Skip to content
Idaho AI Classroom
All lessons
AI FluencyEvaluation and Critical Thinking9-12.AIF.ECT.6

Senses vs. Statistics

9–1050 minutes90 minutes (extend hands-on and discussion)

Standard quoted exactly

Give examples of how humans combine information from different bodily sensory inputs to make sense of the world, and compare that to how AI models learn about the world.

Example from the standards. LLMs largely learn from text written by humans, whereas humans use vision, olfactory, haptics, geolocation, etc., to ground their understanding of the world.

Student-friendly learning targets

  • I can give examples of humans combining sight, sound, smell, touch, and location to make a judgment.
  • I can compare that grounding to how an LLM mainly learns from human-written text.
  • I can explain why a text-only model can miss a shop, fireline, barn, or river reality that a person would notice.

Essential questions

  1. How do my senses work together when I decide something is safe, ripe, or off?
  2. What does an LLM actually 'know' if it mostly learned from text?
  3. When must an Idaho worker trust a body on the ground more than a generated description?

Objectives

  1. Students will list multiple human sensory inputs and show how they are combined in one grounded judgment.
  2. Students will compare that process to LLM learning from text, using the standards example.
  3. Students will complete a two-station activity: text-only description vs. multi-sense scenario cards.
  4. Students will write a comparison paragraph for a welder, nurse, rancher, river guide, or firefighter.
  5. Students will identify a school or work task that should not be decided from text statistics alone.

Key vocabulary

Sensory input
Information the body takes in through senses such as vision, hearing, smell, taste, touch, balance, and a sense of where you are.
Grounding
Tying words and ideas to the real world through experience, senses, and location, not only through other words.
Haptics
Touch and pressure information, such as the feel of a weld bead, a pulse, wet snow, or a too-hot pipe.
Olfactory
Smell information, such as smoke type, a sour silage pile, overheating electrical gear, or a fuel leak.
Geolocation
Knowing where you are in space: slope, aspect, river mile, GPS pin, or which end of the mill you are standing in.
Multimodal
Using more than one kind of signal (text, image, audio, sensor). Humans are multimodal by default; many LLMs are still text-first.
Large Language Model (LLM)
A model that largely learns from text written by humans and predicts likely next words, without a body in the world.
Embodiment
Having a body that senses and acts. Humans are embodied; a chatbot is not, even if it describes bodies well.

Teacher background

The standard is comparative and concrete. Humans combine bodily senses to make sense of the world. AI models, especially LLMs, largely learn from text written by humans. That is the official example, and it is enough. Do not turn the hour into a neuroscience lecture or a robotics unit. Ninth- and tenth-graders already know that smoke smells different from a burn pile, that a weld can look fine and feel wrong, that a patient can say they are fine while their color is not. Put those moments next to a paragraph an LLM could have written from forum posts. The model has statistics about the words 'wildfire,' 'pulse,' and 'ripe.' It has not stood in ash wind on the Payette, felt a thready pulse at St. Alphonsus, or walked a rutted pivot track at dawn. Some newer systems add images or sensors; still, the school-facing tools students meet are text-first, and even camera models lack smell, ache, and responsibility. Use a text-only station and a multi-sense card station. Paper packets carry the lab if you cannot set out objects. End with humility: language models are powerful readers of our writing, not stand-ins for a body on the ground.

Materials and prep

Materials

  • Projector and a short text-only wildfire or shop description
  • Printed fake-output packet: LLM-style text vs. human sense log
  • Optional object box for a safe haptic demo (sandpaper, cold pack, citrus peel) if policy allows; otherwise paper-only sense cards
  • Station cards: firefighter, ER nurse, dairy operator, welder, river guide, ski patroller
  • Chromebooks optional for a teacher-projected text generation; not required
  • Comparison chart handout

Before class

  • Print sense-log sheets with columns: Vision, Sound, Smell, Touch, Location, Judgment.
  • Write a text-only LLM paragraph that sounds expert on wildfire but misses wind on the face, grit in the teeth, and the smell of pitch.
  • If using objects, choose items with no allergens your nurse has flagged. Default to paper cards if unsure.
  • Save a screenshot of a text-only model answering 'How do you know the fire is making uphill run?' as the offline demo.

Instructional sequence

Mystery judgment

5 min
  1. Read a three-sentence text-only description of a shop hazard. Students vote safe or not safe.
  2. Reveal the missing haptic and olfactory clues from the card (hot pipe, ozone smell). Revote.
  3. Ask what the first vote was missing that a body would have had.

Bodies combine; models count words

10 min
  1. List human inputs: vision, hearing, smell, taste, touch/haptics, balance, geolocation. Students add one Idaho example each.
  2. Read the standards example aloud. Translate: LLMs largely learn from text; humans ground meaning with senses.
  3. Diagram two stacks: Human (many senses at once) vs. LLM (text in, likely text out). Mention cameras and sensors exist, but school chat tools are still text-first.
  4. Tie to responsibility: a river guide who 'reads' the Salmon is not doing the same thing as a model summarizing trip reports.

Fill the sense log together

8 min
  1. Project the dairy-parlor card: visual (cow's ear set), sound (off rhythm), smell, haptics (heat, swelling), location (which pen).
  2. Class completes one combined judgment: call the vet vs. wait.
  3. Contrast with an LLM paragraph that only restates generic 'monitor livestock health' advice.

