Interactive Powerpoint_How to Master effective communication
You say you want a revolution? Benchmark Roadmap to solve AI/IA
1. You say you want a
revolution?
October 8, 2017
https://www.slideshare.net/spohrer/revolution-20171008-v23
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 1
2. TED Arai Todai Robot
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 2
… when will
your smartphone
be smart enough to
pass a university
entrance exam?
3. “AI will change the world?
Who will change AI?”
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 3
4. “These amazing technologies must be able to
help people like myself…”
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 4
5. Questions
• What is the timeline for solving AI and IA?
• Who are the leaders driving AI progress?
• What will the biggest benefits from AI be?
• What are the biggest risks associated with AI, and are they real?
• What other technologies may have a bigger impact than AI?
• What are the implications for stakeholders?
• How should we prepare to get the benefits and avoid the risks?
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 5
6. Every 20 years, compute costs are down by 1000x
• Cost of Digital Workers
• Moore’s Law can be thought of as
lowering costs by a factor of a…
• Thousand times lower
in 20 years
• Million times lower
in 40 years
• Billion times lower
in 60 years
• Smarter Tools (Terascale)
• Terascale (2017) = $3K
• Terascale (2020) = ~$1K
• Narrow Worker (Petascale)
• Recognition (Fast)
• Petascale (2040) = ~$1K
• Broad Worker (Exascale)
• Reasoning (Slow)
• Exascale (2060) = ~$1K
610/8/2017 (c) IBM 2017, Cognitive Opentech Group
2080204020001960
$1K
$1M
$1B
$1T
206020201980
+/- 10 years
$1
Person Average
Annual Salary
(Living Income)
Super Computer
Cost
Mainframe Cost
Smartphone Cost
T
P
E
T P E
AI Progress on Open Leaderboards
Benchmark Roadmap to solve AI/IA
7. GPD/Employee
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 7
(Source)
Lower compute costs translate into increasing productivity and GDP/employees for nations
Increasing productivity and GDP/employees should translate into wealthier citizens
AI Progress on Open Leaderboards
Benchmark Roadmap to solve AI/IA
8. Leaderboards Framework
AI Progress on Open Leaderboards - Benchmark Roadmap
2015 2018 2021 2024 2027 2030 2033 2036
Perceive World Develop Cognition Build Relationships Fill Roles
Pattern
recognition
Video
understanding
Memory Reasoning Social
interactions
Fluent
conversation
Assistant &
Collaborator
Coach &
Mediator
Speech Actions Declarative Deduction Scripts Speech Acts Tasks Institutions
Chime Thumos SQuAD SAT ROC Story ConvAI
Images Context Episodic Induction Plans Intentions Summarization Values
ImageNet VQA DSTC RALI General-AI
Translation Narration Dynamic Abductive Goals Cultures Debate Negotiation
WMT DeepVideo Alexa Prize ICCMA AT
Learning from Labeled Training Data and Searching (Optimization)
Learning by Watching and Reading (Education)
Learning by Doing and being Responsible (Exploration)
2015 2018 2021 2024 2027 2030 2033 2036
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 8
Which experts would be really surprised if it takes less time… and which experts really surprising if it takes longer?
9. Icons of AI Progress
• 1956: Dartmouth Conference
organized by:
• John McCarthy (Dartmouth, later
Stanford)
• Marvin Minsky (MIT)
• and two senior scientists:
• Claude Shannon (Bell Labs)
• Nathan Rochester (IBM)
• 1997: Deep Blue (IBM) - Chess
• 2011: Watson Jeopardy! (IBM)
• 2016: AlphaGo (Google DeepMinds)
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 9
11. AI Leaders
• Who is winning?
• Regions China vs USA vs EU vs ROW
• Companies Microsoft vs Google vs IBM
• Leaderboards
• SQuAD – Question Answering
• EFF Measuring AI Progress
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 11
12. AI to IA Timeline: Hard unsolved AI problems
• 2012-2017 AI Pattern Recognition and
Learning from Massive Labeled Data
• Speech, image, translation, driverless, games
• Chatbots as digital assistants
• 2018 Video Understanding
• 2021 Episodic Memory
• 2022 Learning from Watching
• 2024 Commonsense Reasoning
• 2026 Learning from Reading
• 2028 Learning from Doing
• 2030 Fluent Conversation
• 2031-2039 Cognitive Collaborator and
Mediator; Intelligence Augmentation (IA)
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 12
13. AI Benefits
• Access to expertise
• “Insanely great” labor productivity for trusted service providers
• Digital workers for healthcare, education, finance, etc.
• Better choices
• ”Insanely great” collaborations with others on what matters most
• AI for IA = Augmented Intelligence and higher value co-creation interactions
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 13
14. AI Risks
• Job Loss
• Shorter term bigger risk
= de-skilling
• Super-intelligence
• Shorter term bigger risk
= bad actors
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 14
15. Other Technologies: Bigger impact? Yes.
• Augmented Reality (AR)/
Virtual Reality (VR)
• Game worlds
grow-up
• Blockchain/
Security Systems
• Trust and security
immutable
• Advanced Materials/
Energy Systems
• Manufacturing as cheap,
local recycling service
(utility fog, artificial leaf, etc.)
