Evolution and control of artificial superintelligence (ASI): a management perspective

Abstract

Purpose The purpose of this paper is to address both the evolutionary and control aspects associated with the management of artificial superintelligence. Through empirical analysis, the authors examine the diffusion pattern of those high technologies that can be considered as forerunners to the adoption of artificial superintelligence (ASI). Design/methodology/approach The evolutionary perspective is divided into three parts, based on major developments in this area, namely, robotics, automation and artificial intelligence (AI). The authors then provide several dynamic models of the possible future evolution of superintelligence. These include diffusion modeling, predator–prey models and hostility models. The problem of control in superintelligence is reviewed next, where the authors discuss Asimov’s Laws and IEEE initiative. The authors also provide an empirical analysis of the application of diffusion modeling to three technologies from the industries of manufacturing, communication and energy, which can be considered as potential precursors to the evolution of the field of ASI. The authors conclude with a case study illustrating emerging solutions in the form of long-term social experiments to address the problem of control in superintelligence. Findings The results from the empirical analysis of the manufacturing, communication and energy sectors suggest that the technology diffusion model fits well with the data of robotics, telecom and solar installations till date. The results suggest a gradual diffusion process, like any other high technology. Thus, there appears to be no threat of “existential catastrophe” (Bostrom, 2014). The case study indicates that any future threat can be pre-empted by some long-term social measures. Originality/value This paper contributes to the emerging stream of artificial superintelligence. As humanity comes closer to grappling with the important question of the management and control of this technology for the future, it is important that modeling efforts be made to understand the extant perspective of the development of the high-technology diffusion. Presently, there are relatively few such efforts available in the literature.

The risks associated with Artificial General Intelligence: A systematic review

“…The modelling approaches identified in the reviewed articles included fault trees and influence diagrams, Bayesian modelling, diffusion modelling, predator-prey models, hostility models, theoretical all-pay contest model, risk estimate modelling and AI treachery threat model. The outcomes of the modelling studies in the current review have provided information to better understand AGI containment fails Baum et al, 2017), technology diffusion enabling a safer AGI (Narain et al, 2019), and the appropriateness of current risk management procedures to deal with the risks of AGI (Bradley, 2020). Even though AGI does not yet exist, these modelling approaches have demonstrated the capability to derive information to better inform decision making the risks associated with AGI.…”

“…Within the included articles, three described the specific functionality of the AGI. These included 12 different human professions, smart homes, and autonomous vehicles (Chen & Lee, 2019), a malicious AGI (Bradley, 2020), and AGI in the manufacturing, communication, and technology sectors (Narain et al, 2019). The remaining articles referred to generic AGI systems and did not describe a specific AGI functionality.…”

“…Various data driven and theoretical models have previously been used to forecast the behaviour of emerging products and technologies, and will be necessary for estimating the risks of AGI (Narain et al, 2019). The modelling approaches identified in the reviewed articles included fault trees and influence diagrams, Bayesian modelling, diffusion modelling, predator-prey models, hostility models, theoretical all-pay contest model, risk estimate modelling and AI treachery threat model.…”

“…The majority of studies reviewed were non-domain specific and focused on the general risks to humanity. However, specific domains identified in the eligible articles included autonomous vehicles, human professions, and smart homes (Chen & Lee, 2019), manufacturing, communication, and energy (Narain et al, 2019), law (Nindler, 2019), environment and social aspects (Pueyo, 2018). Two notable domains that did not feature in the review were research investigating the risks associated with defence and autonomous weapons systems, and healthcare.…”

“…Currently there are few modelling efforts investigating the diffusion of AGI into society (Narain et al, 2019), which was exemplified in the current review by the limited number of modelling approaches identified. Although our understanding of the risk associated with AGI is limited, they have similar characteristics to other risks that involve integration of humans and technology, and modelling techniques exist that may provide meaningful analyses in relation to AGI .…”