Tag Archives: Weapons of Math Destruction

Predictive policing in Vancouver—the first jurisdiction in Canada to employ a machine learning system for property theft reduction

Predictive policing has come to Canada, specifically, Vancouver. A July 22, 2017 article by Matt Meuse for the Canadian Broadcasting Corporation (CBC) news online describes the new policing tool,

The Vancouver Police Department is implementing a city-wide “predictive policing” system that uses machine learning to prevent break-ins by predicting where they will occur before they happen — the first of its kind in Canada.

Police chief Adam Palmer said that, after a six-month pilot project in 2016, the system is now accessible to all officers via their cruisers’ onboard computers, covering the entire city.

“Instead of officers just patrolling randomly throughout the neighbourhood, this will give them targeted areas it makes more sense to patrol in because there’s a higher likelihood of crime to occur,” Palmer said.

 

Things got off to a slow start as the system familiarized itself [during a 2016 pilot project] with the data, and floundered in the fall due to unexpected data corruption.

But Special Const. Ryan Prox said the system reduced property crime by as much as 27 per cent in areas where it was tested, compared to the previous four years.

The accuracy of the system was also tested by having it generate predictions for a given day, and then watching to see what happened that day without acting on the predictions.

Palmer said the system was getting accuracy rates between 70 and 80 per cent.

When a location is identified by the system, Palmer said officers can be deployed to patrol that location. …

“Quite often … that visible presence will deter people from committing crimes [altogether],” Palmer said.

Though similar systems are used in the United States, Palmer said the system is the first of its kind in Canada, and was developed specifically for the VPD.

While the current focus is on residential break-ins, Palmer said the system could also be tweaked for use with car theft — though likely not with violent crime, which is far less predictable.

Palmer dismissed the inevitable comparison to the 2002 Tom Cruise film Minority Report, in which people are arrested to prevent them from committing crimes in the future.

“We’re not targeting people, we’re targeting locations,” Palmer said. “There’s nothing dark here.”

If you want to get a sense of just how dismissive Chief Palmer was, there’s a July 21, 2017 press conference (run time: approx. 21 mins.) embedded with a media release of the same date. The media release offered these details,

The new model is being implemented after the VPD ran a six-month pilot study in 2016 that contributed to a substantial decrease in residential break-and-enters.

The pilot ran from April 1 to September 30, 2016. The number of residential break-and enters during the test period was compared to the monthly average over the same period for the previous four years (2012 to 2015). The highest drop in property crime – 27 per cent – was measured in June.

The new model provides data in two-hour intervals for locations where residential and commercial break-and-enters are anticipated. The information is for 100-metre and 500-metre zones. Police resources can be dispatched to that area on foot or in patrol cars, to provide a visible presence to deter thieves.

The VPD’s new predictive policing model is built on GEODASH – an advanced machine-learning technology that was implemented by the VPD in 2015. A public version of GEODASH was introduced in December 2015 and is publicly available on vpd.ca. It retroactively plots the location of crimes on a map to provide a general idea of crime trends to the public.

I wish Chief Palmer had been a bit more open to discussion about the implications of ‘predictive policing’. In the US where these systems have been employed in various jurisdictions, there’s some concern arising after an almost euphoric initial response as a Nov. 21, 2016 article by Logan Koepke for the slate.com notes (Note: Links have been removed),

When predictive policing systems began rolling out nationwide about five years ago, coverage was often uncritical and overly reliant on references to Minority Report’s precog system. The coverage made predictive policing—the computer systems that attempt to use data to forecast where crime will happen or who will be involved—seem almost magical.

Typically, though, articles glossed over Minority Report’s moral about how such systems can go awry. Even Slate wasn’t immune, running a piece in 2011 called “Time Cops” that said, when it came to these systems, “Civil libertarians can rest easy.”

This soothsaying language extended beyond just media outlets. According to former New York City Police Commissioner William Bratton, predictive policing is the “wave of the future.” Microsoft agrees. One vendor even markets its system as “better than a crystal ball.” More recent coverage has rightfully been more balanced, skeptical, and critical. But many still seem to miss an important point: When it comes to predictive policing, what matters most isn’t the future—it’s the past.

