Keynote: Act deliberately and preserve things. Academic Libraries in an age of artificial intelligence
Nicole Coleman, Digital Research Architect, Stanford University Libraries and Research Director, Humanities + Design
6. Ray Kurzweil
Director of Engineering at Google
“Within several decades, information based
technologies will encompass all human knowledge
and proficiency, including, ultimately, the pattern
recognition powers, problem solving skills and
emotional and moral intelligence of the human
brain itself. This singularity will allow us to
transcend the biological limitations of our bodies
and our brains. We will gain power over our fates.
Our mortality will be in our own hands. We will be
able to live as long as we want and we will be able
to fully understand human thinking and will be
able to vastly expand and extend its reach.”
7. “Artificial intelligence, the dream of computer
scientists for over half a century, is no longer
science fiction. And in the next few years, it
will transform every industry… Where engines
made us stronger and powered the first
industrial revolution, AI will make us smarter
and power the next. What will make this
intelligent industrial revolution possible? A
new computing model — GPU deep learning
— that enables computers to learn from data
and write software that is too complex for
people to code.”
Nvidia maker of the GPU
8.
9. Robert Harrison
Professor in Italian Literature
Stanford University
“These guys should all take two years out
of their lives to undergo an education in
literature, maybe scripture and the classics
and learn something about how to
improve this really tacky imagination they
have about what the good life is supposed
to consist in.”
10. Douglas Hofstadter
“If you look at situations in the
world, they don’t come framed, like
a chess game or a Go game or
something like that. A situation in
the world is something that has no
boundaries at all, you don’t know
what’s in the situation, what’s out of
the situation.”
12. Despite its name, there is
nothing “artificial” about this
technology — it is made by
humans, intended to behave like
humans and affects humans.
Fei-Fei Li
Associate Professor of Computer Science
Stanford
It is pattern matching
within a very wide space of
possibility.
Edward Feigenbaum
the "father of expert systems"
15. 1997
IBM’s Deep Blue
beats Garry Kasparov
1986
"Learning
representations by
back-propagating
errors”
Rumelhart, Hinton,
and Williams
1998
Larry Page and Sergey
Brin found Google
based on the
“PageRank” algorithm
1989
Yann LeCun applies
deep learning to
reading handwritten zip
codes on envelopes
1999
Nvidia GPU
2009
ImageNet
Andrew Ing begins
using Nvidia GPUs
for deep learning
20082005
Amazon launches
the Mechanical
Turk
1986-2009
16. Google’s AlphaGo
defeats Lee Sedol
20162010
Deep Learning for
Speech recognition
George Dahl
2012
Deep Learning for
computer vision
Geoffrey Hinton
Using ImageNet
2011
Google Brain
project begins
Andrew Ing,
Jeff Dean
2013
Google hires
Geoffrey Hinton
This leads to
Siri, Alexa, etc.
2015
Google RankBrain
applies deep
learning to search
2014
Andrew Ing’s team
applies unsupervised
learning example on
cat videos
Facebook launched
the Newswire app
and acquired
WhatsApp, the
smartphone instant
messaging
application
2010-2016
George Dahl joins
Google Brain
17. Google’s AlphaGo
defeats Lee Sedol
20162010
Deep Learning for
Speech recognition
George Dahl
2012
Deep Learning for
computer vision
Geoffrey Hinton
Using ImageNet
2011
Google Brain
project begins
Andrew Ing,
Jeff Dean
2013
Google hires
Geoffrey Hinton
2015
Google RankBrain
applies deep
learning to search
2014
Andrew Ing’s team
applies unsupervised
learning example on
cat videos
Facebook launched
the Newswire app
and acquired
WhatsApp, the
smartphone instant
messaging
application
2010-2016
George Dahl joins
Google Brain
This leads to
Siri, Alexa, etc.
18. 2. Processing power
I don’t have 1,000 computers lying around, can I even research this?
19. 1997
IBM’s Deep Blue
beats Garry Kasparov
1986
"Learning
representations by
back-propagating
errors”
Rumelhart, Hinton,
and Williams
1998
Larry Page and Sergey
Brin found Google
based on the
“PageRank” algorithm
1989
Yann LeCun applies
deep learning to
reading handwritten zip
codes on envelopes
1999
Nvidia GPU
2009
ImageNet
Andrew Ing begins
using Nvidia GPUs
for deep learning
20082005
Amazon launches
the Mechanical
Turk
1986-2009
20. Google’s AlphaGo
defeats Lee Sedol
20162010
Deep Learning for
Speech recognition
George Dahl
2012
Deep Learning for
computer vision
Geoffrey Hinton
Using ImageNet
2011
Google Brain
project begins
Andrew Ing,
Jeff Dean
2013
Google hires
Geoffrey Hinton
This leads to
Siri, Alexa, etc.
2015
Google RankBrain
applies deep
learning to search
2014
Andrew Ing’s team
applies unsupervised
learning example on
cat videos
Facebook launched
the Newswire app
and acquired
WhatsApp, the
smartphone instant
messaging
application
2010-2016
George Dahl joins
Google Brain
21. Ben Vigoda (MIT)
Today, one million microchips = the compute power of the human brain
at the cost of $1 billion.
(100 Billion Neurons in the human brain. 10 Billion transistors in a modern
microchip. 100,000 transistors to do the work of one neuron.)
In 2026: It will take only 10,000 chips at a cost of $10 million
In 2036: It will take only 100 chips at a cost of $100,000
In 2046: It will be 1 chip at a cost of $1000
One personal computer will have the same compute capacity as one person.
