China
Africa
Explore chapters and articles related to this topic
Introduction
Published in Changliu Liu, Te Tang, Hsien-Chung Lin, Masayoshi Tomizuka, Designing Robot Behavior in Human-Robot Interactions, 2019
Changliu Liu, Te Tang, Hsien-Chung Lin, Masayoshi Tomizuka
In robotics, the state-action-cost pairs provide a straightforward encoding of cluttered data. These pairs describe the evolutional system dynamics, the desired policy principle and the hidden cost mechanism implicitly. The state variables are the smallest possible subset of system variables that can represent the entire system condition at any given time. The action is the control command that can be applied on the system to influence the evolution of the system. The instant cost is a quantitative description of the goodness or badness of the current state and action. For a rational agent, it is supposed to be an optimizer which tends to achieve the highest rewards or the lowest costs by taking appropriate actions.
Ignorance by proxy
Published in Cathrine Hasse, Posthumanist Learning, 2020
In relation to the second surprise of normativity, we can discuss if the textbook by Russell and Norvig belongs to what many have defined as an engineering culture (e.g. Kunda 1992; Hansen 2018; Sorensen 2018), where engineers share particular understandings of what is the “best” or “optimal solution” to problem solving issues and the like. Here we find many statements like, “A rational agent is one that acts so as to achieve the best outcome or, when there is uncertainty, the best expected outcome” (Russell & Norvig 2010, 4). Optimal and best generally refer to aspects of the technical process – the meaningfulness lies in making “learning” paths more efficient, where broad searches for instance can be optimised.
Cognitive Architectures
Published in Ron Fulbright, Democratization of Expertise, 2020
This type of agent is called a rational agent because incorporating utility values into decision processes allows the agent to make decisions yielding the best possible overall performance. When an agent takes the optimal course of action its behavior is considered rational.
Reinforcement learning based optimal decision making towards product lifecycle sustainability
Published in International Journal of Computer Integrated Manufacturing, 2022
Yang Liu, Miying Yang, Zhengang Guo
RL extends machine learning beyond just modelling and prediction to also decision-making. The algorithm is a rational agent that learns how to act in an uncertain environment by trial and error. The agent need only be given the problem state at the current time xt and the immediate reward rt of that state. The reward can be seen as the profit of a business decision, or equivalently the inverse cost, material and environmental. Such profits and costs are condensed into one number, the reward, which the agent uses as feedback on improving. As seen in Figure 19, at each step, the agent will have to select an action at, which will result in a new problem state and a new reward. This is a sequential decision problem as the sequence of actions determines the resulting sequence of states and rewards. The agent will have to plan a number of steps into the future to find the sequence of actions that maximises the total reward over time. In classic RL, no prior knowledge is needed of either the rewards or how the problem state changes in response to an action, xt+1 = f (xt, a).
Normative and descriptive rationality: from nature to artifice and back
Published in Journal of Experimental & Theoretical Artificial Intelligence, 2018
We therefore relabel agents who act merely in a means-ends fashion as ‘ethical’ ones, and call agents ‘rational’ who function under a broader concept of rationality which also encompasses further facets (Cowen, 2004). If we retained the concept ‘rational’ for the utility-maximising agents, we would then still need to find another label for the broader notion; and it turns out that some people already have: In AI, a distinction is often made between ‘rational’ agents, which act in a utility-maximising way, and ‘intelligent’ agents in general, with the former being taken as a sub-class of the latter. Following the hierarchy of agent classes introduced in Russell and Norvig (2003), agent architectures in AI can be ordered on a scale of increasing degree of (perceived) intelligence and overall capability level starting from reflex agents and ranging through purely goal-based agents to utility-based agents and learning agents at the upper end. When speaking of ‘rational’ agents, reference is usually made to utility-maximising agents as the second highest class in the hierarchy. Utility-maximising agents in general are required to come equipped with a utility function as an internal performance measure, which in the case of ‘rational’ agents additionally has to match the performance measures of the environment. In this sense, while a utility function might not be necessary for rationality in general, ‘rational’ agents in AI can be modelled (and are expected to act) as if they are governed by one – thus, for all relevant purposes, becoming equal to ethical agents.