Embed Size (px)
Citation preview
AI 3.0: Is it Finally Time for Artificial Intelligence and
Sensor Networks to be Disruptive?
Have recent advances in mathematical algorithms,
highly sensitive/compact sensors, big data, mobile
communications, and robotry made Stephen
Hawking’s warning that artificial intelligence could
end mankind more eminent?
What does this mean for jobs in the “second
machine age” and AI 3.0?
David Smith
AI 3.0: Is it Finally Time for Artificial
Intelligence and Sensor Networks to be
Disruptive?
As we begin the new millennium science and
technology are changing rapidly
“Old” sciences such as physics are relatively well-understood
Computers are ubiquitous
Grand Challenges in Science and Technology
Understanding the brain
reasoning, cognition, creativity
creating intelligent machines
is this possible?
What are the technical and philosophical challenges?
Arguably AI poses the most interesting challenges and questions in computer science today
“Whoever wins this race will dominate the next stage of the information
age,” - Pedro Domingos, a machine learning specialist and author
of “The Master Algorithm,” a 2015 book contending that
A.I. and big-data technology will remake the world.
ARTIFICIAL INTELLIGENCE
“AI is the study of
techniques for
solving exponentially
hard problems in
polynomial time by
exploiting knowledge
about the problem
domain.“
Elaine Rich
"Once you have a truly massive amount of information
integrated as knowledge, then the human-software system
will be superhuman, in the same sense that mankind with
writing is superhuman compared to mankind before
writing.”
- Technology Review, March 2005
"Compared to Nature we suffer a poverty of imagination; it
is thus much easier for us to uncover than to invent.”
Doug Lenat's Cyc project, is to build
the basis of a general artificial
intelligence by representing
knowledge
What is Intelligence?
Intelligence:
- “The capacity to learn and solve problems” (Webster
dictionary)
- In particular,
• the ability to solve novel problems
• the ability to act rationally
• the ability to act like humans
Artificial Intelligence
- Build and understand intelligent entities or agents
- Two main approaches: “engineering” versus “cognitive
modeling”
Artificial intelligence (AI) is an area of computer
science that emphasizes the creation of intelligent
machines that work and react like humans.
● Artificial intelligence is a branch of computer science
that aims to create intelligent machines.
● Some of the activities computers with artificial
intelligence are designed for include speech
recognition, learning, planning and problem solving.
● Robotics is a major field related to AI.
● Robots require intelligence to handle tasks such as
object manipulation and navigation along with sub-
problems of localization, motion planning and
mapping.
Ex Machina Featurette - New Consciousness
Philosophers have been trying for over 2000 years to understand and resolve two Big Questions of the Universe: How does a human mind work, and Can non-humans have minds? These questions are still unanswered.
Intelligence is the ability to understand and learn things.
Intelligence is the ability to think and understand instead of doing things by instinct or automatically.
Intelligent Machines, or
What Machines Can Do
(Essential English Dictionary)
What’s involved in Intelligence?
Ability to interact with the real world - to perceive, understand, and act
- e.g., speech recognition and understanding and synthesis
- e.g., image understanding
- e.g., ability to take actions, have an effect
Reasoning and Planning - modeling the external world, given input
- solving new problems, planning, and making decisions
- ability to deal with unexpected problems, uncertainties
Learning and Adaptation - we are continuously learning and adapting
- our internal models are always being “updated”
• e.g., a baby learning to categorize and recognize animals
Computers versus humans
A computer can do some things better than a human can
- Adding a thousand four-digit numbers
- Drawing complex, 3D images
- Store and retrieve massive amounts of data
However, there are things humans can do much
better.
Thinking Machines
A computer would have
difficulty identifying the
cat, or matching it to
another picture of a cat.
AI Purposes
"AI can have two purposes. One is to use the power of
computers to augment human thinking, just as we use
motors to augment human or horse power. Robotics and
expert systems are major branches of that. The other is
to use a computer's artificial intelligence to understand
how humans think. In a humanoid way. If you test your
programs not merely by what they can accomplish, but
how they accomplish it, they you're really doing cognitive
science; you're using AI to understand the human mind."
- Herb Simon
“We cannot solve our problems
with the same thinking we
used when we created them.”
