13
Aug

AI Now 2018 Symposium


test. Test.
Test. Test. Test. Test. Test. Test. Test. Test. This is a
test. Test. Test. Test. This is a test. Test. Test. Test.
This is a test. Test. Test. Test. This is a test.
[music] [music] [music] [music] [music] [music] [music]
[music] [music]
[music] [music]
[music] [music] [music] [music] Ladies and gentlemen, the show
will begin in five minutes. Please take your seats. Ladies and gentlemen, the sew
will begin in three minutes. Please turn off your cell phones
and silence your electronic devices.
Thank you. Hello, everybody. Welcome to the third annual AI
Now Symposiums. Somehow we’ve had three. It’s been an
extremely big year. This is the biggest gathering. We talk
about what’s been happening in AI, we acknowledge good work,
and we map some paths forward. We have a stellar lineup for for
you tonight. They are going to be addressing
ethics, accountability, and organizing. Our third panel is going to be
on facial recognition and surveillance. The second panel
is going to look at the problems of rising
inequality, austerity, and politics in AI.
Finally we’re going to look at research and organizing and how
they can work together to develop stronger accountability
and develop some of the vexing issues that we’ve been facing in
2018. Speaking of that, we’re going to give you a tour of what
happened this year. Now AI systems have been increasing in power and reach against a pretty
stark political backdrop. Meanwhile they have been made to
shift in upheaval in both the AI research field and in the tech
industry at large. To capture this, we’ve decided to do
something a little different. We’re going to visualize all of
the stories that happened this year. I want to warn you this
is overwhelming. It is designed to be. It’s been a pretty endless
parade of events. In any year Cambridge would have
been the biggest story. It is just one of many. Facebook
along had a royal flush of scandals, including — let’s go
through them briefly. Just a sample. A huge data breach in
superintendent, multiple class actions of
discriminations, potential violations of the Fair Housing Act in May, and
hosting massive fake Russian accounts all year round. We saw Mark Zuckerberg himself
and others testifying. Facebook was
by no means the only one. News broke in Martha Google was
building AI systems for the department of defense’s drone project,
known as Project Maven. This kicked off the organizing and
defense. Then in June when the Trump Administration introduced the
Family Separation Policy that forcibly removedded kids from
their children, employees from Amazon and
Microsoft asked their companies to end
contracts with I.C.E. It was revealed that I.C.E. had tampered with its own risk assessment algorithm to produce
one result. 100% of immigrants in custody
would be receiving the recommendation of detained.
Zero would get released. Meanwhile this is a big year for
the spread of facial recognition. We saw Amazon, Facebook, and
Microsoft launch facial recognition as a service. We
also learned that IBM was working with the NYPD and secretly built
an ethnicity detection feature to search for people’s faces
based on race. They did that using police camera footage of
thousands of people in New York, none of whom knew they would be
used for this purpose. All year we saw more and more AI systems being used in high-stake
domains with real consequences. Back in March we had the first
fatalities for drivers and pedestrians from autonomous
vehicles. In May we had a voice recognition in the UK which was
meant to be detecting immigration and accidentally
cancel thousands of people’s VISAs. In July it was reported
that IBM Watson was producing inaccurate and sometimes
dangerous cancer treatment recommendations. All of these events have,
pushing a growing wave of tech criticism which is focused on
the unaccountable nature of the systems. Some companies,
including Microsoft and Amazon, have made public
calls to have regulation of technologies like facial
recognition. That’s a first. Although so far we haven’t
really seen any real movement from Washington. I’m waiting on
that one. So that’s a tiny sample of what has been a hell
of a year. Researchers like us who work on issues around the
social implications of AI have basically been talking about the
scale of the challenge that we now face. There’s so much work
to be done. But there are also some really positive changes
too. We’ve seen the public discussion about AI mature in some significant
ways. So six years ago when I was
grading papers from people like Kate and
Cynthia on the topic of bias and AI,
these were outliers positions. Even years ago, some of you have
been at the first AI Symposium. It
was not mainstream. Now it is. There are ways in which AI can
reflect bias. Like Amazon’s machine learning
system for resumé scanning which was shown to be discriminating against woman
that it was down ranking for containing the word woman. Back in July they showed Amazon
our they incorrectly identified 28 members of Congress as
criminal. They also said recognition
performed less well on darker skin women. We’re thrilled that
the co-author will be joining us on stage along with Nicole Ozer
who drove the ACLU project. Overall, this is a big step
forward. People now recognize bias as a problem. But the
conversation has a long way to go. It is already bifurcating into
different camps. In column A, we see the fixing
and bias and solutionism. In column B we see an attempt for
ethical codes that will do the heavy lifting for us. In just the last six months,
speaking of column A, IBM, Facebook, Mwai
cosoft, and others have released tool
kits that promised to mitigate ther issues
using system call methods to achieve
fairness. It is necessary and important. But they can’t fix this problem
alone. Because at this point they are
telling technical methods as a cure for social problems. They
are sending in more AI to fix AI. We saw this logic in action
when Facebook quizzed in front of the Senate repeatedly pointed to AI as the
cure for its allegory rhythmic codes. This
is a big reaction to this year. What should be built? What
should we not build? Who should make the decisions? Google published the principles
and ethics courses emerged with the goal of teaching engineers
to make ethical decisions. But a study published yesterday
called in to question the effectiveness of the poaches. It showed that software
engineers do not commonly change behavior based on
exposure to ethics codes. Perhaps this shouldn’t surprise
us. The current focus on individual choice, thought
experiments, and technical definitions is too narrow.
Instead more oversight and public input is required. Or as Lucy Suchman, a pioneer
thinking in human. Computer interactions put it,
while ethics code are a good start,
they lack any real public accountability.
We are delighted she’s joining us on the final panel. In short, ethics principles can
help, but we have a long way to go before they can grapple with
the complexity of issues in place. The biggest as yet unanswered
question is how do we create sustainable
forms of accounts accountability. This is a major
focus of our work. We are trying to address the kinds of
challenges. We have looked at AI this
large-scale context beyond a purely technical focus to include a wider range of
voices and methods. So, Kate, how do we do this?
Let me share with you. We have seen five themes. We thought we
would give you a quick tour tonight. First of all, there’s
a lot to be learned by looking at the underlined material
realities of the AI system. Last month we published the
project called the anatomy of AI. This is a result of a year long collaboration between myself and
the expert where we saw how many
resources werer required to build a device that responds when you say Alexa,
turn on the lights. Starting with that, we traced through all
of the environmental extraction processes from mining and
smelting and logistics to contain shipping and what they
need to build AI systems to scale to the international
networks of data centers all the way through to the final resting place of so many of our consumer, AI gadgets, which is
buried in Ghana, Pakistan, or China. When we look at the
material realities, I think we can begin to see the true
resource implications that this kind of large-scale AI really
requires for our everyday convenience. But if doing this research, we
discovered something else. There are black boxes on top of
black boxes on top of more black boxes. It is not just at the algorithmic level. I think this
is why the planetary resources that are needed to build AI to
scale is really hard for the public to see. But it is really
important if we’re going to develop real accountability.
Second, we are continuing to examine the hidden labor behind an AI
system. Now lots of scholars are working op the issue issue
right now, including people like Lilia , and we are delighted that we
have Astra Taylor who joined the term
that seem to be AI that can only function with a huge amount of
input. Often when we tend to think about the people behind AI, we might
imagine a handful of highly paid dudes in
Silicon Valley. This isn’t the whole picture. As Adrian Khan showed, there’s
more people who work in the minds of
content moderation than people who work at Facebook and Google.
AI takes a lot of work. As we’re going to hear tonight,
most of that goes unseen. And third we need new legal
approaches to contend with automated decision making. That means looking at what’s
working and looking at what’s needed. Accountability rarely
works without reliability on the back end.
There have been some breakthroughs this year. The
data protection guidelines came in to effect in May. New York
City announced its automated decision task force, the first
of its kind in the country. California just passed the
strongest privacy law in the U.S. Plus there are hosts of new
cases taking algorithms to court. We just recently held a workshop
where we invited public interest
lawyers who are representing the people who thinks they’ve been
unfairly cut off from Medicaid benefits or lost their
jobs due to bias systems and people whose
prison sentences have been affected by
skews perception. This focused on how do you build more due
process and safety net? Later tonight you are going to
hear from Kevin de Liban whose ground-breaking work is
published. This is designed to get the
public sector more tools to inquire as to whether an algorithmic system is
appropriate to be used. And then shoring the community
oversight. Rashida Richardson will be
talking more about it later tonight.
-this brings us to the topic of inequality. Popular discussion
often focused on hypothetical youth cases and promises of
teacher benefit. AI is not a common resource available
equality to all. Looking at who builds the systems, who makes the decision on how they
are used, and who is left out and can help
us see beyond the marketing. These are some of the questions
that Virginia Eubanks wrote about. We’re happy she will be
joining us tonight. The power and insights that can be gleamed
are further skewing the distribution of resouses. That the systems are so unevenly
distributed that they may be driving greater forms of wealth
and equality. A new report from the U.N. published that said while AI
could be used to address major issues, there’s no guarantee it
will align with the most pressing needs of humanity. The
report needs that AI systems are increasingly used to manipulate
human emotion and spread
misinformation. Tonight we have the U.N. special on extreme poverty and
human rights. Phillip Alston will be joining
us to talk about his ground breaking work on inequality in
the U.S. If last year was a big moment
for recognizing bias and the limitations of technical systems
in social domains, this coming year is a big most
for accountability. A lot is already happening. People are
starting to take action. There are new coalitions growing.