Some predictive policing systems incorporate information like the weather, a location’s proximity to a liquor store, or even commercial data brokerage information. But at their core, they rely either mostly or entirely on historical crime data held by the police. Typically, these are records of reported crimes—911 calls or “calls for service”—and other crimes the police detect. Software automatically looks for historical patterns in the data, and uses those patterns to make its forecasts—a process known as machine learning.

Intuitively, it makes sense that predictive policing systems would base their forecasts on historical crime data. But historical crime data has limits. Criminologists have long emphasized that crime reports—and other statistics gathered by the police—do not necessarily offer an accurate picture of crime in a community. The Department of Justice’s National Crime Victimization Survey estimates that from 2006 to 2010, 52 percent of violent crime went unreported to police, as did 60 percent of household property crime. Essentially: Historical crime data is a direct record of how law enforcement responds to particular crimes, rather than the true rate of crime. Rather than predicting actual criminal activity, then, the current systems are probably better at predicting future police enforcement.

Koepke goes on to cover other potential issues with ‘predicitive policing’ in this thoughtful piece. He also co-authored an August 2016 report, Stuck in a Pattern; Early evidence on “predictive” policing and civil rights.

There seems to be increasing attention on machine learning and bias as noted in my May 24, 2017 posting where I provide links to other FrogHeart postings on the topic and there’s this Feb. 28, 2017 posting about a new regional big data sharing project, the Cascadia Urban Analytics Cooperative where I mention Cathy O’Neil (author of the book, Weapons of Math Destruction) and her critique in a subsection titled: Algorithms and big data.

I would like to see some oversight and some discussion in Canada about this brave new world of big data.

One final comment, it is possible to get access to the Vancouver Police Department’s data through the City of Vancouver’s Open Data Catalogue (home page).

Robots in Vancouver and in Canada (two of two)

This is the second of a two-part posting about robots in Vancouver and Canada. The first part included a definition, a brief mention a robot ethics quandary, and sexbots. This part is all about the future. (Part one is here.)

Canadian Robotics Strategy

Meetings were held Sept. 28 – 29, 2017 in, surprisingly, Vancouver. (For those who don’t know, this is surprising because most of the robotics and AI research seems to be concentrated in eastern Canada. if you don’t believe me take a look at the speaker list for Day 2 or the ‘Canadian Stakeholder’ meeting day.) From the NSERC (Natural Sciences and Engineering Research Council) events page of the Canadian Robotics Network,

Join us as we gather robotics stakeholders from across the country to initiate the development of a national robotics strategy for Canada. Sponsored by the Natural Sciences and Engineering Research Council of Canada (NSERC), this two-day event coincides with the 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2017) in order to leverage the experience of international experts as we explore Canada’s need for a national robotics strategy.

Where
Vancouver, BC, Canada

When
Thursday September 28 & Friday September 29, 2017 — Save the date!

Download the full agenda and speakers’ list here.

Objectives

The purpose of this two-day event is to gather members of the robotics ecosystem from across Canada to initiate the development of a national robotics strategy that builds on our strengths and capacities in robotics, and is uniquely tailored to address Canada’s economic needs and social values.

This event has been sponsored by the Natural Sciences and Engineering Research Council of Canada (NSERC) and is supported in kind by the 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2017) as an official Workshop of the conference.  The first of two days coincides with IROS 2017 – one of the premiere robotics conferences globally – in order to leverage the experience of international robotics experts as we explore Canada’s need for a national robotics strategy here at home.

Who should attend

Representatives from industry, research, government, startups, investment, education, policy, law, and ethics who are passionate about building a robust and world-class ecosystem for robotics in Canada.

Program Overview

Download the full agenda and speakers’ list here.

DAY ONE: IROS Workshop 

“Best practices in designing effective roadmaps for robotics innovation”

Thursday September 28, 2017 | 8:30am – 5:00pm | Vancouver Convention Centre

Morning Program:“Developing robotics innovation policy and establishing key performance indicators that are relevant to your region” Leading international experts share their experience designing robotics strategies and policy frameworks in their regions and explore international best practices. Opening Remarks by Prof. Hong Zhang, IROS 2017 Conference Chair.

Afternoon Program: “Understanding the Canadian robotics ecosystem” Canadian stakeholders from research, industry, investment, ethics and law provide a collective overview of the Canadian robotics ecosystem. Opening Remarks by Ryan Gariepy, CTO of Clearpath Robotics.