One job, two columns

8 min
  1. Each student picks welder, nurse, firefighter, river guide, or ski patroller.
  2. Left column: three sensory combinations a human uses. Right column: what an LLM trained on text might say instead.
  3. They star one judgment they would not let a text model make alone.

Real-world examples

  • A firefighter combines smoke color, wind on the skin, radio traffic, and slope underfoot; a text model has trip reports and news stories.
  • An ER nurse combines skin color, breathing sound, a patient's joke, and a monitor number; the chart bot has only the typed notes.
  • A welder feels the puddle through the glove and hears the right sizzle; a generated procedure has sentences about amperage.
  • A river guide on the Salmon reads water noise, air temp, and the boat's vibration, not only a flow-rate number on a site.
  • A dairy operator smells a sour note and feels a hot quarter; a text alert may only flag a missing milk weight.

Hands-on activity

Text station vs. sense station

14 min
  1. Station A (text-only): pairs read the printed LLM wildfire paragraph and list what they still do not know.
  2. Station B (multi-sense cards): pairs complete a sense log for the same event using vision, olfactory, haptics, and geolocation clues.
  3. They write a four-sentence comparison that quotes the idea in the standards example.
  4. Teacher may project a live text-only answer if the district tool is up. Objects are optional. Paper cards are enough when labs are not.
  5. If movement is hard, both stations live in the packet as page A and page B. No student PII and no recording of classmates' bodies.

Discussion questions

  1. Which sense would you least want to give up on a fireline, in a shop, or in a clinic?
  2. Can adding a camera to a model replace smell and touch? Why or why not?
  3. Why might an LLM sound more certain than a tired human who was actually there?
  4. Where should school assignments still require a lab, a shop, or field notes instead of a generated description?
  5. How does geolocation change a weather or fire judgment in the mountains vs. the valley?

Differentiation

Support

  • Provide a labeled body diagram to tap when naming senses.
  • Give sentence frames for the comparison paragraph.
  • Allow the independent job column to be completed with the dairy example already modeled.

Challenge

  • Argue whether a multimodal model with cameras and GPS is closer to a human, and still list what it lacks (ache, fear, duty).
  • Write a shop-safety paragraph that a text model could not have grounded.
  • Connect to ECT.5: some hallucinations happen because words are ungrounded.

Multilingual learners

  • Teach olfactory, haptics, and geolocation with gestures and sketches before the stations.
  • Invite home-language sense words (smell of rain on dust, names for snow types) as evidence of human grounding.
  • Accept a bilingual sense log.

IEP / 504

  • Do not require handling objects. Cards and words are a full alternative.
  • For sensory sensitivities, skip smell items and use vision plus location only.
  • Allow a typed or scribed comparison. Keep stations quiet if needed.

Assessment

Formative

  • Warm-up revote and the class dairy log.
  • Two-column independent job sheet.
  • Station comparison sentences.

Summative

  • Paragraph: Give two human sensory-combination examples and compare them to how LLMs learn from text, using Idaho work or recreation.
  • Exit ticket: Finish 'Humans ground meaning with ___. LLMs largely learn from ___.' and name one task that needs a body on site.

Success criteria

  • I named more than one human sense working together.
  • I accurately described LLMs as largely text-trained, matching the standards example.
  • I showed a judgment a person could make that a text model would miss.
  • I did not claim the chatbot has a body or feelings.

Responsible use, ethics, and privacy

Responsible use

Do not ask a model to diagnose a real person or to 'be' a firefighter. Compare public, teacher-made texts. Students should not upload photos of their faces or homes to prove a sensory point.

Ethics

Ungrounded systems can still be useful as drafts. They become unethical when someone treats a paragraph as if it had been on the fireline, in the barn, or at the bedside. Human senses carry duty; generated text does not.

Privacy

FERPA: no student PII in unapproved tools. No student health details, no photos of identifiable students, no home GPS pins in prompts or packets. Sense logs use role cards, not real patients or family members. Unapproved tools stay off this lab.

Reflection

  1. When did my senses change a decision that words alone would have missed?
  2. What would an LLM get wrong about a place I know well?
  3. Which school task should stay embodied even though a paragraph could be generated?

Homework

On paper, sit somewhere safe for ten minutes (porch, shop door, kitchen). Log five sensory notes and one judgment they support. Then write four sentences comparing your log to what a text-only model might have written about that place. No photos of people. No AI tool required.

Closing

Re-read the standards example. Students tap the desk for haptics, point to the map for geolocation, and point to the packet for text. Collect logs. Next lesson: agents that take actions, still without a body and still in need of a human goal.

Extensions and cross-curricular links

Go further

  • 90-minute block: add a third station with a map and wind arrow so geolocation becomes visible, then rewrite the LLM paragraph with sensory gaps labeled.
  • Outdoor education or ag: field notes that require five senses before any generated summary is allowed.
  • Health sciences: vital signs plus 'how the patient looks and smells' as a human-in-the-loop check.
  • Physics / shop: vibration and heat as data humans feel before a sensor log is even opened.
Biology
Sensory systems and the brain combining signals, kept at a 9–10 level.
English Language Arts
Imagery in writing vs. ungrounded generated description.
Skilled and Technical Sciences
Why a procedure on paper is not a substitute for feel, sound, and smell in a shop.
Geography / Outdoor Ed
Geolocation, aspect, and microclimate as bodily and map knowledge a text model only fakes.