King's College London · FACULTY OF ARTS & HUMANITIES

7AAVCD42 Chap.8 Physical AI, Sensing and Responsibility

- one subject, every graph, every model, every mark
5 Chapters2-page Bible
Our own words - no uploaded lecturer files
Updated for this semester
Chapter 8 of 10 · 7AAVCD42

Physical AI, Sensing and Responsibility

The street model is not the street

Physical AI senses an environment through cameras, radar, lidar or other devices, builds a working model and selects an action. This chapter reconstructs that loop and locates failure in sensing, representation, planning or control. It then links technical stages to operators, designers, organisations and regulators.

The method avoids a simple human-or-machine blame choice and extends responsibility analysis to privacy, hidden labour and deployment decisions.

Reconstruct the sense-model-act loop

Record what each sensor can detect and where conditions make it unreliable. Explain how signals are combined into objects, positions and predicted movement. Identify the planning rule and physical action.

The action changes the environment that the next sensing cycle receives, so delay or misclassification can compound across the loop.

Operating conditions belong inside the system boundary when deployment makes them foreseeable. Rain, glare, occlusion, road layout and unusual movement are not external excuses.

Evaluation should test relevant conditions and specify a safe response when uncertainty rises.

Locate the first failed stage

An object may be detected but tracked incorrectly. A good street model may feed an unsafe planning rule. A correct plan may be executed too late. Distinguish sensing, representation, prediction, planning and control before recommending repair.

The earliest failed stage often determines which team and evidence matter.

Use a timeline. Mark the information available at each moment, the system response and the remaining stopping or intervention time.

This prevents hindsight from assigning knowledge to an operator or system before it was available.

Distribute responsibility carefully

Responsibility can include causal contribution, professional duty, organisational control and legal liability. These categories may point to different actors.

Operators, safety staff, designers, maintenance workers, deployers and regulators possess different information and authority. Do not force the answer into a choice between one human and the machine.

Counterfactuals sharpen the analysis. Ask whether a different speed rule, sensor layout, alert design, testing programme or deployment boundary would likely have prevented the event.

Then identify who controlled that decision and what evidence would have supported the alternative.

Include privacy and hidden work

Street and domestic sensors capture bystanders and intimate spaces. Remote support workers may review difficult scenes, while data workers label previous events.

Describe how this labour sustains apparent autonomy and how data collection extends beyond the immediate user.

For coursework, link a technical stage to an organisational choice in every paragraph. State the failure, actor, available intervention and consequence. Finish with the most plausible prevention route and acknowledge remaining uncertainty rather than announcing that responsibility has been solved completely.

In this chapter

What this chapter covers

  • 01

    Sensor Input

  • 02

    Street Model

  • 03

    Action Planning

  • 04

    Accident Chain

  • 05

    Distributed Responsibility

Worked example · free

Reconstruct a crossing near miss

Q [8 marks]. A delivery robot slows late when a pedestrian appears from behind a parked vehicle. Explain the failure and responsibility. The marks shown are an AskSia study allocation, not an official King's marking scheme.
Sense-model-act loop.Sense-model-act loop. Sensors build a street model that guides action, while each action changes the environment observed on the next cycle.sensorsstreet modelactionenvironment changes the next input
Sense-model-act loop. Sensors build a street model that guides action, while each action changes the environment observed on the next cycle.
  • 3Reconstruct the timeline.
  • 2Locate the first failed stage.
  • 3Distribute responsibility.
Limited visibility shortened detection time, but the audit must test whether uncertainty and an occluded crossing should have led to lower speed. Responsibility includes design, testing and deployment choices.
Sia tip — Place every actor beside the information and control available at that moment.
Glossary

Key terms

Sensor Fusion
Sensor Fusion combines information from multiple sensing systems into a working representation of the environment.
Street Model
Street Model is the system's constructed representation of nearby objects and movement, not the street itself.
Distributed Responsibility
Distributed Responsibility recognises causal and governing roles across people, organisations and technical components.
FAQ

Physical AI, Sensing and Responsibility FAQ

What is the sense-model-act loop?

Sensors collect signals, software constructs a representation, planning selects an action and the action changes the next environment observed. Technical failure can enter at any stage.

Why is the street model not the street?

Sensors are limited by position, conditions and design, while software selects and interprets signals. The resulting representation can omit or misclassify features of the physical environment.

Who is responsible for a physical AI accident?

Examine causal contribution, professional duty, organisational control and legal liability separately. Operators, designers and deployers can hold different responsibilities depending on available information and authority.

Study strategy

Assessment move

Draw an accident timeline with sensing, representation, planning and action rows. Add the actor able to prevent or repair failure at each point. Record environmental conditions and the earliest moment at which uncertainty should have changed behaviour. Build a counterfactual for sensor layout, speed policy, alert design and deployment boundary, then identify who controlled each choice.

Separate causal contribution, professional duty, organisational control and legal liability rather than selecting one nearest person. Add the remote workers, maintainers and data labelers that make the system function. Finally, trace what bystander data is collected and how long it remains useful. A strong revision answer connects one technical failure to one preventable organisational decision.

Re-run the timeline after changing a single condition, such as visibility or operator workload. Preserve unaffected stages and explain why the responsibility analysis narrows or shifts. Close by selecting the earliest feasible intervention and defending that choice.

A+Everything unlocked
Unlocks this Bible + all 18 of your subjects at King's College London, plus 1,000+ more Bibles across the AskSia library.
Sia - your 7AAVCD42 tutor, unlimited, worked the way the exam marks it
The full 2-page Bible + practice bank with worked solutions
Chrome extension - sync your LMS so Sia knows your deadlines
Bilingual EN / Chinese on every Bible and every Sia answer
$0.99 Trial
30-day money-back · cancel in one tap · how it works