Marek Rosa - Inventing General Artificial Intelligence: A Vision and Methodology
Download as PPTX, PDF2 likes700 views
The document discusses GoodAI's mission to develop general artificial intelligence as quickly as possible to help humanity. It outlines several key advantages of developing general AI over narrow AI, including higher return on investment potential and the ability for AI to recursively self-improve exponentially. The document also describes GoodAI's unified brain architecture approach and lists many intrinsic properties and learned abilities they are aiming to develop in artificial intelligence systems to achieve human-level general intelligence.
Marek Rosa - Inventing General Artificial Intelligence: A Vision and Methodology
- 1. Accelerating Towards General Artificial Intelligence: GoodAI and the Future of Humankind Marek Rosa CEO, CTO & Founder GoodAI and Keen Software House
- 2. About GoodAI • Interest in general AI since childhood • Development began in January 2014, within Keen Software House - Space Engineers + Medieval Engineers • My personal $10mil investment • Team of 30 researchers • Team members = co-owners of GoodAI
- 3. Our mission Develop general artificial intelligence as fast as possible, be helpful to humanity, and understand the universe.
- 4. Advantages of AGI • General AI vs narrow AI • Highest “return on investment” (ROI) possible => high-risk & high-reward • Recursively self-improving AI; exponential growth • Could be “our final invention” (in a good sense) • AI scientists, AI programmers, AI astronauts, AI …. • Solve the problems of humankind • Illness, death, climate change, rescue operations, exploring the universe • Everyone will benefit from AI (charities, corporations, individuals…)
- 5. Effect on Jobs • AI will be increasingly more skilled than humans at performing human tasks. Where will this lead? • Increased automation in our economy • Employers will start to prefer intelligent machine workers to human ones • Job replacement • Institute a universal basic income • Exit the human-based economy • No need to work to survive • Altruism • Investing in the AI future
- 6. What is Intelligence? • Learn, adapt, solve problems and achieve goals in a complex environment • Model of the world where only relevant parts are represented • Evolution vs intelligence - Intelligence is both faster and cheaper - Intelligence needs fewer resources • Maximize the chance of achieving goals in the future (goal / resources / time) BrainSensors Motors
- 7. Unified Brain Architecture • Our own all-in-one AI brain architecture - Composed of a network and sub-networks of "experts" (essentially small programs) - Purpose is to make a procedural representation of the world, past experiences, learned skills, predictions, plans, etc. • Not just integrating Deep Learning, Machine Learning, HTM, LSTM, or others as independent modules where each is focused on specifics - Look to the principles of intelligence, aim to understand them, and implement only those principles into our unified brain architecture - Building our own design, not on top of existing technology
- 8. Intrinsic Properties • One system manifests them all: - Over-generalization - Generalize-first - Analogy - Knowledge transfer + context switching: Reuse existing or modified programs (groups of experts) • Conflicting and Parallel Actions - Hierarchical long-term sequences - Actions (hierarchical) –> motor commands - Behaviors: internal, external, general to specific / concrete • Learning to learn • Additive learning, compositional learning • Online, continuous, lifelong learning • Unsupervised and supervised learning, puppet learning, guided and gradual learning, reinforcement learning • Pattern detection • Altering knowledge • Unified memory: working + short term + long term + episodic • Anomalies / novelty detection • One-shot learning • Pattern reconstruction • Detect uncertainty + confidence • Predictions – multiple and simultaneous • Perceptual consistency, continuity
- 9. Learned General Abilities • Visual attention / focus • Feature selection • Language? • Gratification delay • Mental time travel • Imagination • Active logic reasoning • Meta cognition • Third party intervention • Empathy • Mirroring • Recognize himself in the mirror • Abduction • Induction • Deduction • Imitation learning • Planning – long-term sequences of actions • Active learning • Proactive learning • Cooperation • Curiosity • Creativity • Imagination • Exploration • Exploitation
- 10. Learned Skills / Knowledge / Experience • Recognize apple, pear, door (locked / unlocked) • Apple tree produces apples...pear tree produces pears… • Open door • Close door • Day + night • …
- 11. School for AI • Set of simple game environments • Gradual and guided learning • Learning tasks • Communication • Can serve as an AGI benchmark • Sets the requirements for our Unified Brain Architecture!
- 12. Brain Simulator • In-house, collaborative platform for researchers, developers, and high-tech companies • Prototype and simulate artificial brain architectures • Share knowledge
- 13. Integration with Space Engineers
- 14. Future Business Applications • Useful to start with games (safe, low-risk environment) • Easy-to-implement applications • Later commercialization - Automating science, engineering, art, manufacturing, robotics, etc. • Paradox: need to reach human-level AI before commercializing?
- 15. Thank you! Keep in touch or join our teams! We’re hiring: Game Programmers Game Artists Game Writer …and more! marek.rosa@keenswh.com twitter.com/marek_rosa blog.marekrosa.org www.KeenSWH.com www.GoodAI.com