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 15
16. Stakeholders
• Individuals
• Families
• Businesses and
other Organizations
• Industry Groups
• Regional
Governments:
• Cities
• States
• Nations
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 16
17. Be Prepared
• Understand open AI code + data
+ models + stacks + communities
• Leaderboards
• Ethical conduct
• Learn 3 R’s of IBM’s Cognitive
Opentech Group (COG)
• Read arXiv
• Redo with Github
• Report with Jupyter notebooks on
DSX and/or leaderboards
• Improve your team’s skills of
rapidly rebuilding from scratch
• Build your open code eminence
• Understand open innovation
• Communities + Leaderboards
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 17
1972 used
Punch cards
2016 used
IBM Watson
Open APIs to win…
18. Cupertino Teens
• IBM Watson on Bluemix
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 18
AI for NLP
entity identification
19. Courses
• 2015
• “How to build a cognitive system for Q&A task.”
• 9 months to 40% question answering accuracy
• 1-2 years for 90% accuracy, which questions to reject
• 2025
• “How to use a cognitive system to be a better professional X.”
• Tools to build a student level Q&A from textbook in 1 week
• 2035
• “How to use your cognitive mediator to build a startup.”
• Tools to build faculty level Q&A for textbook in one day
• Cognitive mediator knows a person better than they know
themselves
• 2055
• “How to manage your workforce of digital workers.”
• Most people have 100 digital workers.
10/8/2017 19
20. Headlines
• 2017 Popular
• “AI vs People”
• “X-Y team up to invest big in AI”
• 2025 Commonplace
• “People using AI to become better at
their professions, serving others.”
• “Teenagers using AI to solve challenges,
and improve their communities.”
• 2085 Resilience
• “Teams competing to rapidly rebuild
socio-economic-technical systems (wise
service systems) from scratch”
• “U.N. Pluto-base makes major discovery
about nature of universe. U.F.P.
established.”
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 20
IEEE 2017
21. IBM-MIT $240M over 10 year AI mission
10/8/2017 (c) IBM 2017, Cognitive Opentech Group 21
Please feel free to share and cite:
Spohrer, JC (2017) You say you want a revolution? Presentation draft October 8, 2017 for AAAI Fall Symposium. https://www.slideshare.net/spohrer/smart-20171008-v23
Title slide video URL: https://www.youtube.com/watch?v=BGLGzRXY5Bw
Opening ten seconds of the above Beatles video shows the self-astonishment of what a team of people can do with lots of practice and amazingly cool technologies. My brother Scott reminds me of John Lennon – not only in early long-hair look, but love of writing and singing songs about building a better world. Not utopia, he never believed in that – he knew life was hard by design, he just hoped for a world with more open minded people working together for the common good. Perhaps AI systems that know us better than we know are selves can help us all become more open minded.
Thanks to IBM colleagues, and IBM visitors including Daniel Pakkala (VTT Finland)– see http://service-science.info/archives/4741
Thanks to “OECD 2016 – One year later group” as well: Elliot Stuart, Ken Forbus, Moshe Vardi, Frank Levy, Vijay Saraswat, Michael Witbrock, Alistair Nolan, Art Graesser, Charles Fadel, Ernest Davis, Jill Burstein, Michael Handel, Jerry Hobbs, Rebecca Passonneau, Mark Steedman
What is beyond Exascale? Zetta (21), Yotta (24)
Time dimension (x-axis) is plus or minus 10 years….
Daniel Pakkala (VTT)
URL: https://aiimpacts.org/preliminary-prices-for-human-level-hardware/
Dan Gruhl:
https://www.washingtonpost.com/archive/business/1983/11/06/in-pursuit-of-the-10-gigaflop-machine/012c995a-2b16-470b-96df-d823c245306e/?utm_term=.d4bde5652826
In 1983 10 GF was ~10 million.
That's 24.55 million in today's dollars.
or 2.4 billion for 1 TF in 1983
Today 1 TF is about $3k http://www.popsci.com/intel-teraflop-chip
URL: https://en.wikipedia.org/wiki/History_of_artificial_intelligence
URL: http://www.businessinsider.com/infographic-ai-effect-on-economy-2017-8
Today’s infographic comes from the Extraordinary Future 2017, a new conference in Vancouver, BC that focuses on emerging technologies such as AI, autonomous vehicles, fintech, and block
http://extraordinaryfuture.com/e/extraordinary-future-2017-71chain tech.
Nathaniel Rochester: In 1948, Rochester moved to IBM where he designed the IBM 701, the first general purpose, mass-produced computer. He wrote the first symbolic assembler, which allowed programs to be written in short, readable commands rather than pure numbers or punch codes.
