Building Artificial General Intelligence · What about Quantum Computing? •First proposed by...
Transcript of Building Artificial General Intelligence · What about Quantum Computing? •First proposed by...
![Page 1: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/1.jpg)
Building Artificial General
Intelligence
Peter MorganCEO TURING.AI
![Page 2: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/2.jpg)
Outline of Talk
• What is Intelligence?
• Physical Systems
• Biological
• Non-biological
• Deep Learning
• Recap
• AGI
• Overview
• Active Inference
• Building AGI
• Conclusions
© Peter Morgan, QCon March 2019 2
![Page 3: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/3.jpg)
TL;DRUsing the physical principles of active inference, I believe we can build AGI systems over the
coming years.
© Peter Morgan, QCon March 2019 3
![Page 4: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/4.jpg)
Motivation
Why do we want to build AGI?
• To solve (general) intelligence• Then use it to solve everything else
• Medicine, cancer, brain disease (Alzheimer’s ...), longevity, physics, maths, materials science, social
• Naturally extend to superintelligent systems
© Peter Morgan, QCon March 2019 4
![Page 5: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/5.jpg)
What is general intelligence? (Ask the audience)
© Peter Morgan, QCon March 20195
AlphaGo AlphaStar Deep Blue
IBM Watson
![Page 6: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/6.jpg)
What is intelligence?
No matter how impressive, these are all examples of
Narrow AI
© Peter Morgan, QCon March 20196
![Page 7: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/7.jpg)
What is Intelligence?
© Peter Morgan, QCon March 20197
![Page 8: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/8.jpg)
How far have we come?
• Logical-mathematical 50% (calculation not creativity)
• Linguistic 50% (statistical only)
• Spatial 50% (SLAM)
• Bodily-kinesthetic 30% (Atlas)
• Naturalistic 10%
• Musical 50% (Google Magenta project)
• Interpersonal 10%
• Intrapersonal 5%?
• Existential 0%
© Peter Morgan, QCon March 2019 8
![Page 9: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/9.jpg)
How will we get to AGI?
PhysicsComputer Science
Neuroscience
© Peter Morgan, QCon March 2019
It takes a village to create an AGI
Psychology
9
![Page 10: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/10.jpg)
Outline
• What is Intelligence?
• Physical Systems• Biological• Non-biological
• Deep Learning• Recap
• AGI• Overview• Active Inference• Building AGI
• Conclusions
© Peter Morgan, QCon March 2019 10
![Page 11: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/11.jpg)
Intelligence in Physical Systems
• Biological (neuroscience)• Plants, bacteria, insects, mammalian
• Non-biological (computer science)• CPU, GPU, FPGA, ASIC • Neuromorphic
• Quantum (physics)• Quantum processors, algorithms• Biology?
© Peter Morgan, QCon March 2019 11
![Page 12: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/12.jpg)
BiologyWe would like to build human level intelligence
© Peter Morgan, QCon March 201912
![Page 13: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/13.jpg)
Biological Systems are Hierarchical
Intelligence is “emergent”
© Peter Morgan, QCon March 2019 13
![Page 14: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/14.jpg)
Biological Neuron
Microstructure
Very complex
© Peter Morgan, QCon March 2019 14
![Page 15: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/15.jpg)
© Peter Morgan, QCon March 2019 15
![Page 16: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/16.jpg)
© Peter Morgan, QCon March 2019 16
Different Morphologies
![Page 17: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/17.jpg)
Cortical columns in the cortex
Approx. 2 million in human brain
Structure is repeated
© Peter Morgan, QCon March 201917
![Page 18: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/18.jpg)
Human Connectome
General intelligence occurs
at this level
© Peter Morgan, QCon March 2019 18
![Page 19: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/19.jpg)
Central Nervous
System (CNS)
Bodily intelligence
© Peter Morgan, QCon March 2019 19
![Page 20: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/20.jpg)
Social Systems
Cloud roboticsSwarm robotics
© Peter Morgan, QCon March 2019 20
![Page 21: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/21.jpg)
Non-biological Hardware
• Digital• CPU, GPU, FPGA, ASIC
• Neuromorphic• Various architectures
• Quantum• Different types
© Peter Morgan, QCon March 2019 21
![Page 22: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/22.jpg)
Digital Computing
• Abacus (mechanical, 2700 BCE)• Charles Babbage (1830)• Ada Lovelace• Vacuum tubes (electronic, 1900)• Alan Turing (1930’s)• Von Neumann• ENIAC (1946)• Transistor (Bardeen, Brattain, Shockley, 1947)• Intel (1968)• ARM (1990)• Nvidia (1993)• ASICs – TPU (2016)
© Peter Morgan, QCon March 2019 22
![Page 23: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/23.jpg)
© Peter Morgan, QCon March 2019 23
![Page 24: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/24.jpg)
Cray-1 1976
160 MFlops
© Peter Morgan, QCon March 2019 24
![Page 25: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/25.jpg)
© Peter Morgan, QCon March 2019
TPU v3 Graphcore IPU
25
Today
![Page 26: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/26.jpg)
Cloud TPU’s
© Peter Morgan, QCon March 2019
Over 100 PetaFlops - 1 billion Cray-1’s!26
![Page 27: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/27.jpg)
Summit US 3 ExaFlops mixed
precision2 tennis courts area
250 Petabytes storage 13MW power$200million
Announced 5 June 2018
No general intelligence© Peter Morgan, QCon March 2019 27
![Page 28: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/28.jpg)
What are we missing?
