Hacker's Guide to Ethical AI

unitaryHACK Hacker’s Guide to Ethical AI Contributions

At unitaryHACK, we believe AI is a powerful co-pilot, not an autopilot. Quantum computing is hard, and while Large Language Models (LLMs) like ChatGPT, Claude, or Copilot can help you understand a new library, they can also confidently generate code that violates the laws of physics.

To keep the ecosystem healthy and respect our maintainers’ time, we follow a Human-in-the-Loop policy.

🟢 The Green Light: High-Value AI Collaboration

These are tasks where AI adds value without burdening maintainers. Note: All AI-assisted work must still be disclosed and verified.

• Documenting the Complex: Ask an LLM to help you draft a docstring for a complex function, then edit it to ensure the quantum terminology (e.g., "non-Clifford gate count" or "basis state") is precise.
• Boilerplate Unit Tests: Use AI to generate the structural code for a test suite. Crucial: You must then run those tests locally and verify they actually pass and cover the edge cases.
• Explaining Errors: Paste a cryptic stack trace into an LLM to help you brainstorm why a circuit might be failing. Use that insight to write your own fix.
• Code Refactoring: Use AI to suggest more "Pythonic" ways to write a loop or handle data, provided the underlying quantum logic remains untouched.

🔴 The Red Light: "AI Slop" & Cheating

Submitting these will likely result in an immediate rejection and potential disqualification.

• The "Blind" PR: Copy-pasting an issue description into an LLM and submitting the output without ever running the code or understanding how it works.
• Mass-Submissions: Using automation to "spray and pray" AI-generated fixes across dozens of repositories.
• Hallucinated Logic: Submitting PRs that use functions or libraries that don't exist, or that "fix" a bug by simply deleting the failing test.
• Undisclosed AI Use: Passing off AI-generated code or documentation as your own original work. Honesty is a requirement for bounty eligibility.

🛠 The Ethical Workflow: 4 Steps to a Winning PR

If you use AI to help with your contribution, follow this "Human-in-the-Loop" checklist before you hit Submit:

1. Understand: Could I explain this code to a maintainer in a live interview? If no, don't submit it.
2. Verify: Did I run pytest (or the project's equivalent) locally? Do the tests pass?
3. Refine: Did I remove the "AI fluff"? LLMs often add unnecessary comments or verbose explanations. Strip it down to the essentials.
4. Disclose: Add a note in your PR: "I used [Tool Name] to help brainstorm the logic for [Section X], which I then manually verified and tested."

A Note on Burnout: Maintainers are the backbone of the quantum world. When you submit unverified AI code, you are asking a human to do the "hard work" of debugging your bot. Be a contributor, not a burden.

Disclosure: This Guide was built in collaboration with an LLM. We developed the language around our human-in-the-loop policy, building on feedback from maintainers, collaborators, and hackers from previous hackathons. Gemini 3 helped put all of this text into a step-by-step guide, which was then reviewed by multiple UF staff members and previous unitaryHACK maintainers before being shared with the public.