- Albert Einstein
Overview of Artificial Intelligence
Definitions – four
major combinations
- Based on thinking
or acting
- Based on activity
like humans or
performed in
rational way
Systems that think like humans
Systems that think rationally
Systems that act like humans
Systems that act rationally
“The market for enterprise AI systems will increase from $202.5 million
in 2015 to $11.1 billion by 2024.”
- Tractica
• By 2018, 20 percent of business content will be authored by
machines.
• By 2020, autonomous software agents outside of human
control will participate in five percent of all economic
transactions.
• By 2018, more than 3 million workers globally will be
supervised by a "robo-boss.“
• By 2018, 45 percent of the fastest-growing companies will
have fewer employees than instances of smart machines.
• By year-end 2018, customer digital assistant will recognize
individuals by face and voice across channels and partners.
• By 2020, smart agents will facilitate 40 percent of mobile
interactions, and the post app era will begin to dominate.
First Key to Creating Artificial General Intelligence:
Increasing Computational Power
NNow =
• Beating a
mouse brain
• About a
thousandth of
a human
Second Key to Creating Artificial General Intelligence:
Making It Smart
Strategies:
1) Plagiarize the brain.
• Reverse engineer it
• Build chips to simulate it
• Capture its synapses
• “Whole brain emulation”
2) Try to make evolution do what it did before but for us this time.
• Use foresight – just pick what you know will win
• Select for intelligence
• Provide externally what evolution takes extra steps to do, i.e.,
provide outside energy/electricity
3) Make this whole thing the computer’s problem, not ours
• It would do research on AI and code the changes into itself
Now: 1 mm-long
flatworm brain of
302 Neurons
Although artificial intelligence as an independent field of
study is relatively new, it has some roots in the past. We can
say that it started 2,400 years ago when the Greek
philosopher Aristotle invented the concept of logical
reasoning. The effort to finalize the language of logic
continued with Leibniz and Newton. George Boole
developed Boolean algebra in the nineteenth century
that laid the foundation of computer circuits. However, the
main idea of a thinking machine came from Alan Turing, who
proposed the Turing test. The term “artificial intelligence”
was first coined by John McCarthy in 1956.
History of artificial intelligence
Meet HAL 2001: A Space Odyssey
- classic science fiction movie from 1969
HAL - part of the story centers around an intelligent computer called HAL
- HAL is the “brains” of an intelligent spaceship
- in the movie, HAL can
• speak easily with the crew
• see and understand the emotions of the crew
• navigate the ship automatically
• diagnose on-board problems
• make life-and-death decisions
• display emotions
In 1969 this was science fiction: is it still science fiction?
AI 1.0 (1960-1985):
AI applications addressed a single area. In this period, they were high value such as human language
translation and route optimization centered around the high cost of humans. Algorithms were
mechanistic. Heavy demand for IT resources made implementations expensive. Today, single area AI
applications, enabled by more sophisticated mathematics and high performance computing, is labelled
Artificial Narrow Intelligence (ANI).
AI 2.0 (1986 - 2010):
AI applications appeared to address a broad area. In this period, they were capable of doing the work
of an occupation of people such as picking crops, scanning social networks for consumer input, and
classifying images for quicker retrieval. Algorithms became more sophisticated and IT resources much
less expensive. However, the solutions approach mimic how humans thought and still fell short of the
abilities of experts. This class of AI Application is labelled Artificial General Intelligence (AGI).
AI 3.0 (2011 - Now):
AI applications are appearing that can solve problems better than the best human in an area of
interest. Examples of this class of AI application can win a the most complex strategic board games,
perform retrieval and analysis of knowledge to quickly answer questions, and stock market trading.
This generational shift has been driven by high value potential, accumulation of massive data of all
kinds, even faster computers the ability to analyze a single situation across a cluster of computers,
and the algorithms to exploit the new technological resources to analyze problem deeper to
incorporate behavioral/neuro/ social data to perform real time analysis and even learn. This class of AI
application is being called Artificial Superintelligence (ASI).
Artificial Intelligence Generations
The vast majority of AI research practiced in academia and industry today fits into the “Narrow AI” category
Each “Narrow AI” program is (in the ideal case) highly competent at carrying out certain complex goals in certain environments
• Chess-playing, medical diagnosis, car-driving, etc.