We’re really excited to be working with a wide range of
people, many of whom are in the room right now. Including legal scholars,
journalist, health and education workers, organizers, and civil
society leaders. AI research will always include
the technical. We’re working to expand its
boundaries. We’re emphasizes
interdisciplinarity and the perspective of those on the
ground. That’s why we are delighted to
have speakers like Sherrilyn Ifill
and Vincent Southerland, each of whom have
made important contributions to the debate. Because genuine
accountability will require the new coalitions. Organizers and
civil society leaders working with researchers will assess AI
systems and to protect the communities who are most at risk. -So continue we offer you a
different kind of AI system. We’re including people from
different disciplines to support the sectors and build more
accountability. That’s where Meredith and I have founded AI
Now in the first place. It has really drived the work that we
do. AI isn’t just tech. AI is power, it is politics, and it is
culture. So on that note, I would like to welcome our first
panelist of the evening. She will be on facial
recognition, Nicole Ozer. Don’t forget to submit your questions
for any of our panelist. Just two to the Twitters and
use #AInow2018. That’s true of people in the room and everybody
on the live stream. Hello, we see you. Not really. (lalaughter).
Please send your questions in. We’ll look at them. Now on with
the show. (Applause).>>Good evening, everyone. I’m
Nicole Ozer. I’m the tech coming and civil
liberties director at the civil liberties in California. I’ve
lead the cutting-edge work working in the courts with
companies, with policymakers, and in communities to defend and
promote civil rights in the digital age. Our team has worked to create landmarks privacy laws like
Calexa. We developed ordinances in Santa
Clara and Oakland. We’ve worked to expose and stop
social media surveillance of black
activist on Facebook, Twitter, and Instagram. We have started
a national campaign to bring attention to the very real
threat of face surveillance. Tonight we are talking face
surveillance. Of course I can’t think of a more timely topic.
For some quick background, the ACLU has long opposed face
surveillance. We’ve identified it as a
uniquely dangerous surveillance form and
a particularly grave threat because of what it can do to
secretly track who we are, where we go, what we do,
and who we know. And how it can be built so easily layered on to existing
surveillance technology and how those incentives and societal incentives really
combine to make it evermore dangerous once
it gets a foothold. And also how it feeds off and exacerbates
a history of bias and discrimination in the country. Professor Woody Hartog wrote
imagine a technology that’s potently dangerous. So
inherently toxic that it deserves to be rejected, banned, and
stigmatized. It is facial recognition. And consent rules, procedural
requirements, and boilerplate contract is no patch for the structure and incentives for
exploitation. This past winter our team discovered the future
it now. Face surveillance that was
thought of was now being quietly and actively deployed by Amazon
for local law enforcement. Amazon’s recognition product
promised locating up to 100 faces in a
picture. Realtime surveillance across
millions of faces. And the company was marketing its use to
monitor crowds. People of interest and wanting
to turn officer-worn body cameras into
realtime surveillance. When we discovered the use of the
technology, we were shocked to find that nothing was in place
to stop this from being used as a tool to attack
community members, to target protesters,
or to be used by I.C.E. which we know has been taking
out courthouses, schools, and
walking down the aisle of buses to arrest and
deport community members. Organizations came together last
spring to blow the whistle and to start pushing Amazon to stop providing
face surveillance to the government. Our coalition call
was quickly echoed by institutional shareholders,
150,000 members of the public, some of you may be some of them, hundreds of
academics, and more than 400 Amazon employees themselves. The ACLU also reenforced the
public understanding by doing our own test of Amazon
recognition. We used the default matching score that
Amazon sets itself its own product and that we know that
law enforcement has also used. The result? Amazon recognition falsely
matched 28 members of Congress. And disproportion that thely
falsely matched members of color, including John Liu wees, Civil Rights Leader.
They have been asking Amazon for answers. Answers they largely
have not gotten. The reality is we only now the tip of the
iceberg. How the government particularly in the current and social climate is
gearing up to try to use face surveillance
to try to target communities. We don’t know what the
communities large and small are doing or not
doing to protect community members. That brings us to
tonight’s timely discussion. I want to thank AI Now. Their research is helping to
further inform some of the very important work. I want to thank
them for the immense privilege of being here tonight
to discuss the critical issue with
Timnit Gebru. She’s a research scientist. She studies the ethical
consideration undermining data mining and what methods are
available to audit. She’s also a co-founder of black
in AI where she’s working to increase
important diversity and reduce the negative impacts of bias in
training data. Timnit’s BHD is from Stanford. He studied computer vision if
the AI lab. We have Sherrilyn Ifill, the President and director council
on the NAACP legal defense fund. She’s the seventh in history to
lead the legal rights organization. For many, Sherrilyn needs no
introduction. So many of us know and admire her as a legal
thinkinger, author, and true power house in bringing to light
challenging issues of race in the American law and society.
So we’re in for a wonderful conversation tonight to help us
all big deeply into understanding the broader social
context. It is not about solving a technical matter, but
decisions about the future of the technology and how it is
used and not used matters so profoundly to the future of who
we are as communities and as a country. And to focus on what
people that we have, can continue to build, and we’ll
need to wield to push back aggressively on threats to the safety and
rights of communities. So with that, let’s get started. I have
the first question for Sherrilyn. Face surveillance is
a relatively new technology. But it isn’t being developed in
a vacuum. How to you think the threat of
face surveillance fit within the
country’s past of violence and
discrimination. Thank you. Thank you for
inviting me and for recognizing the importance of us getting our
hands around the critical issue. I thank you for teeing up the
first question in this way. I think much of our conversation
about technology in the country happens as though technology and AI in
particular is developing in some universe that’s separate than the universe you
and I all know we live in, which is identified with problem of
inequality and discrimination. So here we are in 2018. It is four years after we all
watched Eric Gardner choked to death. It is four years after Michael
Brown was killed in Ferguson. It is three years after Freddie
Gray was killed in Baltimore it is three years after Walter
Scott was sot in the back of the park in north Charleston. It comes as a time of mass
incarceration, a phrase that everyone foes. When the United States
incarcerates the most people in the world, the overwhelming
percentage of them African-American and Latino. It comes at a time in which we e
are segregated at levels that rival
the 1950’s in the schools, where we live, and it comes a time of
widening and some of the widest income inequality that we’ve
seen in the country since the 1920s. And into that reality we develop
this awesome power that allows us to
speed up all of the things that we
currently do, take shortcuts, and
theoretically produce deficiencies in doing what we
do. If we think about facial recognition technology just in
the context that we most often talk about it as a threat in the
context of law enforcement and surveillance. Who here thinks
the biggest problem of law enforcement is, you know, they
need facial recognition technology? Why, as a matter of first
principle, do we think this is something that’s needed for law
enforcement? Why is this something we would devote our
attention to? Why would we devote our public
dollars to the purchase of the technology when we recognize all
of the problems within law enforcement. What we do is we deposit these
technologies into industries, aspects, and governmental institutions
that have demonstrated they are unable to address deep problems
of discrimination and inequality. That lead to
literally destroy lives. Not just killing of people which
is bad enough, but actual destroyed
lives. We drop it into a period of
racial profiling. We drop it into stop-and-frisk. We are part of the team that
sued the NYPD for stop-and-risk here in New York. We are diligently monitoring
that decree. Now we have a technology that
reports to assist police in doing this
kind of law enforcement activity. When we combine it
with things like the gang database in New York which we’ve
been — I’m trying to get information about. New York
City has a gang database. There were about 34,000 people in the
gang database. They have reviewed it and dropped some
folks out. I think it is down to 17 and 20,000. They made
mistakes since they were able to drop out 10,000 people. The
gang database is somewhere between 95 and 99% African-American,
Latino, and Asian-American. It is 1% wait. We’ve asked the NYPD to tell us
the algorithm or the technique they
put it in the gang database and how to get out of it. If I
discovered I was in, how could I get out of it? They still haven’t provided us
with that information. We just filed suit last week. Now try to imagine marrying
facial recognition technology to the development of a database that
thereatically presumes you are in a gang and that your name
pops up in the database. We know we’re in the age of the
Internet, even when you scrub me out, it exists. We’re
unleashing the technology that has the ability to completely
transform forever the lives of individuals. We e do work around employees
and misusing criminal background checks. Some of the work
demonstrates that your arrest record stays with
you forever. You have employees that won’t employee anyone who
has an arrest. We are talking about a class of
unemployable people, a class of people who are branded with a
criminal tag. We have a number of school districted that have invested in
facial recognition. Identifying students that are suspended and
carrying one of the ten most popular guns in school shootings
when they come in the door. Very often these aren’t students
that are suspended. Now we’re going to brand students within
the school. So the context in which the technology is coming
to us. To me it is a very chilling
context. Yet we talk about facial recognition technology
and all of the other efficiency algorithms and AI
technologies as though they exist, were created, or can be evaluated
separately from the very serious context I just described. -And in terms of the historical
context, I often think of the 1958
Supreme Court case NAACP versus Alabama. They were able to maintain the
privacy of the members list. In the case the Supreme Court
recognized the vital relationship between privacy and
the ability to exercise first amendment rights to be able to
speak up and protest. So, you know, in the current political
context that we’re in, how afraid are you, how worried should we be
about the impact of surveillance on civil rights and activist
movements? We should be worried. When we hear the way the
President or attorney general talks about the group and the
creation of the black identity extremist, the idea that the
technology can be mounted and used on police cameras and that the police can
be taking this kind of data from crowds of
people. The crowds of people who came out to to test against
the confirmation of Brett Kavanagh were characterized by some of the President and some
of the republican leadership. Imagine those kind of protesters
and activist would be subjected to facial recognition in which
they would be included in some kind of database that would
identify them with these kinds of words. Think about what this
means for young people. We’re in a movement period in the country in which young people
are engaged and protesting. Now we have to monitor them. The recognition of NAACP versus Alabama and the reason the NAACP
did not have to give up the membership list is the court
recognized the possibility that the revelation of who these
individuals were would subject them to retaliation within the local
community and their freedom to fully
exercise their facial amendment rights. They have the same that
others want to be out in the public phase. We have to talk
about that. So much of it is about implications and the contested public space
in the country and the way in which we now want to privatize
everyone, because you have to believe if you step into the
public space, you would all thely
surveilled. And the last thing is just to go
back to the point about us being so
segregated. It is not just recognizing the
face, it is also evaluating. We know that police officers tend
to assign five additional years to African-American boys. They
see them being older than they are. Who says that people who have
grown up so segregated are in a positioning to evaluate or tell people apart or
are in a position to know whether someone is a threat or
is dangerous. Once we go down the road without
recognizing the way which in America many of the people who
would be using the technology are ill equipped
to evaluate the face of someone, to recognize, and differentiate
between two black people and two Latino people. Anyone that is asking why aren’t
you happy? Why don’t you smile more? Somebody can look at you
and tell your emotion is not true. We should recognize it is not
just going to click and say that’s
Sherrilyn Ifill. It is going to do more than
that. It is going to try to evaluate my intentions in the moment.
