AI Creator Ethics

The Tradeoff Compass

Scenario:
Artifact Domain
The output and its audience
Creator Domain
You and your creative process
HUMAN-LED BALANCED AI-AUGMENTED 50 AI SUITABILITY
Balanced Partnership
AI as a thoughtful collaborator. Use it for efficiency, but inject your judgment and voice.
Shannon Information Lens
AI Token Generation (Entropy Space)
Human Refinement (Signal Injection)
A balanced exchange: AI provides the raw probability space, while human judgment selectively injects meaning.
Facilitator Discussion Prompts
Select a scenario above to see discussion prompts.
The Shannon Information Lens
A mental model — not technically precise, but hopefully useful — for understanding AI as a communication problem and the human role as faithful decoder
1. Information Source
The World's Knowledge
Trillions of words, images, and patterns from human civilization — the internet, books, code, conversations. High entropy. Massive redundancy. Raw and unorganized.
Energy + Compute
2. Lossy Compression
The LLM as Compressed Knowledge
Training uses enormous energy to compress the world's information into a few trillion parameters. Useful patterns are preserved — the information becomes more organized. But the compression is lossy — misalignments to reality, inherited biases, and hallucinated patterns are baked in.
Prompt = Signal Injection
3. The Human Decoder
You — The Faithful Decoder
Your role is to inject the right signal — context, judgment, domain expertise, intent — to decode the compressed information into a useful artifact. You are the error-correction layer. The quality of the output depends on your ability to be a faithful decoder: honest about what the model gets wrong, and deliberate about what you add.
Artifact Emerges
4. The Artifact
Useful Output in the World
A speech, a design, a document, a piece of code. Its quality reflects both the compressed knowledge of the model and the signal quality of the human decoder.
Energy Shifts — It Doesn't Disappear
Creating something useful from scratch has always required energy — human effort spent on research, drafting, revision, expertise. Training an LLM reorganizes vast amounts of information into compressed, lower-entropy parameters, but this takes enormous energy in the form of compute. The human energy of creation doesn't vanish; it's redistributed. Part moves upstream to the energy spent in training. The rest stays with you — the energy to prompt well, verify the output, and inject the meaning that the model can't.
🔍
Lossy and Misaligned Compression Has Consequences
No compression is perfect. The model's parameters encode useful patterns, but also inherit the biases, gaps, and contradictions of their training data. What we call hallucinations are the probabilistic consequence of this — the model generating statistically plausible output that doesn't align with reality, because its compressed representation of the world is incomplete and shaped by its training context. This is why the human decoder matters: someone must recognize where the compression is misaligned and correct for it.
📡
The Prompt as Signal
In Shannon's model, a message travels through a noisy channel. Your prompt is the signal you inject to extract meaning from the compressed knowledge. A vague prompt gets noisy output — the model falls back on statistical patterns. A precise prompt with domain context, constraints, and intent produces high-fidelity output. The signal-to-noise ratio of your prompt directly determines the quality of the artifact.
🎯
Faithful Decoding Is the Ethical Core
The ethical question isn't "did you use AI?" — it's "were you a faithful decoder?" Did you verify the output against reality? Did you inject your own knowledge, voice, and judgment? Did you catch the compression artifacts? The Compass tool helps you think about how much human decoding each scenario demands — because a eulogy requires different fidelity than a sprinkler part.
The Factor Universe
Every scenario on the Compass tab highlights 5 factors from each domain — because context determines which tradeoffs matter most. Here's the full set of considerations when deciding how much to lean on AI.
Artifact Domain — factors about the output and its audience
Creator Domain — factors about you and your process
Favors AI — higher values suggest AI is appropriate Favors Human — higher values suggest human effort needed Contextual — direction depends on the situation
Showing 5 artifact + 5 creator factors for this scenario cycle scenarios