Robots and autonomous technologies practice questions.

A test vehicle running automated driving software struck and killed a pedestrian in Tempe, Arizona, in March 2018. Investigators found that the software repeatedly reclassified the person crossing and did not brake in time, while the operator in the driver's seat was not watching the road. What does the case illustrate?

• responsibility is distributed across developer, operator and regulator
• an autonomous system leaves no one who can be held responsible
• the person in the driver's seat carries the whole responsibility
• sensor hardware, rather than software, is the weak point of such vehicles

Answer: responsibility is distributed across developer, operator and regulator

Every link in the chain contributed: a classifier that could not settle on an object type, a braking response that had been disabled, a monitoring task that people perform badly, and a testing regime that allowed the arrangement. Accountability in Digital Society is answered by mapping that chain rather than by choosing a single culprit.

Since 2016 a drone service has flown blood products from a central hub to remote clinics in Rwanda. Which of the following is an implication of the service rather than an impact?

• delivery times to remote clinics fall from several hours to minutes
• stock can be held centrally instead of at every clinic in the network
• fewer motorcycle journeys are made on rural roads by couriers
• deliveries reach clinics that rural roads cut off during the rainy season

Answer: stock can be held centrally instead of at every clinic in the network

Faster delivery, fewer road journeys and reaching clinics the roads cut off are direct effects of the service. Holding stock centrally is a change in the way the health system is organised, and it creates a new dependency on one hub, its flight clearances and its operator.

A state operates an air-defence system that identifies, tracks and fires on incoming objects without a human confirming each engagement, because only a few seconds are available. Critics accept that no operator could act within that time and still object to the design. What is their objection?

• That the system will inevitably identify targets less accurately than a trained human operator would
• That responsibility for a wrong engagement cannot be located once no person authorised it
• That such a system cannot be tested at all before it is used in a real operational setting
• That the software is not fast enough to complete an engagement in the seconds available

Answer: That responsibility for a wrong engagement cannot be located once no person authorised it

The argument is about accountability rather than about accuracy: a machine cannot be answerable, and the operator who could be has been designed out of the moment of decision. This is the responsibility gap in its sharpest form, which is why the debate concentrates on meaningful human control rather than on performance.

A hospital introduces autonomous cleaning machines that navigate by mapping the corridors continuously and photographing obstacles so that the fleet's routing can be improved. Which implication should the hospital weigh before deployment?

• That the machines will be unable to clean rooms whose layout changes from one day to the next
• That cleaning staff will have to be retrained each morning in how to operate the navigation software
• That a continuous record of who was in which corridor, and when, is created as a by-product
• That the mapping data will be too large for the hospital to store on its own equipment

Answer: That a continuous record of who was in which corridor, and when, is created as a by-product

A robot that must perceive its environment necessarily records it, so surveillance arrives as a side effect of navigation rather than as anyone's purpose. Deciding who may query that record, and for how long it is kept, is a governance decision that has to be taken before the machines arrive.

A vineyard replaces blanket herbicide spraying with a towed machine that recognises individual weeds and treats each one, cutting herbicide use by about 80 per cent. Which of the following is an implication of the change rather than an impact?

• The quantity of herbicide the vineyard buys each season falls sharply, and the chemical bill falls with it
• The weeds between the vines are removed as effectively as they were under blanket spraying, with no loss of yield
• The machine has to be drawn along every row more slowly than the sprayer it replaces
• Recognising each plant becomes a bought-in service the vineyard now depends on

Answer: Recognising each plant becomes a bought-in service the vineyard now depends on

Lower herbicide purchases, unchanged weed control and slower passes along the rows are measurable effects of the operation itself; dependence on a bought-in recognition service changes what the farm has to buy and whom it must keep a relationship with in order to weed at all. The knowledge that used to sit with the grower now sits in a supplier's model.

Since 2018, Starship delivery robots have carried groceries along pavements in Milton Keynes in the United Kingdom, sharing the space with pedestrians. For residents who never use the service, one implication of the deployment is that

• the robots' cameras make the pavements safer for everyone walking there
• grocery prices fall across the town as delivery costs are reduced
• public walking space now also functions as a private delivery route
• maintaining the pavements becomes the legal duty of the robot operator

Answer: public walking space now also functions as a private delivery route

Non-users are stakeholders too: pavements are a shared public resource, and a commercial fleet operating on them converts part of that commons into logistics infrastructure — which wheelchair users and visually impaired pedestrians must negotiate. The distractors assert benefits and duties that do not follow from the deployment.