Draft AI Policy for Medley Interlisp Project
Larry Masinter
### Medley Interlisp Project: AI & Contribution Policy
The use of AI and agents is a current controversy in the open source community and also within the Medley Interlisp project. The use of AI is novel, and in many circumstances the AI agents do a poor job. We want to understand the applicability of using AI in software preservation.
I then asked Gemini to rewrite this, and got the following:
Here is a revised, polished version of the policy that tightens the phrasing while fully preserving your points about software preservation research, attribution, and licensing:
Medley Interlisp Project: AI & Contribution Policy
The role of generative AI and automated agents is a subject of ongoing debate within both the wider open-source community and the Medley Interlisp project. While AI technologies are novel, they frequently produce suboptimal results. However, as an organization dedicated to historical continuity, we are interested in evaluating and understanding the practical applicability of AI tools specifically within the field of software preservation.
To support this research and maintain project integrity, all submissions and contributions—including source code, documentation, test cases, and tracking issues—must adhere to the following criteria:
Authenticity or Disclosure: Every contribution must be either the original creation of the contributor or work whose precise origin and development process are fully documented and annotated within the submission.
AI Attribution: If a generative AI tool is utilized, the submission must explicitly disclose the identity of the tool used, the date it was accessed, the exact prompts provided, and any operational constraints or parameters applied during its execution.
Licensing & Copyright: Unless an exception is explicitly granted, all contributed work must be labeled with copyright Interlisp.org and made available for publication under the terms of the specific repository's open-source LICENSE to which it is submitted.
Herb Jellinek
In our meeting yesterday (25 May 2026), I recall your saying that you're having second thoughts regarding copyright clause in the ultimate version below. How would you revise it?
Copyright aside, none of the points you and Gemini have written below address my main concern: AI attribution is fine and good, but if a contributor didn't write the content (code, docs, etc.) they're submitting, they're less likely to be able to explain it or maintain it. In the era before generative AI agents, the burden of demonstrating and explaining the value of contributed artifacts was on the submitters. Now the burden of evaluation and consideration is shifting to the recipients. The proportions are out of balance.
Regarding student work, my position is simple: The main "output" I'm looking for is a better-prepared student, so I want to see how well the student understood the problem and how they solved it, or tried to. I don't care how well you prompt an LLM. The code, documentation, whatever the student is secondary, necessary but insufficient.
How would you integrate those concerns in a larger AI policy?
Herb
--
You received this message because you are subscribed to the Google Groups "Medley Interlisp core" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lispcore+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/lispcore/CAKq15vd%2B068jJsw6Yr%2BwAgzfvj5eU4waPdQVpoSZ40ed4J62gw%40mail.gmail.com.
Herb Jellinek
Larry,
In our meeting yesterday (25 May 2026), I recall your saying that you're having second thoughts regarding the copyright clause in the ultimate version below. How would you revise it?
Copyright aside, none of the points you and Gemini have written below address my main concern: AI attribution is fine and good, but if a contributor didn't write the content (code, docs, etc.) they're submitting, they're less likely to be able to explain it or maintain it. In the era before generative AI agents, the burden of demonstrating and explaining the value of contributed artifacts was on the submitters. Now the burden of evaluation and consideration is shifting to
Regarding student work, my position is simple: The main "output" I'm looking for is a better-prepared student, so I want to see how well the student understood the problem and how they solved it, or tried to. I don't care how well you prompt an LLM. The code, documentation, whatever the student produced is actually secondary, necessary but insufficient.
How would you integrate those concerns in a larger AI policy?
Herb
To view this discussion visit https://groups.google.com/d/msgid/lispcore/43085785-e0e3-4e52-a9f6-5166b2ebb199%40newscenter.com.
wstumbo
Herb,
You're asking the right question. I'm not sure I have a great answer for you. I suspect the answer isn't something that ends up in a policy but finds its way into a work process.
Some ideas:
- Definition of Done - I know we've used this before with some of our issues. Maybe, we need to put a bit of a twist on them. For larger submissions were we have people doing work for us, make this the first part of the work product. The individual needs to create the Definition of Done. We could use this as a more rigorous means of defining the work product and setting expectations.
- Required Explanations - What was your rationale for ... (unfortunately the exact questions will require some knowledge of the work product and some level of investment. Maybe the Definition of Done can point towards the right questions to ask. Another approach is "What part of the submission are you most confident or least confident in and why?" Can we, without an exceedingly large amount of effort, come up with some questions that will require human synthesis to answer?
- Self-review - require contributor to leave several ( n > 2) critical comments explaining implementation details. Easier for code then written documents.
None of this solves the problem. At best, it moves it upstream and, ideally, allows us to put the right guardrails in place early and set expectations. But, there is still effort.
And, I'll be honest I don't know how much of this work you've already tried or have in place.
Full Disclosure: I asked Gemini for feedback on how open source communities tried to ensure reviewers weren't inundated by AI slop. It focused on code submissions and the PR process. From there, I pulled out what I thought might be useful and refocused it for our specific scenario.
-- Bill Stumbo wst...@charter.net
To view this discussion visit https://groups.google.com/d/msgid/lispcore/74dc2655-881f-4dbe-87dc-474f7b52fb1c%40newscenter.com.