A&S AI Literacy Module
33 minutes · self-paced · no prerequisites
AI systems run in data centers. Those facilities can use water directly for cooling and indirectly through electricity generation—but a single number cannot describe every workload or every community. You will use three peer-reviewed studies to evaluate a fictional proposal in Central New York.
Instructor
Associate Professor, Department of Earth and Environmental Sciences, Syracuse University. Professor Wen’s teaching and research connect water quality, hydrogeochemistry, data science, and machine learning in coupled human-natural systems.
Distinguish withdrawal from consumption, direct from indirect use, and quantity from quality.
Do not treat a projection, a broad assessment, and screening-level risks as equivalent.
Evaluate a claim and identify evidence still needed for one proposed site.
Opening scenario · 2 minutes
This scenario is fictional. It is designed to show why national estimates cannot answer a local water question on their own.
A company proposes an AI data center near a Central New York community. Its fact sheet says the facility will be “water-smart.” A social-media post replies that “AI data centers drain drinking-water supplies and poison fish.” Local officials ask: What can we conclude now, and what must we investigate?
Professor Wen’s water-scientist lens
A water footprint has a location, a season, a source, a return flow, and a receiving environment. The same facility design can have different consequences in different watersheds.
Video 1 · 7 minutes
Four distinctions keep a water claim from collapsing different processes into one number.
Synthetic narration disclosure: This video uses an authorized local clone of Professor Tao Wen’s voice, generated with Chatterbox and automatically watermarked. Read the narration notice.
Open the full text transcript. Synchronized English captions are available from the video controls.
Water taken from a river, lake, aquifer, or utility—even if some is later returned.
The portion not returned to the same water source, often because it evaporates or is incorporated into a product.
How much water is withdrawn or consumed, when, and from where.
Properties such as temperature, salts, treatment chemicals, and other constituents in a discharge.
Original concept diagram
An AI training or inference workload branches into two pathways. The direct pathway reaches facility cooling. The indirect pathway reaches the electricity supply. Both pathways then pass through local context: the electricity grid, climate, source water, and season.
Professor Wen’s water-scientist lens
Ask whether a number describes withdrawal or consumption and whether it includes direct, indirect, or total water use.
Checkpoint · 2 minutes
Sort all four examples. Feedback explains how the distinctions overlap.
Video 2 · 7 minutes
Each paper answers a different kind of question. Treat its study design as part of the evidence.
Synthetic narration disclosure: This video uses an authorized local clone of Professor Tao Wen’s voice, generated with Chatterbox and automatically watermarked. Read the narration notice.
Open the full text transcript. Synchronized English captions are available from the video controls.
Projected annual U.S. AI-server water use: 731–1,125 million m³.
Limit: a national model projection—not facility meter readings.
Estimated water use across data-center workloads varied by more than 10,000-fold.
Server efficiency, cooling, the electricity grid, climate, and utilization all matter.
Limit: data-center workloads broadly—not one commercial AI service.
Data centers: an emerging threat to freshwater biodiversity in the United States
Identified potential pathways involving thermal effects, lower-quality discharge, water stress, entrainment, and altered flows.
Limit: screening-level risks—not demonstrated harm at a particular site.
Professor Wen’s water-scientist lens
A national projection can frame scale. A screening study can flag pathways. Neither replaces local flow records, chemistry, temperature, permits, or ecological conditions.
Evidence activity · 9–10 minutes
Classify three claims, justify each choice, then write a 50–75-word recommendation grounded in at least two research evidence cards and one concrete question for the developer.
Supporting AI-literacy method
For each claim: identify one place it could be wrong, one thing to check, and one source that could resolve it.
The evidence backs the main idea, but only when you keep an important limiting phrase—for example, “in this modeled U.S. projection.”
The claim may fit some situations, but it sounds too general because important conditions can change the answer.
The studies show what could happen, but information from this proposed facility and its local water system is needed before deciding.
Community recommendation
Use 50–75 words. Cite at least two research evidence cards, distinguish one system boundary, and turn one missing local input into a concrete question for the developer. The AAAS guide helps frame that question but does not count as one of the two research cards.
In plain language: a system boundary tells the reader what an estimate includes and leaves out. You might distinguish on-site direct water from electricity-related indirect water, or a national projection from measurements at this proposed site.
Together, the four entries should total 50–75 words.
| Criterion | Strong | Developing | Revise |
|---|---|---|---|
| Evidence fit | Uses at least two cards accurately. | Uses evidence but blurs one limit. | Makes claims the cards do not support. |
| System boundary—what is counted | Names direct/indirect or national/local limits. | Mentions water use without saying what is included. | Treats one number as universal. |
| Calibrated uncertainty | States one missing local input and a next step. | Signals uncertainty but not what would resolve it. | Claims certain benefit or harm. |
Keyboard users can focus this table region and scroll horizontally if needed.
Five-question quiz · 4–5 minutes
Each question assesses a different concept from the module.
Module complete
A useful water claim states what was measured or modeled, where, over what system boundary, and what evidence is still missing.
Direct cooling and indirect electricity-related water use can point to different interventions.
Projections estimate scale; risk screens identify pathways; neither proves local harm.
Cooling, source water, flow, chemistry, temperature, grid mix, and permit conditions turn a headline into an evaluation.
Professor Wen’s water-scientist lens
“What evidence would change our conclusion?” is not a retreat from expertise. It is how watershed context, hydrogeochemistry, and uncertainty become part of a responsible decision.
Your responses stay in this browser. They are not sent to an external service.