• Brain computation is of order of PetaFlops, 1.5kg, 30W
• Clearly, shear digital horsepower (von Neumann architecture) is NOT going to get us to generally intelligent systems, or we would have them by now (have ExaFlops)
• We already know the brain (biology) uses a different architecture - memory and compute are combined (memristors)
• Perhaps we should be looking at the brain for inspiration - bioplausible architectures
© Peter Morgan, QCon March 2019 28
![Page 29: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/29.jpg)
Introducing Neuromorphic
Computing
• Biologically inspired
• First proposed Carver Mead, Caltech, 1980’s
• Uses analogue signals – spiking neural networks (SNN)
• SpiNNaker, BrainScaleS, TrueNorth, Intel Loihi
• Startups - Knowm, Spaun, …
• Up to 1 million cores, 1 billion “neurons” (mouse)
• Need to scale 100X à human brain
• Relatively low power – 1,000X less than digital
• SpiNNaker is available today in the cloud – try it out!
© Peter Morgan, QCon March 2019 29
![Page 30: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/30.jpg)
SpiNNaker Neuromorphic
Computer
© Peter Morgan, QCon March 201930
![Page 31: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/31.jpg)
© Peter Morgan, QCon March 201931
![Page 32: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/32.jpg)
Analog v Digital
© Peter Morgan, QCon March 2019
BrainScalesGoogle TPU’s
32
![Page 33: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/33.jpg)
What about Quantum
Computing?
• First proposed by Richard Feynman, Caltech, 1980’s• Qubits – spin 1, 0, and superposition states • Nature is fundamentally probabilistic at atomic
scale• Several approaches - superconductors, trapped
ions, photonic, topological • Several initiatives
• IBM, D-Wave, Rigetti, Google, Intel, Microsoft, ...
• Applications – optimization, drug discovery, machine learning, cryptography• Does Nature use quantum computation? • See my talk tomorrow @ 3pm.
© Peter Morgan, QCon March 2019 33
![Page 34: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/34.jpg)
IBM 50 Qubit Quantum Computer
© Peter Morgan, QCon March 2019 34
![Page 35: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/35.jpg)
Google Bristlecone
© Peter Morgan, QCon March 2019 35
![Page 36: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/36.jpg)
Quantum Logic Gates
© Peter Morgan, QCon March 2019 36
![Page 37: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/37.jpg)
Summary – Four classes of physical computation systemsDigital Neuromorphic
Quantum Biological
© Peter Morgan, QCon March 201937
![Page 38: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/38.jpg)
Four Types of ProcessorsAll information processing systems
© Peter Morgan, QCon March 2019
DigitalBiological
Neuromorphic
Quantum
38
![Page 39: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/39.jpg)
Three Non-biological Stacks
Algorithms
Distributed Layer
OS
Hardware
Digital Neuromorphic Quantum
© Peter Morgan, QCon March 2019 39
![Page 40: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/40.jpg)
© Peter Morgan, QCon March 2019 40
![Page 41: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/41.jpg)
Outline
• What is Intelligence?
