McKinsey predicts that AI and robotics will create $50 trillion of value over the next 10 years. Many predict that the recent technology of “deep learning” will be a big part of the transformation. Over 250 deep learning startup companies have attracted more than $1 billion of venture investment in the past year. Deep learning systems have recently broken records in speech recognition, image recognition, image captioning, translation, drug discovery and other tasks. Why is this happening now and how is it likely to play out? We review the development of AI and the pendulum swings between the “neats” and the “scruffies”. We describe traditional approaches to semantics through logics and grammars and the new deep learning vector semantics. We relate it to Roger Shepard’s cognitive geometry and the structure of biological networks. We also describe limitations of deep learning for safety and regulation. We show how it fits into the rational agent framework and discuss what the next steps may be.
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Exosphere Chile Talk: Semantics, Deep Learning, and the Transformation of Business
1. Chile: Semantics, Deep Learning,
and the
Transformation of Business
Steve Omohundro, Ph.D.
PossibilityResearch.com
SteveOmohundro.com
SelfAwareSystems.com
http://discovermagazine.com/~/media/Images/Issues/2013/Jan-Feb/connectome.jpg
3. Multi-Billion Dollar Investments
• 2013 Facebook – AI lab, DeepFace
• 2013 Yahoo - LookFlow
• 2013 Ebay – AI lab
• 2013 Allen Institute for AI
• 2013 Google – DNNresearch, SCHAFT, Industrial Perception,
Redwood Robotics, Meka Robotics, Holomni, Bot & Dolly, Boston
Dynamics
• 2014 IBM - $1 billion in Watson
• 2014 Google - DeepMind $500 million
• 2014 Vicarious - $70 million
• 2014 Microsoft – Project Adam, Cortana
• 2015 Fanuc – Machine Learning for Robotics
• 2015 Toyota – $1 billion AI and Robotics Lab, Silicon Valley
4. McKinsey: $50 Trillion to 2025
http://www.mckinsey.com/insights/business_technology/disruptive_technologies
6. Internet of Things: $15 Trillion to 2025
100 Billion devices by 2025
Cars, Appliances, Cameras, Meters, Wearables, etc.
https://www.summitbusiness.net/images/Internet.jpg
http://www.forbes.com/sites/gilpress/2014/08/22/internet-of-things-by-the-numbers-market-estimates-and-forecasts/
7. Robot Manufacturing: $10 Trillion to 2025
Work 24 hours/day
No breaks, food, medical
Don’t quit, get bored, get depressed
Work anywhere
Hazards OK
Don’t leak secrets
Work well with others
Easy to replicate
http://thisisrealmedia.com/2014/06/19/robotics-and-ethics-the-smart-car-by-ron-parlato/
8. Foxconn Technology Group
• World’s largest contract
manufacturer
• Assembles 40% of all consumer
electronics
• iPhone, iPad, Kindle, Xbox,
Playstation 4, etc.
• 1.3 million employees, $8K
salary
• Employee suicides
• “Foxbot” robots, cost $25K, 2nd
generation now
• Building 30K robots/year
http://www.tomshardware.com/news/foxcponn-apple-iphone-ipad-robot,19088.html
10. March 2015: China Brain
http://www.scmp.com/lifestyle/technology/article/1728422/head-chinas-google-wants-country-take-lead-developing
Robin Li Yanhong, CEO of Baidu proposed a state-level
Chinese initiative to develop AI
“comparable to the Apollo space programme”.
14. https://d185ox70mr1pkc.cloudfront.net/post_image_teaser/1403883171000-uber-force.png
World’s largest job creator: 50,000 per month
http://www.businessinsider.com/uber-offering-50000-jobs-per-month-to-drivers-2015-3
Uber valuation: $51 billion, 20% of fares
http://www.wsj.com/articles/ubers-new-funding-values-it-at-over-41-billion-1417715938
Center for research on self-driving cars
http://bits.blogs.nytimes.com/2015/02/02/uber-to-open-center-for-research-on-self-driving-cars/?_r=0
36 second wait, $.50/mile, 100% of fares
http://zackkanter.com/2015/01/23/how-ubers-autonomous-cars-will-destroy-10-million-jobs-by-2025/
16. April 2014: Chinese WinSun 3D printed
10 houses, 2100 sq ft, $4800
http://m.newsru.co.il/realty/20jan2015/3d_house_i101.html
17. WinSun 3D printed 12,000 sq ft villa
http://3dprint.com/38144/3d-printed-apartment-building/
US Building construction: $1 Trillion/yr
5.8 million employees
23. “Neats” vs. “Scruffies”
http://news.stanford.edu/news/2003/june18/mccarthy-618.html http://www.bbc.co.uk/timelines/zq376fr
1963: John McCarthy
Stanford AI Lab
Mathematically Precise
Thinking = Logical Inference
Semantic Representations
1963: Marvin Minsky
MIT MAC AI Group
Self-Organized
Adaptive Elements
Machine Learning
Emergent Semantics
24. 1957: Rosenblatt’s “Perceptron”
http://www.rutherfordjournal.org/article040101.html
“The embryo of an
electronic computer that [the
Navy] expects will be able to
walk, talk, see, write,
reproduce itself and be
conscious of its existence."
