This paper makes the case that a powerful new discipline, which we term perception engineering, is steadily emerging. It follows from a progression of ideas that involve creating illusions, from historical paintings and film, to video games and virtual reality in modern times. Rather than creating physical artifacts such as bridges, airplanes, or computers, perception engineers create illusory perceptual experiences. The scope is defined over any agent that interacts with the physical world, including both biological organisms (humans, animals) and engineered systems (robots, autonomous systems). The key idea is that an agent, called a producer, alters the environment with the intent to alter the perceptual experience of another agent, called a receiver. Most importantly, the paper introduces a precise mathematical formulation of this process, based on the von Neumann-Morgenstern notion of information, to help scope and define the discipline. It is then applied to the cases of engineered and biological agents with discussion of its implications on existing fields such as virtual reality, robotics, and even social media. Finally, open challenges and opportunities for involvement are identified.
If K is infinite, then the costs must be carefully chosen so that the sum is finite for successful policies; alternatives include discounted cost, average cost, and termination actions (6). If K is finite, then a final cost term lF (K+1, yr K+1, r Now suppose that disturbance-based extensions of f r and hr are introduced for the receiver, to obtain F r and H r, as defined in Section 2.1. In this case, the receiver is no longer predictable, even from the producers perspective. It is thus more effective for the producer policy to be formulated as state-feedback p : X p U p, which even implies universe-state feedback. Acknowledging that this may be extreme in many settings, Section 3.4 removes producer omniscience to obtain other cases of information-feedback policies for the producer. K+1) may be added.
id: 9811b8fabc71e5649933a7e90745f52d - page: 16
Consider characterizing the evolution of the whole system under the implementation of a fixed, state-feedback producer policy p. Under the fully predictable case, a sequence : K is determined from the initial universe state 0. In the case of a nondeterministic disturbance-based receiver, then a set of possible sequences is instead obtained. If is finite, then the process can be imagined as a nondeterministic finite automaton (NFA) over . One should consider worst-case analysis to determine whether a goal can be guaranteed to be accomplished. With a cost model, one can consider minimizing the worst-case perceptual experience. In the case of a probabilistic disturbance-based receiver, a Markov chain is obtained under the implementation of p (also called Markov decision process (MDP) by artificial intelligence researchers). In this case, expected-case analysis could be used to assess the probability that the goal will be satisfied under p. In this case, p can be selected to maxim
id: dca1ae472a3c74b1b3670b3520bc5aa6 - page: 16
A cost model can additionally be used, with the resulting optimization being to find the lowest expected-case cost under the implementation of p.
id: e0ce65a8a229ed43015604e684b44c94 - page: 16
3.4. Producers with Imperfect Information If the producer is not omniscient, then it may not have access to enough information to ensure that the targeted perceptual experiences function as desired. To analyze what might happen between the producer and receiver, it will be helpful to nevertheless introduce a third-person perspective in which we as scientists or engineers have access to more information than the producer. This could be modeled formally as an observer agent.
id: 32a00f9364f3c90794dd85346bcf9ec7 - page: 16