Why Science? | Smart Systems Institute

A smart system works with people, objects and norms in a place. The same model can succeed in one setting and fail in another.

Start with the observable system. An operating-context card records users, tasks, environment, variability and unavailable information. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how situated-AI literacy becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to compare a home robot task in a quiet room and a crowded kitchen. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. General AI language does not replace a defined situation. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Next chapter

Robots and wearables have size, mass, sensors, batteries and contact surfaces. Physical design shapes what intelligence can do safely.

Follow the mechanism before the headline. Co-design links body geometry, computation and task rather than treating hardware as a container. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how embodied-systems reasoning becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to sketch two robot bodies for the same reaching task and compare limits. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. A capable algorithm cannot overcome every physical constraint. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Cameras, microphones, touch and motion sensors each miss or distort information. Lighting, occlusion and noise change reliability.

Make the comparison explicit. Sensor models describe range, resolution, latency and failure, with redundancy chosen around consequences. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how smart-sensing literacy becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to create a sensor table for finding a cup under changing light. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. More sensors can add conflict, calibration burden and privacy exposure. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Computer vision and other perception systems infer objects, pose and activity from measurements. Confidence depends on training coverage and scene conditions.

Keep the evidence chain visible. Evaluation separates detection, localisation and tracking across representative and difficult cases. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how robot-perception judgement becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to design an occlusion test for a household object detector. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. A bounding box is not proof that a system understands intent or risk. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

A robot must consider goals, obstacles, uncertain state and future human movement. Plans need fallback when assumptions fail.

Test the boundary as carefully as the main effect. Belief-state and receding-horizon methods update actions as new observations arrive. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how robot-planning literacy becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to build a three-step plan with an observation that forces replanning. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. A shortest path is not automatically the safest or most legible path. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Controllers convert desired motion into actuator commands while compensating for dynamics and disturbances. Contact makes errors consequential.

Start with the observable system. Feedback, limits and emergency stops are tested from simulation through staged hardware trials. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how robot-control reasoning becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to compare open-loop and feedback control for holding a tray. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. A smooth simulation does not establish safe physical control. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Tactile sensing reveals contact, slip and force that vision may miss. Sensor placement and material affect what can be inferred.

Follow the mechanism before the headline. Manipulation tests vary object texture, shape and fragility with calibrated force and failure logging. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how tactile-robotics literacy becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to design a gentle grasp test for three objects with different surfaces. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. Detecting contact does not guarantee stable or safe grasp. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

People infer robot goals from movement, timing and communication. Robots must also respond to human uncertainty and preferences.

Make the comparison explicit. Studies measure task success with trust, workload, comfort and recovery from misunderstanding. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how human-robot interaction becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to compare two robot approaches to handing over an object. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. Efficiency alone can hide confusion or loss of control. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Speech, gesture, display and motion can explain intention and request help. Channels should match user ability and context.

Keep the evidence chain visible. Design tests include ambiguity, interruption and confirmation rather than only ideal commands. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how robot-communication judgement becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to write a clarification exchange for a robot facing two possible objects. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. Fluent language does not mean reliable grounding in the physical scene. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Assistive systems should align with a person’s goals, strengths and environment instead of treating impairment as one generic deficit.

Test the boundary as carefully as the main effect. Participatory design identifies tasks, fatigue, privacy, stigma and acceptable interaction before technology choices. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how assistive-design literacy becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to create two user journeys for the same visual-assistance idea. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. A technically impressive device can fail if it reduces dignity or independence. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

AR overlays information on the world; VR constructs an immersive environment. Registration, latency and attention affect usefulness and comfort.

Start with the observable system. Evaluation measures spatial accuracy, task performance, motion discomfort and cognitive load. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how XR-systems reasoning becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to design an alignment test for instructions placed on a physical object. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. A convincing visual overlay can still be misplaced or distracting. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Voice, gaze, gesture, haptics and displays can complement one another. They can also compete for limited attention.

Follow the mechanism before the headline. Design maps each message to a channel and tests noise, disability and divided-attention conditions. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how multimodal-interface literacy becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to allocate navigation alerts across sound, vibration and vision. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. Redundant channels help only when they remain consistent and perceivable. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Users calibrate reliance from experience, explanations and consequences. Overtrust and undertrust both create risk.

Make the comparison explicit. Studies compare confidence with actual reliability and observe whether users detect failures. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how human-AI trust judgement becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to plot user reliance against system accuracy across five trials. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. A friendly appearance can increase trust without increasing competence. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Inclusive design considers sensory, motor and cognitive variation from the start. Compatibility and customisation extend usefulness.

Keep the evidence chain visible. Teams test with diverse users and support settings, training and alternative controls. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how accessible-technology literacy becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to audit a smart interface for vision, hearing, reach and literacy. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. Compliance with a checklist does not prove equal practical access. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Laboratory benchmarks isolate variables, while field studies reveal interruptions, habits and social constraints. Both evidence types matter.

Test the boundary as carefully as the main effect. A staged evaluation moves from component tests to integrated tasks and supervised real-world trials. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how smart-system evaluation becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to create a ladder from simulation to field use for a home assistant. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. A successful demonstration is not a validated deployment. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

Always-on sensors can capture bystanders and sensitive spaces. Physical action adds hazards beyond information error.

Start with the observable system. Data minimisation, local processing, access control, operating limits and incident response are designed together. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how responsible-smart-systems becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to build a threat-and-hazard table for smart glasses in public. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. On-device processing reduces some risks without removing misuse or physical harm. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter · Next chapter

A credible portfolio records user, situation, hardware, sensors, model, training data, operating domain, task, uncertainty, human study, accessibility, privacy, safety, fallback and claim.

Follow the mechanism before the headline. The receipt links each demonstration to exact conditions and preserves failures that define the boundary. The important question is not simply whether something can be measured, modelled or built, but which claim that work supports under the stated conditions. A useful explanation names the system boundary, the independent and dependent quantities, the comparison, and the evidence that could change the conclusion.

In a real investigation, connect this idea to the surrounding workflow. Record the starting material or participant context, instrument or model version, settings, controls, raw observation, transformation, uncertainty and decision. That record lets a classmate or future researcher see whether a surprising result came from the phenomenon, the method, the sample or the interpretation. This is how smart-systems career preparation becomes a transferable scientific capability rather than a remembered definition.

A practical way to learn it is to build one page for a fictional assistive robot and ask a peer to find every hidden assumption. Before accepting the result, ask what was held constant, what was deliberately changed, what was actually observed, and whether an alternative explanation fits the same evidence. Add one negative or blank control and one independent cross-check wherever the question permits.

Interpretation needs a stop line. Real trials with people and robots require approvals, supervision and appropriate safety controls. Write that boundary beside the result rather than hiding it in a footnote. Then state what additional measurement, population, comparison, duration or authority would be needed before making the next larger claim. Careful limits make an article more useful because readers can act on what is known without mistaking curiosity for certainty.

Contents · Previous chapter