We may be able to shrink that timeline down to 20 years.
22. 3. Lots of Data
It’s not who has the better algorithm, it’s who has more data.
23. 1997
IBM’s Deep Blue
beats Garry Kasparov
1986
"Learning
representations by
back-propagating
errors”
Rumelhart, Hinton,
and Williams
1998
Larry Page and Sergey
Brin found Google
based on the
“PageRank” algorithm
1989
Yann LeCun applies
deep learning to
reading handwritten zip
codes on envelopes
1999
Nvidia GPU
2009
ImageNet
Andrew Ing begins
using Nvidia GPUs
for deep learning
20082005
Amazon launches
the Mechanical
Turk
1986-2009
24. Google’s AlphaGo
defeats Lee Sedol
20162010
Deep Learning for
Speech recognition
George Dahl
2012
Deep Learning for
computer vision
Geoffrey Hinton
Using ImageNet
2011
Google Brain
project begins
Andrew Ing,
Jeff Dean
2013
Google hires
Geoffrey Hinton
This leads to
Siri, Alexa, etc.
2015
Google RankBrain
applies deep
learning to search
2014
Andrew Ing’s team
applies unsupervised
learning example on
cat videos
Facebook launched
the Newswire app
and acquired
WhatsApp, the
smartphone instant
messaging
application
2010-2016
George Dahl joins
Google Brain
25. ImageNet
Does ImageNet own the images? Can I download the images?
No, ImageNet does not own the copyright of the images. ImageNet only provides
thumbnails and URLs of images, in a way similar to what image search engines do. In
other words, ImageNet compiles an accurate list of web images for each synset of
WordNet. For researchers and educators who wish to use the images for non-
commercial research and/or educational purposes, we can provide access through
our site under certain conditions and terms.
38. Supasorn Suwajanakorn, Steven M. Seitz, and Ira Kemelmacher-
Shlizerman. 2017. Synthesizing Obama: Learning Lip Sync from
Audio. ACM Trans. Graph. 36, 4, Article 95 (July 2017), 13 pages.
Thies, Justus, et al. "Face2face: Real-time face capture and
reenactment of rgb videos." Computer Vision and Pattern
Recognition (CVPR), 2016 IEEE Conference on. IEEE, 2016.
39. And it need not be based on actual recorded audio.
March 7, 2018
41. “The danger I see with machine learning is that
it gives rise to the qualified back-formation
‘human learning,’ for which machine learning
does not bode particularly well.”
John Willinsky
Professor Graduate School of Education
Stanford University
42. Herbert Dreyfus, What Computers Still Can't Do, 1992
“Expertise consists in being able to respond to relevant facts.”
48. “We thought the problem was that
the class ‘basketball’ only contained
pictures of black people. But no.
The class contains as many pictures
of white people as black people.
The bias appears because black
people are under-represented in
the data set as a whole.”
Moustapha Cissé
Research Scientist
Facebook AI Research Lab (or FAIR) Paris
51. AI researchers employ not only the scientific method, but also
methodology from mathematics and engineering. However, the use of the
scientific method - specifically hypothesis testing - in AI is typically
conducted in service of engineering objectives. Growing interest in topics
such as fairness and algorithmic bias show that engineering-focused
questions only comprise a subset of the important questions about AI
systems. This results in the AI Knowledge Gap: the number of unique AI
systems grows faster than the number of studies that characterize these
systems' behavior. To close this gap, we argue that the study of AI could
benefit from the greater inclusion of researchers who are well positioned
to formulate and test hypotheses about the behavior of AI systems.
Closing the AI Knowledge gap
Z. Epstein, B. H. Payne, J. H. Shen, A. Dubey, B. Felbo, M. Groh, N. Obradovich, M. Cebrian, I. Rahwan (2018). Closing the AI Knowledge Gap. arXiv:1803.07233 [cs.CY]
52. The role of the library
“We have the capability, now we need to know what questions to ask.”
53. Peter Norvig
Director of Research at Google
“Engineers need to become more like natural
scientists. The task is no longer bug fixing but
observing and guiding.”
“Applications of AI are not engineering
problems, they are subject matter expert
problems.”
Andrew Ng
Professor of Computer Science, Stanford
54. This signals a paradigm shift in technology from
engineering to design.
In other words, it’s up to us.
56. Philip A. Schreur
Associate University Librarian for Technical and Access Services
from “The Academy Unbound: Linked Data as Revolution” 2012
“Linked data has the potential to revolutionize the academic world of
information creation and exchange. Basic tenets of what libraries
collect, how they collect, how they organize, and how they provide
information will be questioned and rethought. Limited pools of
bibliographic records for information resources will be enhanced by
data captured at creation. By harvesting the entire output of the
academy, an immensely rich web of data will be created that will
liberate research and teaching from the limited, disconnected silos of
information that they are dependent on today.”
61. Deeper and Higher Resolution Layers
Improved Layer Tracing
Schroeder et al. (in prep)
Published Film Record
High Resolution Scan
From “Observing Antarctic Ice-sheet Conditions Using Ice-Penetrating Radar” presented at the American Physical Society April Meeting 2018
Dustin Schroeder, Stanford University
64. The keys to human intelligence are not only to be found in fields
like neuroscience, cognitive science, and psychology. They are
evidenced in how we organize knowledge and how scholars
connect their own ideas to the past and to their contemporaries.
The library holds this information and we can ‘operationalize’ it in
the design of better systems of discovery that provide meaningful
context for information.