Maybe we’re out protesting. Many of us understand it is
different. While you can leave your cell phone at home and not
be tracked, we can’t leave our face at home. On a more complex level, what do
you think it is about face recognition that potentially
makes it different, more dangerous, or risky than other
AI technologies?>>I think you touched on many
of the things I wanted to say. The first thing is the fact that
for example, it doesn’t just
recognize it evaluates you. And let’s think about your emotions. You know, your emotions are — Ranna, who started an emotion
recognition company, said your emotions are some of the most private
things you possess. Even if it worked perfectly,
that would be terrible. If you could just walk around and
people could see your emotions. It also doesn’t work perfectly. People trust algorithms to be
perfect. I might be perfectly happy. Somebody can say I’m
dissatisfied. I just read recently every day I learn something new about where this
different face — automated facial analysis tools are being
used. I read — I forgot the name of
the company. They were used automated facial analysis
instead of the time cards. So then that — then they were
talking about the potential to then do
emotion recognition in aggregate. If their employees
are not satisfied, they can tell over time. This is pretty scary, right? So a combination of things. The
fact that there are some things that are very private to us. We
want to keep private. There are — even my research
with Joy shows that automated facial analysis tools have high error of
disparity for different groups of people. At the same time,
people trust them to be perfect. I think the three combinations
are dangerous.>>So speaking of that research,
there’s been a lot of talk about accuracy or inaccuracy and sort of improving
its function overall. How does some of this conversation really
miss some of the bigger picture around face surveillance? -I think the fact that we showed
high error disparities could start the conversation. The same way the ACLU could show
therer were high error rates for some of the members of Congress; right?
But that doesn’t mean that you can — you should also have
perfect facial recognition that is being used against mostly black and brown people,
like you said. These two conversations have to happen in
tandem. For example, for emotion recognition, again I’ll bring up Ranna —
she’s the only person I’ve talked to about this. She started it to help autistic
kids. People with autism. I’ve talked to people who want to
work with older people who have dementia and use some of the
emotion recognition kind of technology. Now this could be
something good; right? In this case, you don’t want
high error rates and disparities. The conversation
about accuracy should happen. Similarly there are some other computer-vision technologies
being used for melanoma detection. Again you don’t want — you
know, a skin tone that’s very dark to be
— for the AI technology not to work on
it and get misdiagnosed. This conversation
should happen. The solution is not just to have perfect accuracy in facial
recognition. -You said they have drawn
particular attention to government use. I think for a
reason. I wanted to explore with you all sort of the
distinction. Can it be drawn effectively between government
use and corporate use? Or do some of thingers really
bleed and blend together in terms of civil rights and civil
liberties. This has been on my mind this week. Some of you may
have seen some of the press about the Facebook revealed this
week that it thinks that some of the photos have been scraped by
Russian face surveillance firms. This was an issue that we were
particularly concerned about at the ACLU when the Cambridge Analytica
story broke. Around the time they provided
database for the third-party apps, they started to change
their privacy settings. They got rid of them for photos. We’ve been attendantive to the
fact that public photos could become
a really great space for potentially scraping face
surveillance data. What do you all think can and
should we be addressing the issues? Should we bleed
together and blend together and look at the bigger
issues? -I think on the privacy front,
they blend together. This is part of the difficulty of this work is that the pathway in is
usually one or the other; right? So the pathway in is usually
this is a business. I’m running Amazon or running
Facebook. It is wonderful. You know, the owners and
shareholders are making lots of money. People are using the
technology for whatever reason they want to use it. They are
making a personal decision to use the technology. So that’s
supposed to cover a multitude of it. It is not. We know it is often
very close. It gets very close particularly in times of national security high
alert like post 9/11. You have the telephone computers handing
over information for surveillance. We know there’s the symbiotic
relationship. The government relies on corporations to
develop technology for the government to use for a variety
of reasons. I don’t think there’s some place where it is
benign and some place where it is evil. I think the technology
itself is like a monster that once unleashed is
very hard to put back in the box. The problem, I think, is
that where government stops acting like government and starts acting like it is
just another corporation or another client of a corporation.
The government’s responsibility is to protect the public.
That’s why the conversation about regulation is so
important. Because that’s the government’s role. So it can’t
act just like another consumer of a product or client
of a corporation. It is supposed to hold the public
trust. I think what we’re seeing is the government fall
down on the job being so scared and tentative and buying the
story that we have to leave the folks alone because they are the
brilliant ones who are doing all of the wonderful, technology
stuff. If we regulate them, they are going to smash the
creativity. We thought it was crazy that the
government would require us to put a seat belt across our
waist. We didn’t want to do it. My feet were on the floor. We
sat kids on the floor. -I wrote a paper.
-It is true. You could fit as many people in the family as you
could fit in the car. My father was outraged. He thought it was
discrimination with people with big families. Now you can’t
imagine. Wereuate We create the Boogeymen. The government has to reengage
and not just be another client. -I’m so happy what you brought
this up. We wrote a paper with Kate Crawford. We had case
studies. The automobile industry was one
of the case studies. It took many years for them to legislate
that you have to have seat belts. Even when they were doing crash
tests, they did them with dummies that
were male bodies and you ended up
having car accidents that
disproportionally killed women and children.
-Right. It is not the car. There are certain uses that are
dangerous for society. There has to be some interventions
there. I always think of civil rights and civil liberties as
either protected by friction or by law. I think decades ago it wasn’t
possible to monitor using face surveillance. Now the technology has advanced
and much of the friction in terms of what the police can do
and what the government can do has been
eviscerated. They see what types of protection are going to
be built up to protect the public and communities. We at the ACLU have called on
Congress to pass a federal moratorium.
We have to think through the implications. We have been in the large
coalition pushing on Amazon and other
countries to stop providing it to the government because of
some of the threats and dangers that we’ve talked about. As I mentioned, over 450 Amazon employees have themselves,
spoken out in writing about this. Today in Amazon employee called
Jeff Basoz out. Yesterday when he was on a panel
he acknowledged that there’s — tech could be used by crowds,
but suggested the company has no responsibility. Instead we should leave it to society’s eventual immune
response to address the threats in terms of the real threats to
community members. In contrast, Google has new AI
principles that specifically say they will not pursue technologies
that gather for surveillance, violating international norms or
human rights. With the last couple of minutes for both of you to answer one or
both, what do youty the responsibility of companies is?
What should they be doing? What should law headachers be
doing? headache — makers be doing?
Corporations consistent of people. People have values.
People can agitate and change of course of things. I think
people in corporations need to remember the fact they have the
values it is their responsibility to advocate. I’ll just keep it there.
I think the government has the long view. They hold, in many
ways, the responsibility of communicating history to
corporations and other companies that are developing technologies and
setting up the Internet places that have been
created. We know what has happened in the physical public
space in the country. We know that most of the civil rights
movement was a fight over dignity in the public space.
That’s a lesson, right, to communicate as you think through
how you are going to engage that new technology. The same is
true for facial recognition technology if we think about
racial profiling it is the government’s obligation to recognize those
kinds of pitfalls exist and compel corporations to adhere to
some kind of regulatory scheme that guards against what we know are thing access of every
system. There are certain themes that are recurring in
American life. Racial discrimination is one of them.
