AGI

AGI
Name	Artificial General Intelligence
Caption	Conceptual diagram of a general-purpose intelligent system
Type	Technology
Developer	Multiple institutions
Introduced	20th–21st century

AGI

Artificial General Intelligence denotes a hypothesized class of artificial systems capable of performing any intellectual task that a human being can perform. It occupies a central role in discussions involving research institutions, think tanks, technology firms, policy bodies, and academic disciplines across continents. Debates about its feasibility, timeline, control, and impact involve figures, organizations, and events from computer science, cognitive science, philosophy, and international policy.

Definition and scope

Researchers and institutions vary in definitions: some describe AGI as parity with human cognitive capacities cited in publications from Massachusetts Institute of Technology, Stanford University, and University of Cambridge; others frame it relative to benchmarks used by OpenAI, DeepMind, and research groups at Google and Microsoft Research. Definitions often reference historical criteria from thinkers associated with Alan Turing, John McCarthy, Marvin Minsky, and frameworks influenced by work at Carnegie Mellon University and University of California, Berkeley. Scope debates invoke use cases in contexts like scientific discovery at CERN, strategic planning in institutions such as the RAND Corporation, and creative production linked to studios like Pixar Animation Studios. Competing taxonomies draw on models from laboratories including IBM Research and policy analysis by organizations such as the Future of Life Institute and Centre for the Study of Existential Risk.

History and development

The intellectual lineage traces to early computing milestones at Bell Labs, conceptions published by Alan Turing and programs from researchers at Dartmouth College and MIT Artificial Intelligence Laboratory. Key periods include symbolic AI prominence shaped by scholars like Herbert A. Simon and Allen Newell, the statistical turn exemplified by work at Bell Labs and AT&T Laboratories, and the resurgence of machine learning through contributions from Geoffrey Hinton, Yoshua Bengio, and Yann LeCun at institutions such as University of Toronto and New York University. Corporate efforts by IBM with Deep Blue and Watson, and later breakthroughs by Google DeepMind with systems inspired by work at University of Alberta and University College London, marked technological inflection points. Policy and public discourse intensified after incidents and publications involving Elon Musk, Stephen Hawking, Nick Bostrom, and reports from bodies like the National Science Foundation and European Commission.

Approaches and architectures

Contemporary strategies span architectures derived from connectionist traditions at McGill University and University of Montreal, symbolic systems tracing to Stanford University and MIT, and hybrid proposals championed in venues such as NeurIPS and ICLR. Prominent paradigms include deep learning variants with roots in research by Yann LeCun at Facebook AI Research, reinforcement learning advances from DeepMind inspired by work by Richard Sutton and Andrew Barto, and probabilistic programming influenced by projects at University of Cambridge and Princeton University. Neuro-symbolic systems draw on collaborations across Carnegie Mellon University and Massachusetts Institute of Technology, while neuromorphic engineering references efforts at Institute of Electrical and Electronics Engineers conferences and labs like Intel Labs. Distributed and multi-agent architectures relate to studies at Stanford University and applications by companies such as Amazon Web Services.

Safety, ethics, and governance

Safety and governance debates involve think tanks like the Future of Humanity Institute and policy groups including the Brookings Institution and Chatham House. Ethical frameworks reference positions advanced by scholars such as Peter Singer and Martha Nussbaum, and standards proposed in reports from bodies like the Organisation for Economic Co-operation and Development and United Nations panels. Risk analysis draws on scenarios discussed by Nick Bostrom and coordination proposals advocated by leaders from OpenAI, DeepMind, and the Partnership on AI. Regulatory initiatives cite precedents from legislation in jurisdictions like the European Union and consultations involving agencies such as the National Institute of Standards and Technology. Multistakeholder mechanisms often mirror processes used in agreements hosted by the World Economic Forum and intergovernmental dialogues at the G7.

Capabilities and evaluation

Benchmarks and evaluation methodologies emerge from competitions and datasets organized by groups including ImageNet contributors, GLUE and SuperGLUE teams, and testbeds from Allen Institute for AI. Evaluation covers domains demonstrated by systems associated with Google DeepMind (game playing and protein folding), OpenAI (language modeling), IBM Watson (question answering), and robotics research at MIT and ETH Zurich. Metrics incorporate performance criteria developed in conferences like AAAI and ICML, and governance-oriented stress tests proposed by policy units at RAND Corporation and Centre for Strategic and International Studies.

Societal and economic impacts

Analyses of labor and markets reference studies by International Labour Organization, Organisation for Economic Co-operation and Development, and research teams at National Bureau of Economic Research. Discussions of strategic competition and defense relate to publications from NATO and national ministries such as the United States Department of Defense. Cultural effects draw attention from institutions like the Smithsonian Institution and media companies such as Netflix, while education-sector implications involve universities like Harvard University and University of Oxford. Financial sector applications and disruptions reference activity on exchanges like the New York Stock Exchange and firms including Goldman Sachs and BlackRock.

Technical and philosophical critiques

Technical critiques invoke limits explored by researchers at Princeton University and California Institute of Technology, skepticism from scholars such as Hubert Dreyfus and John Searle, and philosophical analysis appearing in journals connected to Oxford University Press and Cambridge University Press. Challenges include questions about interpretability emphasized by teams at Carnegie Mellon University, scalability debated at Stanford University, and embodiment discussed by researchers at Tokyo Institute of Technology and University of Tokyo. Philosophical debates reference classical problems discussed by Immanuel Kant and modern treatments by philosophers associated with University of Pittsburgh and Rutgers University.

Category:Artificial intelligence