools will be removed.

I understand why a technical community would want to discourage automatically generated, inaccurate, misleading, or low-effort contributions.

But a blanket statement like this fails to distinguish between someone asking an AI to generate opinions or technical claims they don’t understand and someone using AI as an assistive technology to communicate their own ideas.

For me, AI isn’t simply a shortcut or a convenience. It’s one of the tools that makes certain forms of communication and participation possible.

Reading that AI-generated content will be removed, without an explicit recognition of accessibility use cases, makes me feel that the assistance I depend on could be grounds for excluding my contributions, regardless of their intellectual substance.

That is profoundly discouraging, especially coming from a community as large, diverse, and influential as Python’s.

I don’t expect every community member to understand the circumstances of every disability. But I do expect accessibility to be considered when establishing rules that determine who can participate and how.

I also don’t think people with disabilities should have to disclose intimate medical information, prove that they cannot use another tool, or justify their financial circumstances simply to be allowed to communicate using the assistance they need.

Accessibility shouldn’t depend on demonstrating that every conceivable alternative has been exhausted.

I want to emphasize that I’m not asking for lower standards of technical accuracy, originality, or intellectual responsibility.

I would expect my contributions to be evaluated critically, just like anyone else’s. If I make an incorrect technical claim, I should be expected to correct it. If I misunderstand an argument, I should be prepared to discuss it.

But those standards should concern what I contribute and whether I understand and take responsibility for it, not whether I was physically capable of producing every sentence without assistance.

I appreciate that there have been subsequent clarifications about responsible AI use. However, the published wording still creates an accessibility barrier, or at the very least a strong perception of one, for people in circumstances like mine.

I would therefore ask the community and its moderators to consider explicitly distinguishing between:

  • AI-generated contributions submitted without meaningful human understanding, review, or intellectual involvement.

  • AI-assisted communication in which the contributor develops the ideas, reviews the resulting text, and takes responsibility for its content, including when AI is used as an accessibility aid.

I would also encourage reviewing the wording of the guidelines themselves. Even if moderators already intend to accommodate assistive uses of AI, people encountering the rules for the first time shouldn’t have to discover that through an individual appeal or an extended discussion.

I genuinely wanted to contribute to this community. I wasn’t looking for recognition or special treatment; I simply thought I had something worthwhile to add to a technical conversation.

Discovering language that seemed to make the assistance I depend on incompatible with participation was unexpectedly painful.

I hope the community can recognize that accessibility is not an exception to meaningful participation. It is part of what makes meaningful participation possible for people with different abilities and circumstances.

I would appreciate not only clarification of the current policy but also a reconsideration of how its wording and enforcement may affect contributors who rely on assistive technologies.

I hope that this concern can be received with the same openness and willingness to reconsider assumptions that I would expect from any constructive technical discussion.

And if we go with the “Content that is generated by AI tools will be removed.”, go ahead and remove this post, if that is, indeed the metric we’re going to use because this was generated using AI tools.

6 Likes

I don’t think there was an intent to discourage assistive technologies. That use is welcome.

AFAICT the AI policy aimed to avoid swamping human participants with content that had little human involvement on the front-end. AI agents can completely swamp human reviewers. There needs to be a balance of human involvement on both the sending and receiving ends.

1 Like

First, I want to clarify that I don’t believe there was any ill intent behind the rule, and I appreciate you confirming that assistive technologies are welcome.

My concern is that the policy doesn’t seem to consider AI as a potential assistive technology, at least not explicitly.

And if AI-assisted communication is indeed permitted for accessibility purposes, that raises another question: should every member of this community who relies on AI because of a disability have to disclose their disability and request permission before participating?

I don’t think that should be necessary. People shouldn’t have to disclose personal medical information or justify their accessibility needs simply to contribute to a technical discussion.

I understand and agree with the concern about overwhelming human participants with automatically generated content that has little meaningful human involvement. But I think that distinction should be reflected in the policy itself.

The relevant distinction should be between contributions with meaningful human intellectual involvement and responsibility, and those without it—not between text produced with or without AI assistance.

A policy can be well-intentioned and still inadvertently discourage participation. That’s what happened to me: I read the guidelines and genuinely believed that the tools I depend on made my contributions unwelcome.

I appreciate your clarification, but I think it would be valuable to make that distinction explicit in the guidelines, so that others don’t have to go through the same experience or disclose their disabilities to find out whether they’re allowed to participate.

Would you mind posting your initial topic, I’m really interested on what you can offer to the community with a new perspective.

1 Like

@ravenhill Welcome to the community. I understand your very human distress, but I think it misplaced. I think it would have be better is you said your interpretation of two sentences distresses you if correct, but then asked if you were correct.

My guess is that you have little idea of the destructiveness of what the new policy is responding to (which is not particularly your fault). Example: human sets up GH account, gives agent credentials, and prompts something like “Find bugs in CPython, file issues, and add PRs fixing the bugs.” In about 10 minutes, agent opens about 20 issues involving as many modules, with 20 corresponding PRs. In one case, the issues honestly included “This report has not been read by a human.” This is what “Generated by AI tools” means. Maybe we should clarify “only by AI tools, without meaningful and responsible human input.” But if we do, I expect people will nitpick the addition to squeeze stuff in.

‘Write’ does not mean ‘type’ (in English). I ‘write’ texts on my phone by speaking them. The draft I read explicitly mentioned that machine translation to English text is permitted. However you input your thoughts is fine.

4 Likes

Your use of AI in writing that post didn’t bother me. Its needless enormous length did.