• Physical Systems• Biological• Non-biological
• Deep Learning• Recap
• AGI• Overview• Active Inference• Building AGI
• Conclusions
© Peter Morgan, QCon March 2019 41
![Page 42: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/42.jpg)
Early papers
© Peter Morgan, QCon March 2019 42
![Page 43: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/43.jpg)
© Peter Morgan, QCon March 2019 43
![Page 44: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/44.jpg)
Deep Learning
Performance
Image classification© Peter Morgan, QCon March 2019 44
![Page 45: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/45.jpg)
Framework Popularity
© Peter Morgan, QCon March 2019 45TensorFlow 2.0 is released this week!
![Page 46: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/46.jpg)
The fourth industrial revolution will be (is) open source
• ML Frameworks – open source (e.g., TensorFlow)• Operating systems – open source (Linux)• Hardware – open source
• OCP (Open Compute Project) & RISC-V
• Data sets – open source (Internet)• Research – open source (see arXiv, etc.)• Is tremendously accelerating progress
© Peter Morgan, QCon March 2019 46
![Page 47: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/47.jpg)
Outline
• What is Intelligence?
• Physical Systems
• Biological
• Non-biological
• Deep Learning
• Recap
• AGI• Overview
• Active Inference
• Building AGI
• Conclusions
© Peter Morgan, QCon March 2019 47
![Page 48: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/48.jpg)
General Theories of Intelligence
• What do we need?• Different approaches• Active Inference• Building AGI
AGI = Artificial General Intelligence
© Peter Morgan, QCon March 2019 48
![Page 49: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/49.jpg)
The Physics Approach The brain is a physical system so what are the
fundamental physical principles?
Newtonian mechanics – three
laws
Special relativity –invariance of laws under a Lorentz transformation
GR – Principle of Equivalence
Electromagnetism – Maxwell’s equations
Thermodynamics –three laws
Quantum mechanics –uncertainty
principle
Relativistic QM –Dirac equation
Dark energy/dark matter – we don’t
know yet
All of the above = Principle of Least
Action
© Peter Morgan, QCon March 2019 49
![Page 50: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/50.jpg)
The Principle of Least Action(All of physics can be derived from this)
© Peter Morgan, QCon March 2019 50
Please don’t worry about the math in
this section – try to focus on the
concepts
Cambridge University Press, 2018
![Page 51: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/51.jpg)
All known physics – Field theoretic
© Peter Morgan, QCon March 2019 51
Pinnacle of human achievement?
![Page 52: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/52.jpg)
Applied to Intelligence
© Peter Morgan, QCon March 201952
![Page 53: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/53.jpg)
We need a system that can model the (whole) world
© Peter Morgan, QCon March 2019
Intelligence is not just about pattern recognition
It is about modeling the world• Explaining and understanding what we see• Imagining things we could see but haven’t yet• Problem solving and planning actions to make these things real• Building new models as we learn more about the world
53
![Page 54: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/54.jpg)
Theoretical Approaches to AGI
• Helmholtz and others (Statistical physics, late 1800’s)
• Friston – Active Inference • Tishby – Information bottleneck
• Bialek – Biophysics
• Hutter – AIXI
• Schmidhuber – Godel Machine
• All of the above have been worked on for thirty+ years
• Many others
© Peter Morgan, QCon March 2019 54
![Page 55: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/55.jpg)
Active Inference
• Professor Karl Friston, UCL Neuroscience• Based on physics and information theory• Uses the Free Energy Principle • Systems act to minimize their expected free energy• Reduce uncertainty (prediction error)
• It encompasses all interactions and dynamics• Completely general• Applies over all time and distance scales
55
![Page 57: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/57.jpg)
Active Inference Summary
• Biological agents resist the second law of thermodynamics
• They must minimize their average surprise (entropy)
• They minimize surprise by suppressing prediction error (free-energy)
• Prediction error can be reduced by changing predictions (perception)
• Prediction error can be reduced by changing sensations (action)
