Creating Beneficial, Democratic Artificial General Intelligence
1 like250 views
This document discusses creating beneficial artificial general intelligence (AGI). It begins by distinguishing between narrow AI, which can only perform specific tasks, and general AI, which could theoretically learn to perform any human task. It describes the OpenCog project and SingularityNET framework, which aim to develop AGI by integrating multiple AI techniques. SingularityNET would create a decentralized marketplace where different AI agents could cooperate and their capabilities could emerge from the network as a whole. The document outlines various applications of AGI and a roadmap for developing SingularityNET over the next few years.
Creating Beneficial, Democratic Artificial General Intelligence
- 3. 3 Narrow AI Can only do one thing (e.g., Deep Blue canʼt drive a car; Google Car canʼt play chess) TRAINED or just PROGRAMMED • Deep Blue • AlphaGo • Facebook Face Recognizer • Google Self Driving Car • IBM Watson for Jeopardy • IBM Watson for Medical General AI Can (in principle) learn to do anything a human can learn and more. EDUCATED OpenCog and SingularityNET are aimed here
- 4. AGIs must Learn from Experience 4 an AI should assign meaning to all its ideas, abstraction and hypotheses via extrapolation from its observations An AGI must build its own model of itself and the world
- 5. 5 Multiple Narrow AI Approaches can play a role in creating AGI Evolutionary learning: Genetic Programming, MOSES, CMA-ES, … Logic systems: Markov Logic Networks, PLN, theorem-provers,… Neural networks: deep and shallow hierarchical networks, heterarchical attractor neural networks (e.g. ECAN),… Stochastic local search: SAT solvers, SMT, … and plenty more …
- 6. 6 The OpenCog framework approaches AGI by tightly integrating multiple AI methods, enabling them all to use a common dynamic knowledge hypergraph
- 7. 7 OpenCog powers the genomic and biomedical inference engine underlying Mozi AI Health, a cutting-edge decentralized cloud service providing advanced intelligence to the discovery of therapeutics and diagnostics and the control of systems biology simulations applying OpenCog to challenging practical problems helps bridge the gap between here and AGI The OpenCog AGI engine is already a core component underlying Sophia, SingularityNETʼs humanoid spokesperson, built by our partner Hanson Robotics. OpenCog applications
- 8. 8
- 10. 10 “Robots that respond with unconditional love“ Hanson Robotics humanoid robot “Sophia” used experimentally as a meditation coach — pilot psychological studies are very promising! Loving AI
- 11. 11
- 12. 12
- 13. 13 EveNet — A generative model for data driven facial expressions https://github.com/elggem/EveNet https://arxiv.org/abs/1609.03499 EveNet Motion capture training data from actor TRAINING INFERENCE (using deep neural nets to give Sophia more realistic facial expressions) Model captures dependencies among muscles and over time Generated output from neural network
- 14. 14 Hanson robot head that holds a simple conversation with obvious awareness of you, itself and its surroundings and context. 2 yrs 4-6 yrs 7-9 yrs Hanson robot that can learn via interaction with humans and the world, similarly to a young child (with some “idiot savant” capabilities) To get to full-on human- level AGI from an “idiot savant” child level AGI is mostly a management and organization problem! The path from human- level to transhuman AGI will be achieved mainly by AGI improving and re- engineering itself
- 15. 15 SingularityNET — The Decentralized AGI Mind-Cloud
- 16. 16 Intelligence exists on the emergent level of the multi-agent network, as well as on the level of the individual AI agents in the network SingularityNET loosely couples multiple AI algorithms, methods and solutions — enabling them to cooperate as a society and economy of minds OpenCog instances, neural networks, and other types of AI agents co-exist and inter-operate within the SingularityNET decentralized cognitive compute fabric
- 17. 17 SingularityNET — a blockchain-based open market for interoperating AIs Transitioning From Narrow AI to Artificial General Intelligence Providing Businesses with Smarter, Broader, More Automated AI Services Ensuring AI Is Applied For the Common Good Key Objectives 01 0302
- 18. 18 Data processing tools AI Algorithms AI Solutions / Services Software & Hardware Applications SingularityNET conceptual workflow MARKET AGI Token DATA Infrastructure abstraction Blockchain / distributed-ledger + offchain messaging/storage 18
- 19. AI Nodes in the SingularityNET carry out diverse cognitive and analytic operations, and exchange diverse types of data 19 This is a “network effect” from both business and cognitive perspectives AI Nodes may join together into “federations” – subnetworks of nodes that habitually work together to carry out particular sorts of tasks.
- 20. 20 AI Nodes in the SingularityNET can also spawn new AI nodes – the network is self-growing and self-pruning
- 21. 21
- 22. 22 Narrow AI Can only do one thing (e.g., Deep Blue canʼt drive a car; Google Car canʼt play chess) TRAINED or just PROGRAMMED • Deep Blue • AlphaGo • Facebook Face Recognizer • Google Self Driving Car • IBM Watson for Jeopardy • IBM Watson for Medical General AI Can (in principle) learn to do anything a human can learn and more. EDUCATED Tightly coupled AI algorithms cooperating for powerful general learning, within individual agents Loosely coupled AI algorithms and services, cooperating together in a marketplace that is an economy of minds and an emergent meta-level mind
- 24. 24 Elon Musk sounds an alarm about AI 2014
- 25. 25 2015 Elon Musk and others commit $1B to OpenAI for beneficial AI development
- 26. 26 2017 Elon Musk hires top OpenAI researcher for Tesla, steps back from OpenAI governance to avoid conflict of interest
- 29. A proactive approach to beneficial AGI 29 Create a framework where scalable, powerful AI can be created by any knowledgeable developer and profitably applied to any problem, even those that donʼt match the business models of large corporations or governments Create a democratically-governed community oriented toward applying AI technology for the broad benefit of sentient beings, and retain this democratic, beneficial emphasis as AGI emerges
- 30. 30 Actively do good and be good, according to your own standards Do your best to deeply understand others’ standards as well “Donʼt Be Evil” was Never good enough anyway
- 31. St. Petersburg & Novosibirsk: Deep AGI R&D, computer vision, reputation systems,… 31 A Global Team to Seed a Global AI Community Hong Kong: head office, robotics/AI, bio-AI Addis Ababa: Decentralized AI applications, deep AGI, AI for agriculture and education Seattle: Scalable platform, AI-meets-blockchain Emerging community: for now, everywhere on EARTH … Bologna: Blockchain wizardry Belo Horizonte (Brazil): Distributed AGI
- 32. 32 Toward a Democratic, Participatory, Generally Intelligent Future
- 33. Next Steps for SingularityNET We're already building the smart contracts and APIs to power the initial network Network Development Weʼre working directly with corporate and research partners to meet industry demand Corporate & Research Partnerships We're aggregating a community of developers and projects to bootstrap SingularityNET Community Engagement 01 0302 33
- 34. 34 Development Roadmap 2016 2017 2018 2019 Outline development & economic logic Assemble team of AI, blockchain & cognition experts Create proof-of-concept and launch live alpha Launch scalable platform First set of commercial agents Massive global AI marketplace Gradual emergence of AGI
- 35. 35 “I set the date for the Singularity- representing a profound and disruptive transformation in human capability- as 2045. The nonbiological intelligence created in that year will be one billion times more powerful than all human intelligence today.” The Singularity is Near When Humans Transcend Biology - Ray Kurzweil (2005)
- 36. by bringing AGI, blockchain, robotics, bioscience and other advanced technologies together, in a participatory and democratic global compute fabric, we are moving rapidly toward realizing Singularitarian visions In 2018