The idea that we’re going to create a new tech following and
don’t have to worry about it is absurd. That’s a good segue into the
audience question. Is it already too late? Aren’t we
already on camera everywhere we go? My thought is companies are
watching people. I don’t know everything. I’m
learning new things. Every day I learn a new thing. Joy had
talked to me about a company that’s interviewing people on
camera and doing emotion recognition on them, and then giving verbal and
non-verbal cues to the employees who are their
customers. I didn’t know this existed until
she wrote an op ed about it. Every day I’m learning something
new. We’re unleashing the technology everywhere without
guardrails and without regulation and without some sort
of standard. I don’t think it is too late.
People wear seat belts now; right? I think — you know, it
it has become standard. I don’t think it is too late. We have
to move fast. Tens of thousands of people were
killed in cars. It wasn’t too late when we got seat belts.
You just do what you have to do at the time that you can. I do
think we have to jump ahead of it. What’s dangerous about all
of this is how deeply embedded it can become. That’s part of
why we don’t know. It is so easy to flip in and
embed in a variety of context and the employment context and
law enforcement context. Then it is hard to get it out.
That’s why I think there’s a sense of urgency that we have to move
very, very quickly.
-I agree. The ACLU has been working to
blow the whistle and bring a lot of national attention. The fact
that there’s been such a great response and people have been moving with he’s to address the
issues. — haste to address the issue. I think Kate and Meredith
talking about what happened in the last year. I work at ACLU.
I never think it is too late. There’s a long — there’s a long
arc of history. I think that we as people can really work
together to influence that history and to make sure that civil
rights and civil liberties are protected. You know,
historically there’s always been an advance in technology. It takes sometime for those
protections to get in place. We should not just leave it to an immune sons — response. We have to push that response
and make sure it happens. Both in companies and by
lawmakers. It takes me think about Tasers. Tasers companies are doing a lot
of facial technology in law enforcement. This was supposed
to stop the use of lethal force. This was greeted at something
that was going to be great. Now police officers didn’t have to
kill you. It didn’t get to the discrimination or use of obsessive force or
brutality. You might not die. That was the theory. Even that
is wrong. Supreme Court just denied a case in which our
client was tased to death. Let’s leave — as terrible as
that is, let’s leave that to the side. It is about the
dignitying issues and all of the issues that surround
law enforcement. Just switching the technology to something
doesn’t get at the problem. I think in that sense, it is not
too late. We keep referring to this again and again and again. We’re nibbling down on the corps
issue issue. We have deep problems. Ultimately, it doesn’t get us
there. Seeing the level of discourse this year versus last year, it is a pretty big
difference. There’s a workshop on the computer vision technical conference. Now people are starting to know. I think there’s a glimmer of
hope. Action and involvement by
everyone in the audience. Our ability to change the
narrative and trajectory has gone through how many people speak up. I think we have done some work
to reinforce that. I want to thank the panelist for joining
us. I want to thank you all for coming. Now it is time for the
first spotlight of the evening. Thank you very much. Now we have our first spotlight.
This is where we invite people whose work we admire and we punish
them by asking three them threes in seven minutes. It is a high-speed, high-stakes
game. I couldn’t be more excited to be
sitting here with Astra Taylor. She’s an author and has a few
film “What is democracy” that’s opening in January. Just around
the corner. And she also coined — I think
this is a really useful term called photomat ion. What is
it? We have to be clear it is
fauxmation. I’ve been writing about issues
and thinking about labor and debt. I wanted to come up with a term
that would name this process. What passes for automation isn’t
really automation. I think — I’ll give a deaf nation. Fauxtomation is the illusion to
maintain machines are smarter than they are. You gave some
great examples in the introduction. We can think
about all of the digital janitors cleaning the Internet
and making it a space we want to be in. They are egregiously under paid. We think of Amazon with the
slogan. Artificial artificial artificial intelligence. The
same issues are at play. They are exposing the fact that the digital assistant is a terribly
under paid human being doing a
mind-numbing task. I was standing in line ordering
lunch. I talked to the human being. The man was clenching
his phone. How did the app know my order
was done 20 minutes early? The girl looked at him and said I
sent you a message. It was the man’s — he was so
willing to believe it was a robot. He was so willing to believe
this all-seeing artificial intelligence system is overseeing his
organic rice bowl. He couldn’t see the human labor in front of
our eyes. We do that all the time. We’re not curious about
the process. We’re so ready to devalue and
under estimate the human contribution. I think that’s
really dangerous. -Where is this comes from? Who has the most to gain from
the perfect automated system? -Automation is a reality. It is
happens. It is also an ideology. The point is to
separate those two things. To be very sort of up front when
the idiological component comes to
play. Somewhere right now an employee
is saying someone or something is willing to do your job for
free; right? The idea of sort of inevitable human will be obsolete. He helped take out an ad in the
“Wall Street Journal.” If you people asked for $15,
robots are going to replace you. He wrote another piece. It has
happened. He cried some crocodile tears.
When you watch the video of how the troublesome workers has been
done away is not anything that’s automation. It was just
customers doing their work and putting the orders into iPads.
That’s not autoHaitian. automation.
Newark Airport. Any time you want to buy something. – Capitalist are making
investments in robots to weaken workers and replace them.
-It is less catchy. -If somebody can make that
catchy, we can co-brand the revolution. -Exactly. One of the things I
love is you walk us through the history of
automation, but really the feministic history. -Yes. We’re lead astray. In
the robot future everything is done for us. Instead of looking
at that, the people who can give us insight are socialist
feminist. Because women have a long history of being told domestic
technologies, there’s a book about how the labor-saving
devices ramped up the cult of domestic cleanliness. The tools
created more and more work. They offer a deeper insight than
that. And the socialist, feminist. They are wrestling with the
question of what is work? They grow and contain themselves and
not paying for as much of it is possible. They don’t want to
pay. Capitalize don’t want to pay the
full value of work. One is the assembly line. You are involved
in monetary exchanges. The underlying is all of the work that’s tone to reproduce daily
life and make the workers who can work the jobs for wages. Women have always been told
their work doesn’t matter and it doesn’t
deserve a wage. Because it’s been compensated. There’s something there for us.
There’s going to be a future where we’re — you know, there’s
no work for humans to do. The insight was made probable to me. I was in a lecture who is the
amazing scholar who also features very prominently in my
film who is democracy. And the grad student — we were
talking about reproductive labor and the value of it. Aren’t we
heading to the future where there would be no jobs? You
know, the reserve army of labor. The image that we would be
sitting there with nothing to do. You know, on the margins. Sylvia’s response is bracing. Don’t let them convince you you
are disposable. Right? Don’t let — don’t believe it. Don’t
believe that message. And, you know, I think there’s a really
valuable point there. If the automated day of
judgment, they wouldn’t have to invent all of the apps to fake it. -Not only did you go from
McDonalds to Sylvia, but you did it in seven minutes. Can we applaud Astra? That was
amazing. You did it. Next up we have our panel on inequality , who is chaired by Vincent
Southerland of NYU. Please welcome him and our
panelists. (applause).
Good evening. I’m Vincent Southerland. I’m the director of the 6:00
center of Race and inequality. I also serve as the criminal
justice lead for AI Now. Because both of those words I’m
thrilled to be part of the conversation with our panelist. Both of whom are at the
forefront of work being done AI in a time of rising austerity
and turmoil. Help me in welcome Phillip
Alston and — Phillip let me start with
you. You report on extreme poverty in the United States.
It was the first in such a venue to really include AI in the
conversation about inequality. Why was it important for you to
do that in this report? -My focus is on the human rights
area on the issue issues. In the AI area where we’re
accustom to talk about inequality and the range of
other issues. I think the human rights dimension comes in very
often. I tend to see something of macro
and micro terms. If you are looking at
inequality, then it is a macro focus. What sort of major government
policies or other policies can we adjust
in order to improve the overall situation? But if you do a human rights
focus, then you are really going down
to the grassroots. You are looking at the rights of the
individual who is being discriminated against, usually
for a whole range of different reason or who has simply been
neglected. I think one of the problems is there’s neglect on both sides that the
AI people are not focused on human rights. There’s a great
tendency to talk about human ethics which is undefined
and unaccountable. And on the human rights side, there’s a
tendency to say the stuff is all outside of expertise and not to
really want to engage with it. So in my report on the United
States, down at the end of last year, I made a big effort to try to link thing
issues of inequality, human rights, and
the uses of AI AI. Great. I have a question. -I want to respond to Phillip.
This report was so important. It was so important for movement
organizing in the poor people’s movement to have this kind of
vision of the United States. The 43 million foreign working
folks who are really struggling to meet their needs day to day and are finding
that these new tools often create more barriers for them than lowering
those barriers. One of the things that I think is so
important about what Phillip just said is we often —
particularly in my work in public services, we often —
these tools get integrated under the wire in a way. We see them
as just being administrative changes and not as the
consequential political decisions. We absolutely have
to reject that narrative that these things are just creating efficiencies and
optimizing systems. We are making a profound
political decision to say we have to
triage. These tools will help us make a decision. That’s
already a political choice that buys into the idea that there’s
not enough for everyone. We live in a world of abundance.
