présentation de bruno schroder au 20e #mforum (07/12/2016)
TRANSCRIPT
Le mobile est-il soluble dans l’intelligence
artificielle?
Bruno Schröder, National Technology Officer Microsoft BeLux
20ème #mforum. Nouveaux territoires mobiles2016
Computer pioneer Alan Kay:
“The best way to predict the future is to invent it.” In the A.I. context, it basically means Stop predicting what the future will be like and create it in a principled way.
Microsoft forms new 5,000-person AI divisionMicrosoft CEO Satya Nadella on how AI will transform his company “AI is at the intersection of our
ambitions,” Nadella said, noting
how it will allow us “to reason
over large amounts of data and
convert that into intelligence.”
“We are on the cusp of a
paradigm shift in computing that
is unlike anything we have seen in
decades,”
He likened AI to the arrival of
books and the web and joked
that we will soon create so much
data that “we are getting to a
point where we don’t even know
what to name things.”
Current State of AI
• General AI ( Artificial General Intelligence)• Kind of HAL in Movie “2001”
• Won’t come for decades
• Goal and/or Threat: Technological singularity (Superintelligence)
• Narrow AI ( addresses specific tasks )• i.e language translation, self driving cars, assistants, image recognition, health
• Remarkable progress in last decade
• Broad societal benefits and economic agility
• Examples: Machine Learning, Deep Learning
• No magic black box. It is a statistical process
• All current products are here
Bing maps
launches
What’s the
best way
home?
Microsoft
Research
formed
Kinect
launches
What does
that motion
“mean”?
Azure Machine
Learning GA
What will
happen next?
Hotmail
launches
Which email is
junk?
Bing search
launches
Which
searches are
most relevant?
Skype
Translator
launches
What is that
person saying?
Microsoft & Machine LearningAnswering questions with experience
1991 201420091997 201520102008
Machine learning is pervasive throughout Microsoft products.
Inflection point: Why now?
• 2010 algorithmic breakthrough
• Cloud: Cheap compute (Quiz)
• Cloud: Cheap storage++ (Quiz)
• Huge data sets of various kinds for statistical learning
• Monetization opportunity & available business models
• Skills and tooling
Classic Programming vs. Machine LearningProgram = Algorithm
Written by humansSpecific to defined taskAlgorithm is “fixed”Algorithm “easy” to describe
Written by softwareGoal: ability to generalizeAlgorithm depends on training dataAlgorithm can “morph” over time
A major paradigm shift
• Solutions based on logic and crafted by hand
Solutions based on probabilities and learned from data
Developing a Machine Learning Program
Learning ExperimentingTesting
Deployment
Phase Process Learning Styles
SupervisedUnsupervised
Self reinforcedHybrid /w humans
28,2
25,8
16,4
11,7
7,3 6,75,1
3,5
ILSVRC 2010
NEC America
ILSVRC 2011
Xerox
ILSVRC 2012
AlexNet
ILSVRC 2013
Clarifi
ILSVRC 2014
VGG
ILSVRC 2014
GoogleNet
Human
Performance
ILSVRC 2015
ResNet
ImageNet Classification top-5 error (%)
8 layers
19 layers22 layers
152 layers
8 layers
Fish or
stone???
Cognitive services
• Principles• Augment human abilities & experiences• Trustworthy • Inclusive & respectful
• Participants• People, Digital Assistants, Bots
• Platform• Human language is the new UI• Bots are the new apps; digital assistants are meta apps• Intelligence infused into all interactions
• “Democratizing AI”
AI Playgrounds
• The race for Digital Assistants (MS, Google, Apple, FB, Amazon)
• Tapping the Enterprise Opportunity (MS, IBM, AWS)
Co-opetition mode: i.e. Partnership on AI (MS, Google, FB, IBM, Amazon)
Nouveaux champs d’action
• Fusionner les perceptions humaines et machines
• Modèles du monde et des humains
• Complémentarité de l’intelligence humaine et de la machine
• Coordination et initiative (de la machine)
• Les surprises probables
2016/12/15 24
1. Empathy
2. Education (knowledge and skills)
3. Creativity
4. Judgment and accountability
Data Ethics Principles for Humanistic Approach to AI
https://news.microsoft.com/cloudforgood/resources.html
Cloud for Global Good – AI Pillar• Modernize laws and practices to enable AI
Data access, copyright, trade secrets Safety and liability Transparency Publicly available data
• Assess privacy law in light of the benefits of AI Unlimited innovation Repurposing data AI’s predictive power
• Ethical principles Transparency Non-Discrimination Multi-stakeholder collaboration Industry standards Cloud-powered AI
Bringing it all together The Seeing AI App
Computer Vision, Image, Speech Recognition, NLP,
and ML from Microsoft Cognitive Services
Watch Video HereRead Blog Here
AnnouncementsPartnership on AI
Sept 29th: AI & TnR reorg press release
Sept 29th: AI & TnR announcement (Harry Shum)
Oct 6th: 5m$ data science MSRC & The Alan Turing Institute
Nov 15th: MS and OpenAI (Harry Shum)
MS Research – AI and ML group
Papers• Stanford – Artificial Intelligence and Life in 2030
• Whitehouse – Preparing for the future of AI
• Whitehouse – Responses
• Whitehouse – A Research agenda for AI
• Eric Horvitz – AI supporting people and society
• QMU – Machine Learning with Personal Data
• QMU – Responsibility and Machine Learning – Part
of a Process
• QMU - Responsibility, Autonomy and Accountability
– Liability for ML
Microsoft productsAzure ML
Cortana Intelligence Suite
Microsoft Cognitive Services
Language Understanding Intelligence Services (LUIS)
AI in Office 365
Office MyAnalytics
ProjectsCaptionbot.ai
How-Old.net
Mimicker Alarm
TwinsOrNot.net
Emotion Demo
Microsoft Pix
MileIQ
SwiftKey