Narrow AI
“The ability to achieve complex goals in complex environments using limited computational resources”
• Autonomy
• Practical understanding of self and others
• Understanding “what the problem is” as opposed to just solving problems posed explicitly by programmers
Artificial General Intelligence (AGI)
Artificial General Intelligence (AGI)
Artificial Intelligence Generation Comparison
Factor \Generation AI 1.0 AI 2.0 AI 3.0
Period of Time 1960 to 1985 1986 to 2010 2011 to Now and
beyond
Type of AI App
Introduced
Artificial Narrow
Intelligence (ANI)
Artificial General
Intelligence (AGI)
Artificial
SuperIintelligence (ASI)
Value Proposition Human Efficiency Human Effectiveness Human Substitution
Human Ability
Acquired
Fast manipulation of
text and data
Incorporation of
knowledge,
Audio/visual recognition
Understanding,
Reasoning
ANI Roadmap Batch processing Complex data/math Real time
AGI Roadmap Longitudinal data,
Pattern recognition
Data warehouses,
Non-SQL data bases
ASI Roadmap Deep Neural Nets,
Big Data, Robotics
Different Types of Artificial
Intelligence
Modeling exactly how humans actually think - cognitive models of human reasoning
Modeling exactly how humans actually act - models of human behavior (what they do, not how they think)
Modeling how ideal agents “should think” - models of “rational” thought (formal logic)
- note: humans are often not rational!
Modeling how ideal agents “should act” - rational actions but not necessarily formal rational reasoning
- i.e., more of a black-box/engineering approach
Modern AI focuses on the last definition - we will also focus on this “engineering” approach
- success is judged by how well the agent performs
-- modern methods are also inspired by cognitive & neuroscience (how people think).
A Human vs. Machine Comparison
Category Attribute Man Machine
Hardware Processing speed Max @ 200 cycles/sec Already 2 billion cycs/sec
Interconnect speed ~ 120 meters/second Speed of light
Size/Storage Size of skull; any
bigger we’d think more
slowly
Greatly expandable in short
term/working/long term
memories; has error
detect/self-correct bits
Reliability/durability Get easily fatigued; will
deteriorate over time
Transistors more accurate
that neurons; can be
repaired or replaced; can
run non-stop 24/7
Software Programmability Human brain is not
“updatable”
Can be optimized to suit its
role; improvable; fixable
“The Collective” Our ability to build vast
collective intelligence
and apply it collectives
has made us the top
species
All computers could work
together on a single
problem; whatever is
learned can be instantly
“assimilated” by all
Overall Self Improvement ??? Yes
• Fast computers internetworked
• Decent virtual worlds for AI embodiment
• Halfway-decent robot bodies
• Lots of AI algorithms and representations
• often useful in specialized areas
• often not very scalable on their own
• A basic understanding of human cognitive
architecture
• A cruder but useful understanding of brain
structure and dynamics
• A theoretical understanding of general intelligence under conditions of massive computational resources
What We Have Now
Artificial Intelligence in the Movies
The Intelligence is in the Connections
Connections between people
Co
nn
ec
tio
ns b
etw
ee
n In
form
ati
on
Social Networking
Groupware
Javascrip
t Weblogs
Databases
File Systems
HTTP
Keyword Search
USENET
Wikis
Websites
Directory Portals
2010 -
2020
Web 1.0
2000 - 2010
1990 - 2000
PC Era 1980 - 1990
RSS Widgets
PC’s
2020 - 2030
Office 2.0
XML
RDF
SPARQL AJAX
FTP IRC
SOA
P
Mashups
File Servers
Social Media Sharing
Lightweight Collaboration
ATOM
Web 3.0
Web 4.0
Semantic Search
Semantic Databases
Distributed Search
Intelligent personal agents
Java
SaaS
Web 2.0 Flash
OWL
HTML
SGML
SQL
Gopher
P2P
The Web
The PC
Windows
MacOS
SWRL
OpenID
BBS
MMO’s
VR
Semantic Web
Intelligent Web
The Internet
Social Web
Web OS
Natural Language: Translation
“The flesh is weak, but the spirit
is strong”
Translate to Russian
Translate back to English
“The food was lousy, but the
vodka was great!”