There’s plenty for everyone. I think that’s important to point
out. I wanted to respond. Thank you. Applause is good. What does
that look like on the ground? What are the types of thins that
you’ve seen over the last year? What do you see going forward
that squares you when you think about automate and technology? -Phillip and I had a
conversation about this earlier today. We talked about how
important it is to listen to the people who are facing the
consequences directly. We tend to talk about the systems as if
the harm might come in the future in the abstract way. But
the reality is the systems have been integrated into public
assistance since the early 1970s. They are having effects
on people right now and really profound material ways. So for
folks who aren’t familiar with a box automating and equality,
what I do is for the last eight year, but
more intensely the last three years looked at the way that new automated
decision system are being integrated across public service
programs in the United States. I look at lee if the book. One is an attempt to automate and privatize all of the
processes for Indiana. And another is the housing system in
Los Angeles. And the third is the system call model that’s
supposed to be able to predict which children might be victims
of abuse or negligent sometime in the future in Allegheny county, which is where
Pittsburgh is in Pennsylvania. What I saw is despite the fact
that there are some incredible
potentials to integrate services, lower barriers, and provide easier access to social
services in the country, because the tools are built on what I
think of the deep social programming of the United States
which is deeply — deep economic
division, a deep and long history of
murderous, racial discrimination that rather than creating tools to ease the
burden, we’re creating tools that divert people from the
resources that they need. They are legally entitled to and
they need to survive and protect their families. That the tools
often criminalize people as part of their process
of deciding who is deserving enough to get access to the basic human
rights. Then in the end all of the data is used to create crystal balls in
order to know who is going to be risky in the future in order to
deny them resources foe. In the last year one of the
thins that stands out post to me as really important to pay
attention to is if you look in the 2019 federal budget,
the Trump Administration budget, it says they are going to save $188
billion by increasing data collection and analysis in
middle-class programs. I look at welfare and child
services that are poor and working people. Now they are
talking about disability and social security and
unemployment. So it very much seems like poor
and working people have been in the
canaries in the coal mine and the experimental population for
the tools. They are looking to implement
them on everyone. Yeah. What about you, Phillip?
What types of things have you seen that give you cause for concern? I echo what Virginia said. I
think it is important to recognize to the extent that we
take a social welfare system, a system for social protection,
and think we can simply put on top of that AI-type technics
which will paque it more efficient and effective, we’re
doubles down on injustices in a great many way. Because the
system is based on racial discrimination and based on
gender discrimination, is based on discrimination against
nationalities and so on. A whole range of different
problems which are not being addressed. And the people who
are promoting most the efficiency motif of those
who want to slash the programs. So I think there’s a firmtive
responsibility on AI people not to say, hey, we’re just a tech
people. What can we do for you? Where do we put our product? You’ve got to start thinking
proactively. How can we build those in and point out to those
who are hiring us what the existing problems are. And the
second point, of course, is very straightforward as Virginia said what we’re going to see is it is
going to come to EU. The house in EU
can affect your apartment. After the midterm elections what
the administration has been doing in the last two years is
to build up the massive deficit. We know there’s only one way
available to solve that deficit. It is going to be cutting middle
class entitlements. And all of the — all of the
pioneering experiments that are are being done on the poor right
now are soon going to be ratcheted up. That’s the only place the saving
can come from. You guys are going to pay for
the tax cuts. We’re going to see it in all of the services
available to us on a regular basis. I see a year that has
been more or less a lead up tax cut on tax
cut. Tax cuts can only go in one direction. You have
absolutely miserable public services, social services, you
have a much greater burden placed on
women in particular, because they are the ones who always
have to pick up the slack when the State pulls back. And
there’s just going to be the huge push to say it is unavoidable.
We’re sorry. It’s got to happen. -Right. You mentioned earlier taking on
human rights framework. What would that look like in AI? Talking to an person audience,
one has to be sensitive about what we —
we in the rest of the world — call
social rights . -I’ve been accused of these. -If we take a stand, you have a
whole range of non-discrimination issues. You
have equal due process. All of the rights are heavily imply
ated. It is present to go back and
make the point that I made earlier — I don’t really
mean it. It is true that ethics are completely open ended. You
can create your own ethic. But human rights you can’t. They are in the constitution and
in the Bill of Rights. There are certain limits and
have been interpreted by courts. Until we bring those into the AI discussion, there’s no hard
anchor. One of the things that I think
is important that I heard over and over again from the point of
view of administers and designers of the tools is they would say to me the
systems are necessary systems of triage. That we just don’t have
enough resources. We have to make these really difficult
decisions. These tools help us make them more fairly. Again this is one of those
political decisions I was talking about at the top of our conversation which is
triage is actually really bad language to use to describe what
we’re doing. Because trionly assumes that there are more
resources coming. If there aren’t more resources
coming, we’re not triaging. We’re not diagnosing, we’re
rashing. We’re automating sources of
rashing resources and people’s access to the shared wealth. I think it is incredibly
important to think beyond these values of efficiency and
optimization to some of the principle that is are enshined in the
economic human rights. The universal declaration of
human rights. As an American, I can say I
don’t care if we signed on the dotted line. As a political community we’re
allowed to say there’s a line nobody
goes below for any reason. Nobody starves to death.
Nobody sleeps in a street on the tent no family is broken up
because they can’t afford a child’s
medication. That’s pretty baseline. We need to get there and get at
the tools right around what we are doing to each other. I’m wondering how the trend fits
into a larger historical trend. Can you speak to that at all?
This is the first time it’s ever happened where technology has squeezed people.
I use the metaphor. The tools that I’m seeing in
public services are more evolution than revolution. And
that the sort of desocial programming of the tools goes
way, way back in the history. The reason that moment is really important is there’s the human
economic depression in the 1819 depression. Economic elites got really freak
out. Poor people were demanding things like food and houses.
They did what economic elites to. They commissioned a bunch
of studies. The question that the study was supposed to answer, what’s the
problem: poverty, lack of resources, or
pauperism? The ladder was the problem. The dependence on public
benefits, not poverty itself. They built them that the
trade-of is if you want to request public services, you
have to move into the poor house to receive them. You have to
give up your right to vote, your right to hold office, your right to marry, and often your
children. The death rates was something
like 30% annually. A third of people who went in every year
died. This was no joke. This is a really horrifying
institution. And I think really this is the moment in our history that we decided
social service programs first and most important job is to do the kind
of moral diagnosis. To decide whether or not you are deserving enough of help whether
or not we’re going in the direction of
forward. Which is happening many places around the world,
but not here. That feels to me the programming that underlying
the new tools. If we don’t address it by
instituting sort of equity gears we are
going to amplify and speed up.>>I’m wondering where does race
fit into all of this. I know we touched on it a little
bit. I’m curious. How does race exacerbate the
problems? Do you want to start? All right. Thanks. I think there’s a close
relationship between race and attitude to
welfare. There are a lot of studies done. As soon as you
talk about poverty, you have a vision of a black family. They
are the ones that are poor and trying to live off of we whites.
We’re not going to let this happen. I think the narratives are
carefully interwoven. They will try to stigmaize welfare and what I
prefer to call social protection if motivated by the racial
stereotypes which are fairly constant. One of the issues that Virginia
and I talked about this is briefly earlier is that when I started doing the
research, I started to look for indicators of class in the U.S. And by that I — I boiled it
down very simply to looking at whether
income statistics have been matched with different groups of
so on. Of course there’s very little of that. Suddenly race
comes in there. That’s the sole factor. To the
extend that other people are poor, that’s not the main focus.
That’s good in some ways. Because the heritage of racial
inequality is so deep and so powerful. It is also bad. Because it is again
stereotyping. This is something for the black community and not
for the whole community. That’s part of the whole them and us
mentality which, of course, is central to the whole welfare
area. Those people, mainly black, are
not contributing. They are not taxpayers. I, of course being
white, am a taxpayer. I’m not going to put up with supporting
these people. We’ve got to move beyond all of
the narratives. Yeah. It is interesting because
in both of your work make the distinction between the deserving and undeserving
beneficiaries. I think that tracks along the lines. I’m
curious your response.>>One of the most important
places that it comes in, it is so important to keep our eye on
this. Phillip was talking earlier about how much — how often they
rationalize. There’s not enough. We need to be more
efficient. But the other reason that
administers and designers give is for
combating bias, particularly racial bias. It is crucial to acknowledge
that the public service system has had a deep and lasting problem with racial
inequality. We blocked all people of color
from receiving from 1935 to 1970 when
they fought back and won. One of the things that folks will
tell you around why we should move to the tools is because it makes it
possible to identify discriminatory patterns
of decision making and pull it out of the system. It is a
bias-fighting machine. The problem with the narrative is
that the assumption that there’s no
bias in the computer which we all — this is an audience that knows that we
build our biases into machines just as we build them into our
children. As also the problem is about the
way bias gets defined is the
automated decisions that I’ve been working on. Bias there was under understood
by a racial choice. It wasn’t a stemmatic and
structural factor. Let me give you a concrete
example. In Allegheny County the
screening tool, they, like every other
place in the United States has a serious
problem with racial disporalty. 18% of the youth population is in foster care. One of the things is to keep an
eye on the intake who receive reports
of abuse and make sure they are not making discriminatory
decisions the problem is that the county’s own research shows
that almost all of this disproportion, all of this discrimination is
entering at a totally different point in the process. It is not
entering at point where the caseworkers are screening calls,
it enters at the point where they are calling on families. So the community reports black
and biracial families three and a half times more often than
white families. 350%. Once that case gets to the intake,
there’s a little wit of additional discrimination that’s
added. Screeners screen in 69% of cases around black and biracial
families. But the reality is the great
majority is coming in from community referral. That’s not necessarily a data
amenable problem. That’s a cultural problem. That’s a
problem around what does a good family look like and in the
United States the family looks rich and wait. And the problem
— one of the problems with it is if you are removing
front line discretion you are removing
their ability to correct for the massive misrepresentation that
comes in. We’re using the idea of
eliminating individual and irrational bias
to allow this vast, structural bias to
sneak in the back door of the system. I think that’s really, really
dangerous. -That brings me to another
question. Is there a way to use these
automated tools for good? Like with homelessness in L.A.
county. You describe the match.com of homeless services.
