The future of AI by Oren Etzioni from AI2

Presented at AI NEXTCon Seattle 1/17-20, 2018
http://aisea18.xnextcon.com
join our free online AI group with 50,000+ tech engineers to learn and practice AI technology, including: latest AI news, tech articles/blogs, tech talks, tutorial videos, and hands-on workshop/codelabs, on machine learning, deep learning, data science, etc..

1. 1
AI NEXTCon Seattle ‘18
1/17-20th| Seattle
#ainextcon
http://aisea18.xnextcon.com

2. Oren Etzioni
CEO, Allen Institute
for Artificial Intelligence (AI2)
allenai.org
(Professor, University of Washington)
The Future of AI

3. Outline
I.AI, AlphaGo, and Deep learning
II.Is AI an Existential Risk?
III.AI for the Common Good (AI2)
SemanticScholar.org
3

4. What Is AI?
AI is machines doing what humans do:
•Gave rise to a lot of Hollywood scripts
•AI as a noun: “the AI” took over the world…
AI in practice:
•The science & engineering of software to accomplishnarrow tasks once thought
to require human intelligence
4

5. WE HAVE
CREATED
NARROW AI
AI SAVANTS

6. 6
AlphaGoCase Study

7. What Is the Significance Of AlphaGo?
7

8. Haven’t We Seen This Movie Before?
8
Deep Blue beats Kasparov in 1997

9. Go Is Far More Difficult Than Chess
9
Chess Go
BoardSize 8 x 819 x 19
Estimated Branching Factor~35 ~250
Estimated Game Tree Depth!" #$"
Est. Number of possible
games
#"!% #"#%"
Go board position evaluation is challenging

10. Myth: AlphaGoShows The Way To AI
Board games are “black and white”
-Discrete moves
-Win/lose èevaluation function
Labeled data is essentially “infinite”
AlphaGo Zero: Self play.
10

11. Arthur C. Clarke
“Sufficiently advanced technology is
indistinguishable from magic...”
11
Deep Learning isn’t magic!
Source: Etzioni (Wired Magazine 2016)

12. MachineLearning Is 99% HumanWork
Deep Learning inputs:
•Target concept
•Algorithm
•Neural network design
12
“Figuring out howto optimize
something is a CS problem. But
figuring out whatto optimize is not.”
(Nate Silver, paraphrased)

13. My Questions for AlphaGo
•Can you play again?
•Can you play poker?
•Cross the street?
•Can you tell us about
the game?
13
No autonomy.
No.
No.
No.
Super-human performance on a
narrow task, does not translate to
human-level performance in
general!

14. WinogradSchemas (Levesque, 2011)
The large ball crashed right through the table because itwas
made of styrofoam.
It = table
The large ball crashed right through the table because itwas
made of steel.
It = ball
Common-sense knowledge &
tractable reasoning are necessary
for basic understanding!
14
2016 Challenge: 58.33% Correct

15. Paraphrasing Winston Churchill
Deep Learning is not the end,
it is not the beginning of the end,
it's not even the end of the
beginning!
15

16. Outline
I.AI, AlphaGo, and Deep Learning
II.Is AI an Existential Risk?
III.AI for the Common Good (AI2)
SemanticScholar.org
16

17. Artificial
Intelligence (AI) is
Like ‘Summoning
the Demon’
17Source: Terminator Salvation (2009)
Elon Musk Warns…

18. What do the Experts Think?
18Source: Etzioni (MIT Tech Review 2016)

19. Working to Prevent AI from turning evil is like disrupting
the Space Program to prevent overpopulation on Mars.
19
Andrew Ng

20. Rodney Brooks
20
“If you’re worried about The Terminator,
just keep the door closed.”

21. Hollywood Myth: SkyNet
AI will be hegemonic, monolithic, and evil
One AI can be utilized to check
and counterbalance others (Etzioni & Etzioni, 2016)
Research on AI Guardians
Fact: Intelligence ≠Autonomy
“Autonomous cars” is a misnomer
21

22. Outline
I.AI, AlphaGo, and Deep learning
II.Is AI an Existential Risk?
III.AI for the Common Good (AI2)
SemanticScholar.org
22

23. 23
M
Common Good
Allen Institute For AI (AI2)

24. 24
Intelligent Cars Will reduce accidents

25. 25
Overcoming Information Explosion

26. “It's theabsenceof AI
technologies that is
alreadykilling people.”
26
Eric Horvitz

27. Moshe Vardi
“We have a moral imperative to
study AI in order to save
people’s lives.”
27

28. IMPREGNABLE OFF SWITCH
Proposed Framework for Regulating AI
Source: 2001: A Space Odyssey (1968)
(Etzioni Op-Ed, 9/1/17)

29. AI Field Is Fast
Moving, Amorphous
AI CARS
AI TOYS
ROBOTS
Regulate AI
Applications

30. Responsibility: an AI is
subject to the laws that
apply to its human
operator
My AI “did it” is not an excuse.
THE DOG ATE MY HOMEWORK…
(EtzioniOp-Ed, 9/1/17)

31. Disclosure: an AI shall disclose that it is not human.
Source: TV Series Finale Humans
(EtzioniOp-Ed, 9/1/17)

32. an AI shall not retain or disclose confidential
information without approval from the source.
Privacy:
AI Barbie
Amazon Echo Roomba
(EtzioniOp-Ed, 9/1/17)

33. AVOIDING BIAS
An AI shall not
amplify the
Bias in its
training data

34. BOTTOM LINE
AI IS NOTGOOD OR EVIL;
AI IS A TOOL;
A TECHNOLOGY,
Loss of Jobs
Major concern:

35. We Are Hiring! (visit Allenai.org)
35

36. 36
Our incubator helps entrepreneurs build
large, impactful, AI-fueled companies.
CTO Residency
Program
Early-Stage
Startups
Entrepreneurs
in Residence
We trainand mentor talented
engineers on the latest in
DeepLearning,Computer
Vision,ML, and more.
We helpearly-stage
startups buildouttheir
AI teamsandaccelerate
their AI capabilities.
We help successful
entrepreneurs build their
next billion-dollar startup
with AI at the core.

37. 37
Companies We’ve Incubated To-Date:
Your StartupHere
Ai2incubator.com

38. 1 23
38
CTO Residency Program (January 23!)
Get up to 12-monthsof
highly exclusive training
on Deep Learning, NLP,
Computer Vision, ML, etc.
Pair with a successful
CEO tocollaboratively
start building a new
company together.
Exit the incubator as
CTO of your new startup
with a large amount of
co-founder equity.