Automation versus machine autonomy

Abstract: When planning support to people in their living environments using technology based on AI, the first approach is to look for the possibility of automation of functionalities favoring independent living. For example, it is possible to install sensors to automatically switch on or off the lights, to turn on the television when a program of interest starts, to remind people the drugs to be taken at the right time. This is supposed to support people in maintaining their autonomy, i.e., their ability to live in their environment without unwanted external interference. When an Artificial Intelligence system detects abnormal behavior of the person, for example through physiological and/or environmental parameters, may be technologically capable to act: for example it can send an alert to the person, via a sound or a message on an electronic device, can proceed to warn relatives or caregivers, sending an appropriate signal through the TLC network, can also administer medicines, or make equally impactful decisions, via home robots or equivalent devices integrated into the home network.The main problem is that not all people, even if supported by automation, are able to live autonomously, due to physical or mental limitations. Therefore, it is necessary to consider the possibility that the supporting systems become partially autonomous, i.e., able to reach a predefined goal according to the current situation without recourse to human control. Such systems are supposed to perceive their environment via sensors, proactively create a plan of action according to the situation and related constraints and execute the planned actions safely and reliably via actuators.Different possibilities of autonomy are possible. The support system offers no assistance - the human must make all decisions and actions, offers a complete set of decision/action alternatives; narrows the selection down to a few alternatives; suggests one alternative. Then it may execute that suggestion if the human operator approves, or allows the human a restricted time to veto before automatic execution, or executes automatically, and necessarily informs the human, or informs the human only if asked etc. This implies several ethical problems, as shown in the following examples. It is necessary to decide who can check that the state of the person is such as to require more advanced autonomous decision-making systems based on AI? Which people and with what priority should those who have a relationship with the person in the house be allowed to contribute? Who becomes responsible for the malfunction of the equipment, in the event of a breakdown, which may cause personal injury? What is the level of decision that can be entrusted to the machine compared to that which is delegated to a caregiver? The answers are related to the person's capacity for autonomy, the responsibility that those who care for this person have, but also e.g., the person's will, the readiness to intervene, the type of warning. The problem will be discussed with reference to three applications implemented at IFAC: cooking in an intelligent kitchen (activity level), support for the solitude of people at home (social support); pedestrian mobility (activities in public areas).

Keywords: Ethics, Artificial Intelligence, Autonomy, Automation, Well-being

DOI: 10.54941/ahfe1004517

Cite this paper:

​