That’s not how you describe it, that’s how they describe it it.
Is there a way to view the systems for good, so to speak?
If so, what do we have to do? What do solutions look like is a
better way of asking the question? Phillip, do you have any
thoughts? -My problem is as someone said
to me in L.A. , the solution to my problem
doesn’t lie in tech, it lies in a house. But it is true. In other words
the basic political decision to provide more money
for housing is absent. We’re not prepared to pay for the losers down and out, refuse to
work, dirty people to get any sort of housing. They are
awfully generous. It’s it. If you start with that, and that’s
where I think we are. No amount of good intentions,
such as the coordinated entry system. We’re really going to
do this scientifically. We’re going to bring in every possible
factor. The decision maker has to know
how to distinguish. The problem is you have to address that
underlying political problem and not think that tech can solve
it. So just a similar quote from
Gary Blais. I think this is one of the best lines in the book. You know, I think I often find
myself in rooms where it feels like what people want is a
five-point plan for building better technology. I think
there’s a lot of room to do that work. I think that’s important
work. I am also really comfortable
spoiling that notion by saying that we have some really deep
work to do. I think we really have to change the story around
poverty in the United States. The narrative that we tell, the narrative that we tell is an
aberration. It is just a small amount of people. It is a
parent. 51% of us will be below the
poverty line. Two-thirds of will need welfare. Creating space to see themselves
within the identity of poor is an important part of the work.
That’s incredibly difficult work. You also have to address
race, gender, and migration. You have to do all of those
things at the same time. It is super hard work. It is change-making, critical,
unsettling work. I think we need to move morer to
a universal system than
thermometers. One of the gut-check questions
I tend to ask engineers and designers
is does the tool increase the self-determination and dignity
of the targets? And if it was aimed at anyone but poor and
working people, would it be tolerated at all? If the answer
to that is no, then it is unacceptable from the point of
view of democracy.
Right. Now we are going to transition to questions from the Twitter
sphere. Maybe folks want to know from
you, Phillip, and you, Virginia as
well. From my vantage point, there’s
an extraordinary similarity. I did a visit at the end of last
year. At the end of this month I’m
doing a similar mission from the United
kingdom. I’m currently totally immersed
in UK public policy and the role that AI is playing if that. I
also happen to be Australian. I see exactly what’s happening
in Australia. In those three countries which are similar in
many ways but also very different in others, we see
exactly the same sort of trends. We see the same phenomenon in
terms of treating welfare — I shouldn’t
call them that. People like most of us who at
some stage needs forms of social
protection that only government can provide. I see them being demonized and
stigmatized. I see the policy which keeps
saying the solution is to get out and
work. It is employment. It is not welfare. It is not
government assistance. What we’re seeing is that even
in a full employment economy and most of those three are, the sort of
jobs that are being created are not just very precarious, but very often by
design do not offer enough income in order
for those people to survive. There are tens of thousands of American military people who are
on food stamps. There are a million and a half
retired people on military with food stamps. I met with workers who work full
time but need food stamps. So the narrative which is trying
to demonize the people is really problematic. I just want to say
one other thing which is when you said what is the role of AI.
Virginia and I both answered you have to go back to basics. You
have to have a serious commit to welfare before AI can do much.
That’s not the answer for this audience. AI is going to be
there. You can either exacerbate the
trends that are going on, or you can call attention to them. You can build in ways of high lighting what the difficulties
are. That’s what needs to be done as much as the more creative and
innovative ways that you are currently working
on. One of the ways that I tend to
try to describe the solution is, you know, we have a tendency to think that
designing in neutral is designing fairly. But, in fact, designing in
neutral gives us no gears to deal with
the actual hills and valleys and twists and turns of the landscapes that we
live if. It is like building a car with
no gears and sitting at the top of the hill in San Francisco and
being surprised when it crashes at the bottom of the hill. For me it is about building
those equity gears into the system from the beginning on
purpose bit by bit and bite by bite. I think a bit part of that is
really speaking directly to the folks who are going to be most
impacted by the systems. I think their voices are too
rarely in rooms like this. Of the tokes that see themselves as
targets of the systems. I think they have the best information
about them. I think they are also the most likely to be accountable for good
solutions, good-lasting solutions. The idea that people have
solutions for the problem. Just one quick thing to wrap up is it seems like automation undermines
the empathy that we need to drive
home the social problems. Would you agree with that?
One of the things I talk about in the book is that the system act as
empathy overrides. They are the release valve
allowing us to outsource the most difficult
as a political community to technology. Again we see that
as an administrative solution not a political choice. I think
we need to sit with the fact that we’re making these choices
based on the assumption there’s natural
austerity and not enough for anyone, and
there’s nothing we can do. We have to challenge that and
recognize that we’re making political choices and move on from there.
Any last words? I think Virginia’s book is
superb in terms of the stories that she
tells and brings from the grounds. That brings the negligence which
is not — dimension. You can privatize a lot of
social provision. You are turning it over to companies that want to have a set of box. It is not — tell me, Virginia,
how is your husband doing? What’s the problem with the
children? Is that getting any better? How can I factor that in?
That’s what social protection and welfare are all about. They can’t be done by
automation. Please join me in thanking our
two wonderful panelist. -(Applause). Our text speaker is Kevin de
Liban. I got to know your work through
the algorithms. I was hearing about the amazing cases that
you’ve been working on. I thought maybe we would start
with our time on the clock by asking about the extraordinary
case you’ve been working on around Medicaid in
Arkansas.-legal aid attorneys are in the trenches helping those who can’t afford
lawyers with the day-to-day needs. It could be things like health
and Medicaid. They come to us when there’s no other option.
What we were seeing is early 2016 we started getting an inordinate
number of calls of people chain complaining of the same issue. A Medicaid will pay for a
caregiver for someone who has physical
disabilities, help with eating and turning and getting out of
bed. People said I’ve been getting eight hours of day of
care for 15 years. Somebody just came and said the
computer said I could only get five hours a day of care. The state of Arkansas had
instituted algorithms to decide how much in-home care the
people were going to get. The best case is 5.5 hours of care. For someone who has cerebral palsy, you are lying in
your own waste. You are getting pressure sores,
because nobody is there to turn you. It didn’t make a lot of
sense. Staying at home is not only
better for the dignity, but it is better for the bottom line.
Because it costs less than nursing home care. -Tell me how you have started to
investigate this. How do you look in to it? ->>All we know the clients are
saying the nurses that came out said the computer did it. We finally got the algorithm. I have no background in computer
coding. It was just a lot of cozy time
with the algorithm in the evenings. Very fun. -Did you teach yourself to code?
No. I could. It’s a bunch of if then statements. I got to
the point where if somebody gave me an assessment, I could
figure out where they would fall in the algorithm. What does accountability look
like for these systems? -It is not just lawyers in suits
carrying big sledge hammers. That’s a big part of it. We
knew that no judge is going to tell the state that you have to
provide eight hours a day of care. Which is barely enough.
No judge is going to say you have to provide eight hours or
ten whatever it might be. They might say you can’t cut
down. They are not going to build policy. We knew from the
starts of limits. That’s an important thing. We used the litigation as a
rallying point for a passive public education that engaged
the people post affected. We put out educational
information. We did all sorts of
presentations. We produced videos of our client’s lives
with their on sent and approval. All of this information act I
havely pushed through social media and traditional media and
so forth ended up empowering our clients to take that and then go
run with it. They were calling legislators
and doing change. org petitions and doing some
mutual aid sharing. Once you have the people most
affected complimented by litigated and complimented by
policy analysis and everything else. You’ve got some sort of
structure to make sure that substantive justice prevails.
The people — my clients, some of whom are watching, get the
care they need. Not that we just default to some
sort of procedural fairness kind of
posture. Excellent. How do you think
lawyers and researchers can work together?
This is key. I figure out the key, and how to
open the can with one. I can’t build one. And that’s where the
researchers come in. Is if validated? Is the software
correct? Are all of the projections and
underlying assumptions on point? That’s the information that I
was lacking that limited my legal challenge to more
procedural bases that I thought I could win, because I didn’t
know how to prove or have the expertise to prove right away the algorithm is
crap. It doesn’t do what it was supposed to do. I could track
the ways it was arbitrary. -You are bares the lead here. How did the case resolve?
Okay. It is a win. All of us do gooders we have
very limited ability to appreciate wins. We know the
next terrible thing is right behind the stage curtain right
over there. Not Meredith. But —
(Laughter). We invalidated the algorithm.