• Perception recurrent message passing in brain - optimizes predictions
• Action makes predictions come true (and minimizes surprise)
© Peter Morgan, QCon March 2019 57
![Page 58: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/58.jpg)
© Peter Morgan, QCon March 2019 58
Internal and external states separated by a Markov blanket
![Page 59: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/59.jpg)
Active Inference - Information theoretic approach, uses generalised free energy
( ) argmin E [ ( , )] [ ( ) || ( )]
( ) argmin ( , )
( , ) E [ln ( | ) ln ( , | )]
ln ( ) ( , )
( , ) E [ln ( | ) ln ( , | )]
Q
Q
entropy energy
Q
entropy energy
Q F D Q P
Q s F
F Q s P o s
P G
G Q s P o s
t
t t
t t t
t
t t t
p p t p p
p p t
p t p p
p p t
p t p p
= +
=
= -
= -
= -
åå
å
!"#"$ !""#""$
!"#"$ !""#""$
Perceptual inference
Policy selection
( , | )
( , | )
( , ) E [ ( , )]E [ln ( | ) ( | ) ln ( , )]
E [ ( | ) || ( | )] [ ( , ) || ( | ) ( | )]
Q
Q o s
entropy energy
Q o s
expected cost epistemic value (mutual informat
G FQ o Q s P o s
D Q s P s D Q o s Q s Q o
t t
t t
p t t t t
p t t t t t t
p t p t
p p
p p p p
=
= -
= -
!"""#"""$ !"#"$
!""""#""""$ion)
!"""""#"""""$
Generalised free energy – with some care
( | ) :( | )
( ) :
( ) :( | )
( | ) :
P o s tQ o s
o t
P s tP s
P s t
t tt t
t
tt
t
td t
tp
p t
>ì= í £î
>ì= í £î
© Peter Morgan, QCon March 2019 59
Warning: the maths gets ugly
![Page 60: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/60.jpg)
© Peter Morgan, QCon March 2019 60
Active Inference
We have code
![Page 61: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/61.jpg)
From cells to brains
Active states
( , , )aa f s a µ»!External states Internal states
Sensory states
( , , )f s aµµ µ»!
( , , )s ss f s ay w= +!
( , , )f s ay yy y w= +!
Cell
Brain
© Peter Morgan, QCon March 201961
![Page 62: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/62.jpg)
Can we build general intelligence?
• We have • Candidate theories – active inference ✓• Algorithms & software ✓• Hardware – ASIC, neuromorphic ✓• Data sets ✓
• Need to build a complete framework with libraries • A “TensorFlow for general intelligence”• We need software engineers for this part J
• Apollo Project of our time – “Fourth (Intelligence) Revolution”• Steam à electricity à digital à intelligence• Human Brain Project, Deepmind, BRAIN project, China, …
• Should we build AGI/ASI? – safety, ethics, singularity?• Topic for another talk(s) – see ethics track
© Peter Morgan, QCon March 2019 62
![Page 63: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/63.jpg)
Some AGI Projects
© Peter Morgan, QCon March 2019
Deepmind – Demis Hassibis (UK)
OpenCog – Ben Goertzel (US)
Numenta – Jeff Hawkins (US)
Vicarious – Dileep George (US)
OpenAI – Sam Altman (US)
NNAIsense –Jurgen
Schmidhuber (Swiss)
AGI Innovations –Peter Voss (US)
GoodAI – Marek Rosa (Czech)
Curious AI –(Finland)
Eurisko – Doug Lenat (US)
SOAR – CMU ACT-R – CMU Turing.AI (UK) Sigma – Paul Rosenbloom, USC
Plus many, many more
63
![Page 64: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/64.jpg)
Conclusions
• It is obvious to most that Deep Learning is lacking the foundations needed for a general theory of intelligence – it is based on statistics not physics• Active inference is a theory of general intelligence• Deep Learning research groups are now (finally) turning to biology for inspiration• Bioplausible models are starting to appear• Some groups are starting to look at active inference• Real AGI systems in one year? five years? ten years?• Still have to wait for hardware to mature• Neuromorphic might be the platform that gets us there.
© Peter Morgan, QCon March 2019 64
![Page 65: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/65.jpg)
TL;DR
Using the physical principles of active inference, I believe we can
build AGI systems over the coming years.
© Peter Morgan, QCon March 2019 65
![Page 66: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/66.jpg)
Final Word …
https://www.youtube.com/watch?v=7ottuFZYflg© Peter Morgan, QCon March 2019 66
![Page 67: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/67.jpg)
Report downloadhttps://start.activestate.com/oreilly-ebook/
Companywww.turing-ai.co
© Peter Morgan, QCon March 2019 67
![Page 68: Building Artificial General Intelligence · What about Quantum Computing? •First proposed by Richard Feynman, Caltech, 1980’s •Qubits –spin 1, 0, and superposition states](https://reader033.fdocuments.us/reader033/viewer/2022042111/5e8c0d68105ba34e03071337/html5/thumbnails/68.jpg)
© Peter Morgan, QCon March 2019
AMA 10:35am in Guild Room
68