The Future of Humanity Institute have published a technical report by K. Eric Drexler entitled Reframing Superintelligence: Comprehensive AI Services as General Intelligence.


Studies of superintelligent-level systems have typically posited AI functionality that plays the role of a mind in a rational utility-directed agent, and hence employ an abstraction initially developed as an idealized model of human decision makers. Today, developments in AI technology highlight intelligent systems that are quite unlike minds, and provide a basis for a different approach to understanding them: Today, we can consider how AI systems are produced (through the work of research and development), what they do (broadly, provide services by performing tasks), and what they will enable (including incremental yet potentially thorough automation of human tasks). Because tasks subject to automation include the tasks that comprise AI research and development, current trends in the field promise accelerating AI-enabled advances in AI technology itself, potentially leading to asymptotically recursive improvement of AI technologies in distributed systems, a prospect that contrasts sharply with the vision of self-improvement internal to opaque, unitary agents. The trajectory of AI development thus points to the emergence of asymptotically comprehensive, superintelligent-level AI services that— crucially—can include the service of developing new services, both narrow and broad, guided by concrete human goals and informed by strong models of human (dis)approval. The concept of comprehensive AI services (CAIS) provides a model of flexible, general intelligence in which agents are a class of service-providing products, rather than a natural or necessary engine of progress in themselves. Ramifications of the CAIS model reframe not only prospects for an intelligence explosion and the nature of advanced machine intelligence, but also the relationship between goals and intelligence, the problem of harnessing advanced AI to broad, challenging problems, and fundamental considerations in AI safety and strategy. Perhaps surprisingly, strongly self-modifying agents lose their instrumental value even as their implementation becomes more accessible, while the likely context for the emergence of such agents becomes a world already in possession of general superintelligent-level capabilities. These prospective capabilities, in turn, engender novel risks and opportunities of their own. Further topics addressed in this work include the general architecture of systems with broad capabilities, the intersection between symbolic and neural systems, learning vs. competence in definitions of intelligence, tactical vs. strategic tasks in the context of human control, and estimates of the relative capacities of human brains vs. current digital systems.




Sorted by Click to highlight new comments since:

Thanks for posting! You can also see a transcript of an EA Global talk on this topic here.

See also my summary and Richard Ngo's comments.

Curated and popular this week
Relevant opportunities