Then the state wanted to bring it back. We’ve been using this
illegal thing for so long that we have no
other way to do it other than using the illegal thing. They
said we’re not going to provide them services if you don’t let
them do this. They let them reinstate it for
two months. This is not holistic. We don’t want it for
the people of Arkansas. It is gone. That algorithm is dead. What is the state going to do in
January? Another algorithm. We’re hoping that at this point
not only are we smarter, we’re hoping that the state has
learned some lessons. Now we’re in a position where we’ve got
more resources, we’ve got knowledge, we have active
community members where we can go and transmit the
message. Do right by people. You are not going to wear us out
and not going to out work us. We’re not going away. We’re
smarter than you. And we’re coming for you.
(Laughter). (Applause). It is such a pleasure to hear
about this work. It’s been extraordinary to meet you and so
many of the public interest lawyers. Can you please give a
big round of applause to Kevin? (Applause). I hate to say it, but we’re
facing the final panel for tonight. We could keep going
for many hours. We’re going to close out with one last panel.
It is about the relationship between research, activism, and
accountability. Would you please welcome
Meredith Whittaker who will be chairing
my co-founder of all things.>>Hello. Good evening. It is
a delight to be here on the last panel. We’re looking at research,
organizing, and accountability. We have a lot of work to do. You can spend some cozy times,
that’s not going to give you the social implications. That speaks to the need to join
the people on the front ground
living the impact of the systems with an understanding of how the
systems are designed and, you know, the intention of these
systems. So I am just delighted to be
here with, you know, two people who I think really exemplify this kind of
socially engaged work. Lucy Suchman is a pioneering
research one of the architects of
human-computer interaction. She spent 20ees at Xerox where
you defined that field and focused on the technical aspects
where humans met machines. Her recent works looks at
autonomous weapons and the danger of an AI-enabled
military. It is engaged with organizing
around the topics. I’m really delighted to be
joined by our own Rashida Richardson. She joins us from the ACLU and
is our policy director. She looks at the lived experiences
of the systems and figuring out ways to empower activist, organizers,
and civil society leaders to push back
against some of the problems that the systems cause. So it
is great to be with you and before we get started with
questions, I just want to remind you #A
Inow2018. We’ll be taking one of your questions at the end of
the panel. With that, Lucy, I would love
to invite you to get us started. Your work looked at the context
of sort of human interactions with the systems. Where did the best laid code go
wrong when it meets the fresh air? That really shaped the
field of understanding how do we live amongst these systems?
What lessons do you think we can kind of learn from your
approaches as we strive to create better accountability on the ground? First of all, thanks to Kate and Meredith and AI Now and also to
all of you. This has been a marathon section. It is
fabulous that you are still with us. So I mean — Meredithing you characterized me as an architect
of AI. I was a very accidental
researcher in area of human and computer
reaction. I went to Xerox as a PhD student
in anthropology. It would be way too long to tell
you how I ended up there. It was at one of the moments of
AI’s AI’s asen dents. Also the idea of AI and
interactive computing. I was really intrigued by the
idea of intelligence and interactivity as they were being reworked through a
kind of computational imaginary there. I guess a lot of my work
since then has been about trying to
articulate both the incredible — incredibly
intimate relations that we have with our machines and also the
differences that matter between humans and machines. And I’ve been in that context
tracking developments in AI and robotics. And I guess for me there’s a
really important distinction between
projects in humanoid robotics which are attempting to create machines
until the image of the individual
autonomous human subject and develops more recently in — that are really tied to, you
know, Moore’s Law is all cited as the thing that’s going to take us
inevitable to the similarity. It is really about speed and
storage capacity. I think it is the speed of
computation, the storage capacity, and the extent of
networking which has made the real difference. There we’re
talking about data analytics and a lot of the things we’ve been
talking about tonight. When we come to humanoid
robotics, AI and robotics which I’ve been tracking quite
carefully, I would argue that practically no progress has been
made. And I think the reason for that
is the difficulty of actually incorporating into technologies
what we think of as — basically knowing what’s going on around
you. So the idea of context which isn’t, you know, if we’re
in a container and we have to recognize it. It is something that we’re
actively and in an ongoing dynamic way
co-creating together. And that sense of interactivity continues to escape projects in
AI robotics. Maybe if we come back to talk
about weapon systems. I think that’s exreamly consequential. I’ve started to engage as the
military refers to as situational awareness. It is
fundamental to — in particular to all military operations
presuppose the identification of a legitimate target of the
enemy. And that’s the place where these
problems that we’ve been talking about, that’s the thread that
connects the systems that we’ve been engaging with. Who is being identified as an imminent threat and on what
grounds? -Yeah. It gets to the heart of
the question when is human party has situational awareness.
Rashida, I would love to turn to you there. Obviously that’s
going to require a lot of different perspectives. You’ve
worn a lot of different hats. You come from AI Now from the
ACLU, you’ve been engaged in social justice and now AI
research around policier issues that reality to the deployment
of these systems. How do you see the joining of
these perspectives and the movement and research being done
given you’ve occupied so many of the positions? A lot of what we’re talking
about is the application of complex, social,
political, and economic issues. I think there’s the temptation
to treat researchers and advocates
separate not collaborative. They both have something to
bring to the table. It shouldn’t be a blank slate that
needs to be taught something. There’s a few things to keep in
mind with both groups. I’ll end with what they can do. The
first is I think it is important for both — I’ll say we. I
guess I wear both hats. For us to be both introspective
and honest about what we know and what we don’t know. Because
I think that there are a lot of, sort of, blind spots and
inherit biases that we have from our individual positions from
power and privilege that leads to a lot of the unknown. No one wants to be honest or
open or open to criticism about that. I also think that all of
the work whether it is research or advocacy needs to be grounded
in reality. Both Kevin’s panel and the panel before that showed
that the harms and consequences of what these issues that we’re talking about are very
real. They impact real people. There’s also a temptation to
think about the issues only
theoretically. That’s extremely harmful right
now. It is sort of to break apartment
the group. It needs to be collaborative in order to inform the advocacy that’s collaborative and diverse.
That’s a great answer. We’re seeing a lot of movement push
back against — we’ve met a lot of people who are pushing back
from instances of harm. We’re also seeing kind of tech
work that’s organized. That’s something I’ve been involved and
contributing my research to. Lucy, I know you’ve been engaged
in it as well. I would love to sort of, you know, two back to
history a little bit; right? Because while this recent wave
of concerned ethical employees at big tech companies has been
fairly surprising and heartening to had many of us, it is not the
first. I would love for you to tell us a little bit about the computer
professionals for social responsibility. Which can be
seen as a predecessor to what we’re seeing now. What lesson does the movement
have for social issues?
-Sure. I went in 1978. And in 1981 an electrical
engineer sent out a possessage to an
e-mail list called anti-war it is called
anti-war up arrow. We used them to differentiate
distribution list from individual persons. He sent out the message to
anti-war up arrow. They have worked on the stage system. It is the semiawe thon mouse
ground environment that was developed from 1950 to 1980. This was the Norad early warning system against incoming Soviet
missiles during the Cold War. He had worked on stage. He was
looking at what was happening in the development of launch on
warning systems within nuclear weapons
systems. And he was tremendously concerned about
launch on warning. It was very destabilizing in terms of the sort of so-called logics of
mutually assured destruction. It was
really crucial in that respect. But also based on what we knew
from having worked on the stage system, we saw how dangerous the
hair trigger unreliable and there were
arguments warning system was how inherently about the inherit
unreliability of the system which carry forward. To our
situation today. So he — a few years later we
founded computer professionals for social responsibility with
people at P.A.R.K. and Stanford. We basically tried to make the technical arguments for the
inherit dangers and reliabilities on launch on
warning. Interestingly in 1983, there was the kind of companion. It was Ronald Reagan’s strategic defense
program. Star Wars. And now we have the Jedi and
have returned to the sequel. It was basically an AI
initiative with something for each of the branches of the
armed services. And with two others I wrote a
piece for the bulletin of the
scientists where we critiqued the strategic computing
initiative. That was published in 1983. It was pretty well be
recycled and it would still apply.
Sadly. I would love to stay on this for a moment with you Lucy.
Just about practical perspective. I’m sure there are
people in the audience that work for the companies that are
confronting the issues. What works? What would you advice
the people to do who are often times really
feeling tensions within themselves and
employer? -Right. Our actions were not directed at
Xerox itself. We were working in a base with
strong ties to universities. That’s where the academy and
industry and that — in those networks
were very closely joined. So we weren’t threatening our
employer directly. That’s a really important issue. It was
— I think — very much through the alliance with the academic
networks that that really grew. And, you know, it is really
heartening to see what’s happening with the
tech workers and with Google and your
incredibly important work in relation to the organizing against Project
Maven and then I would part of the group
who put together a kind of academic, scientist letter in
support of that effort. So those coalitions, I think, are
great. Of course, you know, we didn’t really have the kind of network-based
organizing tools. We had e-mail lists. That was about it. So the advent of the web. You
know, the way that things now travel from a letter that then
gets posted on the web that gets picked up by the media and turns
into the online op ed. That’s really accelerated and
facilitated that as well. -Yeah. In the case Maven, they have
visibility into what these are. Rashida, a lot of your work is
kind of grappling with systems that are incredible by design.
