Evidence Activity: Advise a Community About an AI Data Center
Estimated time: 9–10 minutes
Scenario
A company proposes an AI data center in a fictional Central New York community. Supporters call its water use efficient. Critics predict harm to drinking water and fish. The planning board asks you to separate what the current evidence supports from what must be investigated locally.
Use the four cards in the evidence pack. You are evaluating evidence, not deciding whether data centers are simply “good” or “bad.”
Part 1: Evaluate Three Claims — about 4 minutes
Choose the best label for each claim:
- Supported with qualifier: the evidence backs the main idea, but you must keep an important limiting phrase such as “in this modeled U.S. projection.”
- Overstated / context-dependent: the claim may fit some situations, but it sounds too general because important conditions can change the answer.
- Needs site-specific evidence: the studies show what could happen, but information from this proposed facility and its local water system is needed before deciding.
Then cite at least one card and explain your choice in one or two sentences.
| Claim | Best label | Card and justification |
|---|---|---|
| 1. “Every AI prompt consumes approximately the same amount of water.” | ||
| 2. “In one U.S. projection, most AI-server water use is indirect through electricity generation.” | ||
| 3. “This proposed local data center will damage drinking water and fish populations.” |
Part 2: Community Recommendation — about 4 minutes
Write 50–75 words advising the planning board on its next step. Your recommendation must:
- cite at least two evidence cards using “Card 2,” “Card 3,” or “Card 4”;
- distinguish at least one system boundary, such as direct/indirect, facility/grid, projection/local measurement, or quantity/quality;
- name at least one missing local input and phrase it as a concrete question for the developer or operator;
- use calibrated language such as “modeled,” “suggests,” “potential risk,” “could,” or “cannot yet determine.”
A system boundary is simply the line a study draws around what it counts. A narrow boundary might include only direct water used at the facility; a wider one might also include indirect water used to generate electricity. A national projection also has a different boundary from measurements at one proposed site.
The AAAS EPI Center community guide can help you frame the local question. It is a practitioner guide, not a research evidence card, so it does not replace either of your two paper-card citations.
Write your recommendation here:
Equivalent bullet-list option
Instead of a paragraph, submit four concise bullets:
- Recommendation: the next step and one missing local input phrased as a question for the developer.
- Evidence 1: what one card supports.
- Evidence 2: what a second card supports.
- Boundary: what the studies count or leave out.
Part 3: One-Minute Check
Before submitting, underline or mark:
- your two card citations;
- the words showing uncertainty or evidence limits;
- the local input you requested.
After any AI-generated explanation, also identify one place it could be wrong, one fact you would check, and one source that could support that check.
Three-Criterion Rubric
| Criterion | 3 — Evidence-ready | 2 — Developing | 1 — Needs revision |
|---|---|---|---|
| Evidence fit | Uses at least two cards accurately and keeps each claim within the card's scope | Uses two cards, but one connection is vague or slightly misstates scope | Uses fewer than two cards or materially misstates the evidence |
| System-boundary reasoning | Correctly distinguishes a relevant boundary and explains why it matters to the recommendation | Names a relevant boundary without explaining its importance | Treats unlike quantities or evidence types as interchangeable |
| Calibrated uncertainty | Separates potential or modeled effects from local conclusions and requests a decision-relevant local input | Uses cautious wording or names a local input, but not both clearly | Claims certainty about local benefit or harm without local evidence |
Prose style and grammar are not scored unless they prevent the meaning from being understood.