https://en.wikipedia.org/wiki/Perceptron
https://upload.wikimedia.org/wikipedia/commons/3/31/Perceptron.svg
http://bio3520.nicerweb.com/Locked/chap/ch03/3_11-neuron.jpg
32. 1962: Roger Shepard Cognitive Geometry
http://link.springer.com/article/10.1007/BF02289630
https://psychlopedia.wikispaces.com/mental+rotation
33. Word2Vec – Mikolov 2013
• Distributional Semantics – Firth 1957
• Represent words by vectors
• Close vectors represent similar contexts
• Certain relations represented by translation:
King – Man + Woman = Queen
• Also tense, temperature, location, plurals,…
http://deeplearning4j.org/word2vec.html
34. Why? Same context shift for all male -> female
https://drive.google.com/file/d/0B7XkCwpI5KDYRWRnd1RzWXQ2TWc/edit?usp=sharing
2013 Mikolov:
The man ate his lunch.
The king ate his lunch.
The woman at her lunch.
The queen ate her lunch.
35. More Semantic Relations
• Paris – France + Italy = Rome
• Human – Animal = Ethics
• Obama – USA + Russia = Putin
• Library – Books = Hall
• Biggest – Big + Small = Smallest
• Ethical – Possibly + Impossibly = Unethical
• Picasso – Einstein + Scientist = Painter
• Forearm – Leg + Knee = Elbow
• Architect – Building + Software = Programmer
https://code.google.com/p/word2vec/
http://byterot.blogspot.com/2015/06/five-crazy-abstractions-my-
deep-learning-word2doc-model-just-did-NLP-gensim.html
http://arxiv.org/pdf/1301.3781.pdf
http://deeplearning4j.org/word2vec.html
37. Deep Neural Net Face Recognition
Google FaceNet, June 2015
Record accuracy 99.63% on Labeled Faces in the Wild dataset
Cuts best previous error rate by 30%
22 layer feedforward net, 140M weights, 1.6 GFLOP/image, conv/pool/norm
Trained on triples pushing same faces together, different apart
http://arxiv.org/abs/1503.03832
https://github.com/cmusatyalab/openface
CMU OpenFace, Oct. 2015
Open Source version of FaceNet
84.83% accuracy, <.1 training faces
38. Cheap Cameras
+
Face Recognition
+
Body Recognition
=
Brin’s “Transparent Society”
http://www.ebay.com/sch/i.html?_nkw=cmos+image+sensor&_sop=15http://www.aliexpress.com/cheap/cheap-image-sensor-module.html
$3.20 on Alibaba $2.95 on ebay
http://fossbytes.com/facebook-can-now-recognize-you-in-photos-without-even-seeing-your-face/
http://thenextweb.com/dd/2015/10/15/watch-this-open-source-program-recognize-faces-in-real-time/
https://www.newscientist.com/article/mg21528835-600-cameras-know-you-by-your-walk/
http://www.amazon.com/Transparent-Society-Technology-Between-Privacy/dp/0738201448/ref=sr_1_1
42. Recurrent Net Hallucinates C Code
http://karpathy.github.io/2015/05/21/rnn-effectiveness/
Karpathy: 464MB of C code, 3 layer LSTM, 10 million parameters
43. The rat escaped.
The rat the cat attacked escaped.
The rat the cat the dog chased attacked escaped.
http://arlingtonva.s3.amazonaws.com/wp-content/uploads/sites/25/2013/12/rat.jpg
48. Other Issues
• Typically have problems to solve rather than
reinforcement signals
• Want confidence that system solves problem
• Want confidence in no unintended behaviors
• Systems often have to obey legal, corporate,
or design constraints
http://78813809ba6486e732cd-642fac701798512a2848affc62d0ffb0.r60.cf2.rackcdn.com/465DAB1D-1F8E-4164-8D18-3BFD150E02F4.jpg
49. Technology Needs Semantics!
• Analyzing camera, sensor, weather data
• Better search, question answering, info
• Analysis and optimization of business processes
• Health monitoring, medical diagnosis
• Financial markets trading, stabilization
• Autonomous cars, trucks, boats, subs, planes
• Pollution monitoring and cleanup
• Improved robotic manufacturing
• Software and Hardware design
50. Approaches to Semantics
• Montague – map into Typed Lambda Calculus
• Denotational – map into CS Domains
• Mathematical – map into Set Theory
• Categorical – map into Category Theory
• Distributional – Statistics of Contexts
http://engineering.missouri.edu/wp-content/uploads/australian-cloudy-sky.jpg
Representation, Encoding, Learning,
Communication, Reasoning
51. New Possibilities Coming Soon!
PossibilityResearch.com
http://flinttown.com/wp-content/uploads/2015/07/Fireworks.jpg