We talked about the black boxes on black boxes. These are still
deeply, powerfully, affecting people’s lives. I would love
your incites on how do you do work around those systems and
how do you marshal the communities who can go their many dimensions? -Transparency and
accountability. I would add work as hard. It’s been in part because
there’s so much we don’t know. One way is to engage in the
research with advocates. The issues vary by jurisdiction. How much we know may vary based
on your local lawyers or
procurement processes. One of the issues is frying to
understand what’s going on. I think as far as strategy and
solutions, that needs to be informed by the local context. And that work — I’ll use New
York as an example, not only because it
is our hometown, but because it is a good example. Here in your intro with Kate you
guys mentioned the New York task force that’s looking at government use of
automated decision systems. We’ve been involved in that work
from getting the legislation passed through to when the task force was
announced. Just so the audience has an understanding, the task
force has to write a report by the end of next year with
recommendations on a number of issues like what should the
public know about where and how these systems are being used, if a group or
individual is harmed, what should readdress
look like? And more technical questions
like how can we or can we archive the systems especially given that municipal
budgets are not ever-expanding. In that work we constantly met,
and I think a common stream both within government and within
civil society that this is an important issue. But I think
everyone kind of struggled with how do you actually address
these issues? And I don’t want to sort of fall
into the let’s punt to the task force until they can figure it
out in December of 2019. So what we ended up doing is
trying to do some of that hard research
and work the summer and engage people from researchers to the
average New Yorkers. I just talked to my family or people on the train about these issues
and trying to figure out what the redress standard should be. If we’re not grappling with what
could work and what doesn’t work, we’re not going to get very far a year
from now. That process was helpful. It allowed us to
engage others. We listed a group of both individuals and
organizational experts where the task force to use. That was a
useful process. Even though we did list three
pages as people, I remember one person pointing out, hey, under
child welfare you listed a lot of great lawyers in people who
understand family law. You have no one representing the parents’
voices. That was just an omission. You
only know within your network. It was a great opportunity to
get more voices at the table and more perspective and understand
what is going on here and what’s the best way to address it.
-Yeah. What was the most illuminating
insight? Is there something that someone told you that you
wouldn’t have figured out if you didn’t engage across the
coalition? -I think no one has the right
answer. Often to try to figure out the
redress issue the place is legal theories on impact. There’s a
lot of different theories. U.S. law is regressive. None of the
prevailing ones are the ones I would choose. But keeping that
in mind and trying to be idealistic, I just talked to
some of the prevailing standards that are under housing
discrimination laws. Others are in employment law.
And the EOC standards. I talked to employment and
housing lawyers. What don’t you like? I’ve never — we’re so used to
working in regressive, it is hard to imagine alternatives.
That was the struggle in the process of how do I imagine a solution
that is not within the oppressive society that I live
in. The answer is I don’t know — but we tried. I think it is
— I think it is forcing that conversation in trying to
imagine alternatives. And sort of being resistant to
the idea that that’s not practical
as a good counterargument. -Yes. I love that point. It is a great onramp to return
to the topic of autonomous weapons. It is the negative example of
the excess and danger of the technology. Lucy, you’ve worked
so much on the issue. You’ve pushed back against the
autonomous weapons. You’ve organized against them and
researched them. I would you to connect the issue of rejecting or holding these
systems accountable with the broader
issues of accountability. What lessons can we learn about
fighting against the frightening and oppressive society’s terms are
wrapped into the development of these systems. -yeah. We haven’t talked about
militarism. These things are incredibly joined up. To me I think it was Phillip
Alston who mentioned the us/them —
this is about — as I said, a lot of the
ways in which technology has been framed
as the avenue for security and the
possibility of a sort of perfect defensive system which was the Star Wars vision,
fantasy is now — I mentioned Jed i, it’s the joint enterprise system that will join
up all of the U.S. pill tear operations around the world in
which they are as you know as many of the — an overwhelming
number. So this is — and the premise
that these systems are going to give
us the ability to discriminate between
us/them, the good guys and the bad guys. Who constitutes a threat? That runs through all of the
issues of sorting and classification and
discrimination that we’ve been talking about. And in the case
of the lethal autonomous weapon systems, the
idea is that — this is an issue that’s being debated at the U.N. in the context of what’s called
the convention on certain conventional weapons. These are weapons considered to
be indiscriminate. Land mines would be an example.
The critical functions that have been identified by the
international committee for the Red Cross and others who are
campaigning around this are are target identification and the
initiation of attack. The idea is that we should ban the
automation of those two critical functions. We should not create weapons
where we think we have delegated to the
weapons system sufficient discriminatory
capacities that it can identify who constitutes an eminent threat. To me that’s directly tied to
the Project Maven work. It is responsive to the vast network
of drone surveillance that the U.S. military has created which is
producing orders of video footage that’s completely
unusable because of the massive size. And so the idea is that there
will be a kind of triage. We heard that word before. A kind of algorithmic automation
of the first path of analysis of the video footage. That it will
be — there will be certain classes and certain categories of objects, vehicles,
buildings, persons, or configurations of persons that
will then lead to the — lead to the claim is the handing off of
the things to human analysts. And this will make the whole
system more efficient. Those of us who have been
following the drone — the U.S. drone program, and again Phillip
Alston has played a really crucial role
here — know that the — the so-called precision of those systems is a complete
fallacy. That, you know, there are —
there’s evidence that, for example, in
the U.S. drone program in Pakistan from
about 2006 to 2014 about 75% of the
people killed by that program we don’t know who they were. And another — about another 20%
were positively identified as women
or children therefor civilians. And there’s a 5% remainder who
are the people who we actually know who was killed, and there’s an argument
that they were a legitimate target. We’re talking about a
massively inaccurate program that the proposal now is to automate. Really bad
idea. And I think the consistent
thread — -yeah. -I think the consistent thread
is profiling. Profiling of various kinds. The crudest forms of profiling
that’s the basis for the proportedly
cutting-edge technological systems. They are informed by
the absolutely crudest and in many cases most long-standing forms of
discriminatory stereotyping that we’ve been — had with us for a
very long time. time. -Yeah. They are so glossy. I want to ask Rashida, how do
they push back? -I think there’s a tendency to
think that people don’t understand the issues. I think most people, especially
those who are abouted not only by the systems but the prior
existence — sort of forms of the systems so government
systems with humans making bias
decisions understand the problem. I think being open to
the fact that people could know the solutions to their own
problems and respecting the intelligence experience and
expertise of everyone so that way they can
engage and have a place at the table to come up with solutions.
-Yeah. I just polled a question from the audience. I would love
your thoughts on this question. What does it mean to organize
tech workers who already have so much privilege? Are people
actually putting their jobs on the line?
-Okay. Great question. Great question. I mean first of all
— I think part of my response to that would be not all tech
workers have a lot of privilege. If we include among tech workers
the people who Virginia Eubanks was
talking about who the tech workers
coalition incorporates many of those people. So if we treat
tech workers really broadly, then there’s a highly heterogeneous group. It is a
legitimate question. Some tech workers who are
speaking out are risking their jobs. You could argue they are
relatively well positioned to go and find other issues. So
that’s — then, of course, you know, I think we all risk in
proportion to the amount of privilege or
precarety that we have otherwise in life. And so, yeah, I think it is a
legitimate question. But I also think again that, you
know, there are some really
interesting possibilities for the
mobilization of the — of a coalition of workers
inside big tech companies. -I think tech workers are in a
unique position to have influence. I’m all hesitant to put all of
my faith in the tech community as
the leaders of some resistance. It if we’re being honest about
how homogenous the group is, they don’t understand. We all
don’t understand a lot. Which was in my first response. They don’t understand the
history or the social, political, and economic context of the ways the tools
they are creating will be implemented. I think some of the advocacy is
great. You have people who are in higher positions of power and
have higher access to privilege and power
that can move the ball forward and move it in some direction.
But I don’t want to think that’s sort of what’s going to change
things. They think it needs to come from everyone else who is
affected by these systems and those who bring other
perspectives from the table. I don’t think it is any one group
that’s going to result in any revolutionary change is what we
really need. It needs to — I think it is also upon the tech
workers to realize they have certain power and influence.
But they are not the ones that should be deciding what is ethical and
what is not and what is moral and what
is not. -Yeah. Hard agree. I think
with that we’re going to wrap it up. Thank you so much. It was
lovely to have you on stage here. I’m going to stay up
here. Thank you. I will remain for a moment and
invite Kate Crawford, co-founder. Thank
you. All that’s left from us are some closing words. A big thanks to the speakers. A
big thanks to NYU. A big thanks to everybody in the community
who has backed the effort, offered support, and guidance to
set up an institute here. I want to thank the John D. and Catherine Foundation for
their support. And a particular thanks for his amazing work on the
visualization tonight. That was Varoon. I want to thank the first cohort in
production. Particularly Emily and Kate and
the Good Sense Team and the NYU
families team. We are really grateful as well
to the volunteers and friends and everyone who helped and contributed so much
energy and enthusiasm and made this
happen. Finally a colossal and huge word
of thanks to the director of
operations, our lead producer in everything, Mariah Peebles. We couldn’t do any of this
without you. -I’m also going to throw in a
final thanks to Meredith for being
amazing and an extraordinary person to do this work with.
The final thanks is going to go with you. This year has thrown off some of
the visions. We see the enormous potential
for action and change. This community here in the room and
on the live stream you have a powerful role to play right now.
Stay in touch. We’re on Twitters. We have old school mailing
addresses. Thank you for being part of it. Have a great night.
Thank you.

Tags: , , , , , , , ,

There are no comments yet

Why not be the first

Leave a Reply

Your email address will not be published. Required fields are marked *