Your Assignment
Let’s start with an easy one:
Chair
Chair?
Chair?
Chair?
Chair?
Chair?
Chair?
Chair?
Chair?
Chair?
Chair?
Chair?
Chair?
Chair?
Chair?
Chair?
The bottom line?
Bill Gates on AI Issues and Potential
[Bill Gates] weighed in on the issue of artificial
intelligence when a Redditor asked him how he felt
about regulating artificial intelligence. Gates said he
agrees with Elon Musk and physicist Stephen
Hawking that, "when a few people control a platform
with extreme intelligence, it creates dangers in terms
of power and eventually control."
When asked about his early motto of putting a
computer in every home, Gates said that today, the
challenge is to make computers more intelligent.
"Software still doesn't understand what thing I should
pay attention to next," he wrote, "in fact the
proliferation of various tools like texting and email and
notifications mean the user has a lot of complexity to
deal with. Eventually the software will understand
what you should pay attention to by knowing the
context and learning about your preferences."
Source: Puget Sound Business Journal, March 8, 2016
Prof Stephen Hawking, one of Britain's pre-eminent
scientists, has said that efforts to create thinking
machines pose a threat to our very existence.
Prof Hawking says the primitive forms of artificial
intelligence developed so far have already proved very
useful, but he fears the consequences of creating
something that can match or surpass humans.
"It would take off on its own, and re-design itself at an
ever increasing rate," he said.
“Humans, who are limited by slow biological evolution,
couldn't compete, and would be superseded."
Geek, May 13, 2015. Renowned physicist
Stephen Hawking appeared at the
Zeitgeist 2015 conference in London and
confirmed the fears held by anyone who
has watched a movie with a robot in it
since 1927’s Metropolis when he said,
“Computers will overtake humans with AI
at some within the next 100 years. When
that happens, we need to make sure the
computers have goals aligned with ours.”
63
Artificial Intelligence: Current Status
Approaches
- Symbolic, statistical, learning algorithms, physical/mechanistic, hybrid
Current initiatives
- Narrow AI: DARPA, corporate
- Strong AI: startup efforts
Near-term applications
- Auditory applications: speech recognition
- Visual applications: security camera (crowbar/gift)
- Transportation applications: truly smart car
Format
- Robotic (Roomba, mower, vehicles)
- Distributed physical presence
- Non-corporeal
Kismet
Stanley
AI State of the Art - Applications
AI achievements:
- Facilitate and replace human decision making World-class chess and game playing
- Robots
- Automatic process control
- Understand limited spoken language
- Smarter search engines
- Engage in a meaningful conversation
- Observe and understand human emotions
- Solving mathematical problems
- Discover and prove mathematical theories
- …
“I set the date for the Singularity-
representing a profound and disruptive
transformation in human capability- as
2045.
The nonbiological intelligence created
in that year will be one billion times
more powerful than all human
intelligence today."
Ray Kurzweil
The Singularity is Near (2005)
World Robot Population
World Robot Population
Isaac Asimov’s Three Laws of Robotics (1940)
69
First Law: A robot may not injure a human or
through inaction, allow a human to come to harm.
Second Law: A robot must obey the orders given it
by human beings, unless such orders would conflict
with the first law.
Third Law: A robot must protect its own existence,
as long as such protection does not conflict with the
first or second law.
Are the 3 Laws the Answer? Extending the Laws
70
Zeroth law: A robot may not injure humanity or through
inaction allow humanity to come to harm.
(due to Asimov, Olivaw, and Calvin).
David Langford’s extensions, acknowledging military funding
for robotics:
4. A robot will not harm authorized Government personnel but
will terminate intruders with extreme prejudice.
5. A robot will obey the orders of authorized personnel except
where such orders conflict with the Third Law.
6. A robot will guard its own existence with lethal antipersonnel
weaponry, because a robot is bloody expensive.
Will They Be Like Us?
71
Like us, AI systems...
...will talk to us in our languages.
...will help us with our problems.
...will have anthropomorphic interfaces.
Unlike us, AI systems...
...will compute and communicate extremely quickly.
...will have bounds for learning and retention of knowledge
that will soon surpass ours.
...might not be well modeled by the psychological models
that work for people.
