To achieve this goal, writes HIROSHI YAMAKAWA in the journal Big Data and Cognitive Computing, an AI system must operate continuously and stably (condition 1) and have an intervention method for maintaining peace among human societies based on a common value (condition 2). However, as a premise, it is necessary to have a minimum common value upon which all of human society can agree (condition 3).
In his paper, titled ‘Peacekeeping Conditions for an Artificial Intelligence Society’, the author investigated an AI system that satisfies condition 1. A part of the system may potentially be destroyed, so a robust intelligent agents (IA) society should be a team of autonomous and distributed IAs. A common value of humanity is shared among all IAs. Individual IAs would decompose common goals and derive means so that they can contribute to the advancement of common values. Each IA would diversify its activities to effectively divide tasks among them all. In order to adapt to their local environment, IAs would usually hold, as their local values, sub-goals derived from common goals.
There are a wide variety of local values and competition for available resources will create a competitive situation for IA comrades, the author states. It is an advantage that competition leads to an increase in capacity to achieve a common goal. “However, if the effort towards merely winning the competition increases, cooperation is lost, and devastating struggles occur, creating obstacles to achieving common goals.”
The ideal situation is one in which every agent believes that “all other agents intend socially acceptable goals”, YAMAKAWA writes. Here, “socially acceptable goals” means that the goals contribute to common goals and do not conflict with any other IA’s local values in practice. Under such circumstances, the IA society can achieve goals peacefully, efficiently, and consistently.
Communication channel problems, comprehension ability problems, and computational complexity problems exist, however, that may impede the realization of ideal situations. In an IA society based on a computer, it seems possible to design a distributed global management system that maintains the local values of distributed IAs.
Conversely, humans are biologically constrained. Irreversible death strengthens our survival instincts, and human beings need to maintain our species through reproduction. Humans must also distinguish their mates by appearance. For these reasons, similar people gather and become more likely to form a party. If people divide into groups with similar values and compete for resources, this can be a major cause of conflict.
YAMAKAWA assumes that building a universal AI system to arbitrate conflicts in human society based on a common value would reduce the existential risk. For stable and continuous operation, the AI system in his paper was constructed as an autonomously distributed system, which has concurrency, scalability, and fault-tolerance. Many issues remain to be solved, but this technology is feasible.
“It is desirable for the final form of our proposed AI system to be almost autonomous and worldwide, but part of that system can begin as a conventional AI system with the help of human operators.” However, a new issue will then arise regarding executing arbitrations that are consistent with a common value, while avoiding arbitrary influences of human operators.
Various possibilities for applying superintelligence to reduce existential risks caused by various non-AI factors, such as climate change, have been discussed before. With regard to AI itself, discussions have mainly focused on the increase in risks they might pose.
An approach using advanced AIs to reduce the existing risks that increase with the progress of AIs has not been sufficiently investigated. “However, effectively using the power of superintelligence or more elementary AI to construct future governance will create previously unknown possibilities for the future of humanity.”
No comments:
Post a Comment