1950 Nathaniel Rochester (IBM) 701 first commercial computer that did super-human levels of numeric calculations routinely. He worked at MIT on arithmetic unit of WhirlWind I programmable computer.
Dota 2 is most recent August 11, 2017 as a super-human game player in Valve Dota 2 competition – Elon Musk’s OpenAI result.
Miles Bundage tracks gaming progress: http://www.milesbrundage.com/blog-posts/my-ai-forecasts-past-present-and-future-main-post
DOTA2: https://blog.openai.com/more-on-dota-2/
Who is winning: https://www.technologyreview.com/s/608112/who-is-winning-the-ai-race/
Leaderboards and reproducibility:
Hugo Larochelle (Google Brain) (@hugo_larochelle) 8/21/17, 7:36 AM
My slides for my talk at ICML 2017 Reproducibility Workshop, on incentives for open source and on open research:
https://drive.google.com/file/d/0B8lLzpxgRHNQZ0paZWQ0cTcxMlNYYnc0TnpHekMxMjVBckVR/view
Slide 20: Conclusions: "Open source is the key to better reproducibility"
Chatbots: http://kasisto.com/
Learning by doing related
The team has successfully applied the One Button Machine in various data science competitions where it outperformed most human teams and ranked among the top 16螄% of participants
One Button Machine works by traversing the graph defined by the entities (tables) and relations (primary/foreign keys) of a relational database. The aggregation functions can be specified by the user, or chosen generically for certain data types. To deal with the combinatorial explosion of related entities, the One Button Machine deploys heuristics and sub-sampling strategies. Scalability to big databases is achieved by dynamic caching of intermediate results and a parallelisable implementation in Apache Spark, a distributed computing framework for analysing massive amounts of data.
http://www.eetindia.co.in/news/article/ibm-develops-low-level-task-automation
Learning by doing related.
The nature of reality changes when there is more than one intelligent species, and we are not the smartest.
The nature of reality also changes when the cost of exploring alternate experience pathways are made less risky – the notions of time and identity changes as a result.
Mitigate risks and harvest benefits of existence, by learning to evermore efficiently and rapidly rebuild from scratch to higher states of value and capability of entities.
The evolving ecology of service system entities their value co-creation and capability co-elevation mechanisms, as well as their capabilities, constraints, rights, and responsibilities at each stage in time. Human progress as well as the development of individuals, and the arc of institutions can be viewed in this way. Entities exist as individuals and populations. Generations of entities, generations of species (populations), generations of individuals (cohorts).
URL: http://www.mercurynews.com/2016/08/04/cupertino-teens-score-20000-for-24-hours-of-work/
Karan Mehta and Anish Krishnan
Here is what I tell students....
... to try to provoke their thinking about the cognitive era:
(0) 2015 - about 9 months to build a formative Q&A system - 40% accuracy;
- another 1-2 years and a team of 10-20, can get it to 90% accuracy, by reducing the scope ("sorry that question is out of scope")
- today's systems can only answer questions, if the answers are already existing in the text explicitly
- debater is an example of where we would like to get to though in 5 years: https://www.youtube.com/watch?v=7g59PJxbGhY
- more about the ambitions at http://cognitive-science.info
(1) 2025: Watson will be able to rapidly ingest just about any textbooks and produce a Q&A system
- the Q&A system will rival C-grade (average) student performance on questions
(2) 2035 - above, but rivals C-level (average) faculty performance on questions
(3) 2035 - an exascale of compute power costs about $1000
- an exascale is the equivalent compute of one person's brain power (at 20W power)
(4) 2035 - nearly everyone has a cognitive mediator that knows them in many ways better than they know themselves
- memory of all health information, memory of everyone you have ever interacted with, executive assistant, personal coach, process and memory aid, etc.
(5) 2055 - nearly everyone has 100 cognitive assistants that "work for them"
- better management of your cognitive assistant workforce is a course taught at university
In 2015, we are at the beginning of the beginning or the cognitive era...
In 2025, we will be middle of beginning... easy to generate average student level performance on questions in textbook....
In 2035, we will be end of beginning (one brain power equivalent)... easy to generate average faculty level performance on questions in textbook....
http://www.slideshare.net/spohrer/spohrer-ubi-learn-20151103-v2
By 2055, roughly 2x 20 year generations out, the cognitive era will be in full force.
Cellphones will likely become body suits - with burst-mode super-strength and super-safety features:
Suits - body suit cell phones
Cognitive Mediators will read everything for us, and relate the information to us - and what we know and our goals.
Think combined personal coach, executive assistant, personal research team....
The key is knowing which problem to work on next - see this long video for the answer - energy, water, food, wellness - and note especially the wellness suit at the end:
https://www.youtube.com/watch?v=YY7f1t9y9a0&index=10&list=WL
Do not be put off by the beginning of the video - it is a bit over hyped and trivial, to say the leasat... but the projects are really good if you have the patience to watch.
URL: https://spectrum.ieee.org/static/ai-vs-doctors
U.N. = United Nations
U.F.P. = United Federation of Planets