7 Likes

For reference, here is the topic that explained the policy modifications: Clarified policy on AI / LLM use. The usage as assistance was discussed, and is fine.

That was my understanding as well. I’m sure some people use voice recognition software (or other kinds of input mechanisms) to “write” their messages.

But besides, I’m also genuinely curious about the physical difference between “writing” a message directly (perhaps using voice recognition software) vs. “writing” instructions for, and interacting with, a LLM. What makes the latter more accessible? (is a LLM more forgiving?)

1 Like

IMO, the forum guidelines should be better aligned with what we have in the CPython repo as a guideline: Guidelines for using AI tools.

Use of AI tooling is not forbidden in CPython, so why should it be forbidden in this forum ? There are good reasons to set up guidelines for their use, but outright stating that “Content that is generated by AI tools will be removed” is not in line with how we operate in CPython today.

Related to this, I do believe that the forum community guidelines should be reviewed and accepted by the PSF Board before being put into action, with the usual open discussion round when making changes (similar to the process we have for PEPs in CPython). Otherwise, we’ll end up with a fractioned view of how things should be done, what is acceptable, what not, etc. which is not helpful.

Since I couldn’t find an alias for the board, I’ll ping our chair directly: @jezdez
Hope you don’t mind, Jannis.

(And yes, if this side line picks up, we should split out that discussion into a separate topic)

3 Likes

For code, there is an objective critery of quality. Does it fix the issue? Does it corretly handle corner cases? You run the tests and they say if it works or no. This is why the current AI is pretty good in writing code for given specifications.

But forum is a different thing. The written message should be read and evaluated by huments, not machine. Excessive verbosity common for LLM generated texts harms readability. Each readers needs to spend efforts to process it. Human time is more expensive resource than machine time.

5 Likes

One very unfortunate aspect of the current trends in AI development is that basically everything is called “AI”. There’s a huge difference between using a neural net for voice dictation (where it’s doing the exact same job as speech-to-text has done for decades, but now with increased accuracy and no need to train it on specific words and phrases) and asking an LLM to write your post for you… but both of them are touted as “the power of AI”.

If every word in your post was composed by you, but rather than putting fingers to keyboard, you dictate to a computer, then the post was yours, and that’s completely fine in my book. But if you write a brief set of points and then ask an LLM to flesh it out into a convincing post, that’s what creates the problems. (And Vodafone, I’m looking at you, spewing AI-generated chat messages at me even when I’m notionally talking to a real person. Not impressed.)

6 Likes

Agreed. That’s why I said “There are good reasons to set up guidelines for their use”.

Maybe, but this is something that is difficult to inerpret correctly. It has been very clear since the earliest version of ChatGPT that many humans will dump lengthy LLM output while claiming (and likely believing) that there was much more human intellectual involvement than there really was. The guidelines were written after plenty of actual experience in these forums of the problems that this caused and I think that the wording is carefully chosen based on that experience.

2 Likes

I answer this message particularly because I do share your concerns, and I share many of them, but it’s kind of an answer to (at least most) of the comments.

First, thank you to everyone who clarified the policy. I’m genuinely glad to know that assistive uses of AI are welcome.

I also want to emphasize that I’m well aware of the problems caused by irresponsible AI use. As both a researcher and an educator, these are issues I actively engage with, including teaching students how to use AI critically and responsibly. I appreciate the community’s efforts to address them.

Regarding suggestions to use other assistive technologies, I don’t think people with disabilities should have to justify their choice of tools or explain everything they’ve tried. AI significantly reduces the physical effort required for me to communicate, and among the options available to me, it’s what has worked best.

My concern was never about obtaining individual permission to use AI, but about how categorically the guidelines are worded.

Statements such as “Write your posts in your own words” and “Content that is generated by AI tools will be removed” (specially when seen together) can discourage people who depend on AI for accessibility, even when their contributions involve substantial human thought, review, and responsibility.

I’m glad the intended policy is more nuanced, but I believe that nuance should be reflected in the guidelines themselves. Nobody should have to disclose a disability or request clarification just to know whether their assistive tools are welcome.

AI isn’t always a shortcut or a substitute for understanding. Sometimes it’s what makes participation possible.

That’s all I hoped to bring attention to, and I hope it can be considered in future revisions of the guidelines.

3 Likes

I do agree that the current guidelines feel too strict.

But at the same time, some AI posts are such bloat.

If you use AI for speach-to-text I 100% support that. And it is bad that the guidelines (read literally) forbid that. If you use AI to navigate the website it is the same to me. But that is all that is required by your disability, or is more required?

I cannot tell for sure whether your post was dicted by yourself, or if you gave an AI some keywords and broken sentences and told it to make a post. I can tell you that this feels like that latter, like an AI diluted the meaning in your post in order to make it smoother. And that’s not good. Because then us as readers have to put in mental work to extract the meaning again.
Apologies if you genuinely speak like this, because then I’m guilty of yet another form of ableism.

9 Likes

Agreed. Give us the unpolished version and let us read your original intent. That’s not ableism; that’s seeing what you truly wrote, not what some AI mutated into its own form.

7 Likes

He litterally said:

That’s not just a “polishing” problem. From what I understand, he’s using tools to express his original intent.

That’s what’s unclear. If the tools used are dictation tools, then the original intent is the dictated words, and there’s nothing wrong with a neural net (AI) being used for the transcription.

But if it’s not, then what IS the original intent being expressed in? Is it point form text? Is it something else? How are we supposed to say whether the use of AI is reasonable or not with such vagueness?

5 Likes