Atlas, The Next Generation Robot
A new version of Atlas, designed to operate outdoors and inside buildings. It is specialized for mobile
manipulation. It is electrically powered and hydraulically actuated. It uses sensors in its body and legs
to balance and LIDAR and stereo sensors in its head to avoid obstacles, assess the terrain, help with
navigation and manipulate objects. This version of Atlas is about 5' 9" tall (about a head shorter than
the DRC Atlas) and weighs 180 lbs.
Domestic Robots
Military robots
The Future?
Idea of Artificial Intelligence is being replaced by Artificial life, or anything with a form or body.
The consensus among scientists is that a requirement for life is that it has an embodiment in some physical form, but this will change. Programs may not fit this requirement for life yet.
Arms race for the future of intelligence
Machine Human
Blue Gene/L 360 teraFLOPS (≈.36+ trillion
IPS) and 32 TiB memory1
Unlimited operational/build knowledge
Quick upgrade cycles: performance
capability doubling every 18 months
Linear, Von Neumann architecture
Understands rigid language
Special purpose solving (Deep Blue,
Chinook, ATMs, fraud detection)
Metal chassis, easy to backup
Estimated 2,000 trillion IPS and 1000
TB memory2
Limited operational/build knowledge
Slow upgrade cycles: 10,000 yr
evolutionary adaptations
Massively parallel architecture
Understands flexible, fuzzy language
General purpose problem solving,
works fine in new situations
Nucleotide chassis, no backup possible
1Source: Fastest Supercomputer, June 2007, http://www.top500.org/system/7747 2Source: http://paula.univ.gda.pl/~dokgrk/bre01.html
ADVANTAGES (Factual Changes)
Smarter artificial intelligence promises to replace human jobs, freeing people for other pursuits by automating manufacturing and transportations.
Self-modifying, self-writing, and learning software relieves programmers of the burdensome task of specifying the whole of a program’s functionality—now we can just create the framework and have the program itself fill in the rest (example: real-time strategy game artificial intelligence run by a neural network that acts based on experience instead of an explicit decision tree).
Self-replicating applications can make deployment easier and less resource-intensive.
AI can see relationships in enormous or diverse bodies of data that a human could not
Analysis of the Risks
• Mass unemployment?
historical evidence is negative
• Loss of income?
productivity creates wealth, jobs, & ownership
• Idleness & boredom?
the rich are seldom idle or bored
• Loss of control over destiny?
freedom to pursue interests
• Overpowered by superior intelligence?
might bring world peace and economic justice
SuperIntelligence Has Already Arrived!!!
In the Stock Market: October 2, 2013 automated computer
buy/sell programs, on news of an offer to buy Blackberry for $9
a share, touched off a flurry of orders reducing the company’s
stock to $7.92.
At Chess: May 11, 1997 IBM’s Deep Blue beat Garry Kasparov,
the then world chess champion. Kasparov had beaten Deep Blue
a year earlier.
At Go: March 14, 2016 Google’s DeepMind beat leading Go
player Lee Sedol 4-1. Lee won in the fourth game by forcing his
opponent into an error. However, in the fifth game the AI program
made a similar error but recovered to win the game..
In Conversations: June 8, 2014 A Russian chatterbot named
"Eugene Goostman" became the first to pass the Turing Test
by convincing 1 in 3 judges that it was a 13-year-old non-
native-English-speaking Ukrainian boy.
“machines will eventually overtake us, as virtually everyone in the A.I. field believes
…The only real difference between enthusiasts and skeptics is a time frame.”
- NYU research psychologist Gary Marcus
Paul Allen, Microsoft Co-founder:
“We can see that overall AI-based capabilities haven’t been exponentially increasing
either, at least when measured against the creation of a fully general human
intelligence…individual AI systems…have always remained brittle—their performance
boundaries are rigidly set by their internal assumptions and defining algorithms, they
cannot generalize, and they frequently give nonsensical answers outside of their specific
focus areas.”
… But It Won’t Be Self Aware
Murray Shananhan, Imperial College of London cognitive roboticist:
“Consciousness is certainly a fascinating and important subject—but I don’t believe
consciousness is necessary for human-level artificial intelligence,” he told Gizmodo. “Or,
to be more precise, we use the word consciousness to indicate several psychological and
cognitive attributes, and these come bundled together in humans.”
Peter McIntyre, Future of Humanity Institute at Oxford University:
“By definition, an artificial superintelligence (ASI) is an agent with an intellect that’s much
smarter than the best human brains in practically every relevant field. It will know exactly
what we meant for it to do.”
McIntyre believes an AI will only do what it’s programmed to, but if it becomes smart
enough, it should figure out how this differs from the spirit of the law, or what humans
intended.
McIntyre compares the future plight of humans to that of a mouse. A mouse has a drive
to eat and seek shelter, but this goal often conflicts with humans who want a rodent-free
abode. “Just as we are smart enough to have some understanding of the goals of mice, a
superintelligent system could know what we want, and still be indifferent to that,”.
Richard Loosemore, AI researcher and founder of Surfing Samurai Robots:
Thinks that most AI doomsday scenarios are incoherent and argues that these scenarios
always involve an assumption that the AI is supposed to say “I know that destroying
humanity is the result of a glitch in my design, but I am compelled to do it anyway.”
Loosemore points out that if the AI behaves like this when it thinks about destroying us, it
would have been committing such logical contradictions throughout its life, thus
corrupting its knowledge base and rendering itself too stupid to be harmful.
… And Artificial Super Intelligence Will Make Mistakes
Stuart Armstrong, Future of Humanity Institute at Oxford University:
“Many simple tricks have been proposed that would ‘solve’ the whole AI control problem,”
Examples include programming the ASI in such a way that it wants to please humans, or
that it function merely as a human tool. Alternately, we could integrate a concept, like love
or respect, into its source code. And to prevent it from adopting a hyper-simplistic,
monochromatic view of the world, it could be programmed to appreciate intellectual,
cultural, and social diversity.
But these solutions are either too simple—like trying to fit the entire complexity of human
likes and dislikes into a single glib definition—or they cram all the complexity of human
values into a simple word, phrase, or idea.
“That’s not to say that such simple tricks are useless—many of them suggest good
avenues of investigation, and could contribute to solving the ultimate problem. But we
can’t rely on them without a lot more work developing them and exploring their
implications.”
It Will Be Difficult to Mitigate Those Mistakes
Philosopher Immanuel Kant believed that intelligence strongly correlates with
morality.
David Chalmers, Professor of Philosophy, New York University, and Fellow of the
American Academy of Arts & Sciences:
“If this [Kant’s belief] is right...we can expect an intelligence explosion to lead to a
morality explosion along with it. We can then expect that the resulting [ASI] systems will
be supermoral as well as superintelligent, and so we can presumably expect them to be
benign.”
Stuart Armstrong, Future of Humanity Institute at Oxford University:
“Smart humans who behave immorally tend to cause pain on a much larger scale than
their dumber compatriots,” he said. “Intelligence has just given them the ability to be bad
more intelligently, it hasn’t turned them good.”
“We’d have to be very lucky if our AIs were uniquely gifted to become more moral as they
became smarter,” he said. “Relying on luck is not a great policy for something that could
determine our future.”
Know that ASI Won’t Be Friendly.
However, we won’t be destroyed by ASI.
Eliezer Yudkowsky, Research Fellow, Machine Intelligence Research Institute:
“The AI does not hate you, nor does it love you, but you are made out of atoms which it
can use for something else.”
Peter McIntyre, Future of Humanity Institute at Oxford University:
“An AI might predict, quite correctly, that we don’t want it to maximize the profit of a
particular company at all costs to consumers, the environment, and non-human animals.
It therefore has a strong incentive to ensure that it isn’t interrupted or interfered with,
including being turned off, or having its goals changed, as then those goals would not be
achieved.”
Elon Musk, Founder and CEO of Tesla and SpaceX:
Points out that artificial intelligence could actually be used to control, regulate, and
monitor other AI. Or, it could be imbued with human values, or an overriding imposition to
be friendly to humans.
Super Intelligent Assistants Will Be More Helpful Than Your
Spouse
Know more about your habits
Anticipate your next move
Prepare you for your next event
Provide the right information for events
Communicate your thinking customized for
each recipient
Follow-up on the impact of your decision
Even do the heavy lifting