Thursday Evening Program: Sponsored by Clearpath Robotics  Workshop participants gather at a nearby restaurant to network and socialize.

Learn more about the IROS Workshop.

DAY TWO: NSERC-Sponsored Canadian Robotics Stakeholder Meeting
“Towards a national robotics strategy for Canada”

Friday September 29, 2017 | 8:30am – 5:00pm | University of British Columbia (UBC)

On the second day of the program, robotics stakeholders from across the country gather at UBC for a full day brainstorming session to identify Canada’s unique strengths and opportunities relative to the global competition, and to align on a strategic vision for robotics in Canada.

Friday Evening Program: Sponsored by NSERC Meeting participants gather at a nearby restaurant for the event’s closing dinner reception.

Learn more about the Canadian Robotics Stakeholder Meeting.

I was glad to see in the agenda that some of the international speakers represented research efforts from outside the usual Europe/US axis.

I have been in touch with one of the organizers (also mentioned in part one with regard to robot ethics), Ajung Moon (her website is here), who says that there will be a white paper available on the Canadian Robotics Network website at some point in the future. I’ll keep looking for it and, in the meantime, I wonder what the 2018 Canadian federal budget will offer robotics.

Robots and popular culture

For anyone living in Canada or the US, Westworld (television series) is probably the most recent and well known ‘robot’ drama to premiere in the last year.As for movies, I think Ex Machina from 2014 probably qualifies in that category. Interestingly, both Westworld and Ex Machina seem quite concerned with sex with Westworld adding significant doses of violence as another  concern.

I am going to focus on another robot story, the 2012 movie, Robot & Frank, which features a care robot and an older man,

Frank (played by Frank Langella), a former jewel thief, teaches a robot the skills necessary to rob some neighbours of their valuables. The ethical issue broached in the film isn’t whether or not the robot should learn the skills and assist Frank in his thieving ways although that’s touched on when Frank keeps pointing out that planning his heist requires he live more healthily. No, the problem arises afterward when the neighbour accuses Frank of the robbery and Frank removes what he believes is all the evidence. He believes he’s going successfully evade arrest until the robot notes that Frank will have to erase its memory in order to remove all of the evidence. The film ends without the robot’s fate being made explicit.

In a way, I find the ethics query (was the robot Frank’s friend or just a machine?) posed in the film more interesting than the one in Vikander’s story, an issue which does have a history. For example, care aides, nurses, and/or servants would have dealt with requests to give an alcoholic patient a drink. Wouldn’t there  already be established guidelines and practices which could be adapted for robots? Or, is this question made anew by something intrinsically different about robots?

To be clear, Vikander’s story is a good introduction and starting point for these kinds of discussions as is Moon’s ethical question. But they are starting points and I hope one day there’ll be a more extended discussion of the questions raised by Moon and noted in Vikander’s article (a two- or three-part series of articles? public discussions?).

How will humans react to robots?

Earlier there was the contention that intimate interactions with robots and sexbots would decrease empathy and the ability of human beings to interact with each other in caring ways. This sounds a bit like the argument about smartphones/cell phones and teenagers who don’t relate well to others in real life because most of their interactions are mediated through a screen, which many seem to prefer. It may be partially true but, arguably,, books too are an antisocial technology as noted in Walter J. Ong’s  influential 1982 book, ‘Orality and Literacy’,  (from the Walter J. Ong Wikipedia entry),

A major concern of Ong’s works is the impact that the shift from orality to literacy has had on culture and education. Writing is a technology like other technologies (fire, the steam engine, etc.) that, when introduced to a “primary oral culture” (which has never known writing) has extremely wide-ranging impacts in all areas of life. These include culture, economics, politics, art, and more. Furthermore, even a small amount of education in writing transforms people’s mentality from the holistic immersion of orality to interiorization and individuation. [emphases mine]

So, robotics and artificial intelligence would not be the first technologies to affect our brains and our social interactions.

There’s another area where human-robot interaction may have unintended personal consequences according to April Glaser’s Sept. 14, 2017 article on Slate.com (Note: Links have been removed),

The customer service industry is teeming with robots. From automated phone trees to touchscreens, software and machines answer customer questions, complete orders, send friendly reminders, and even handle money. For an industry that is, at its core, about human interaction, it’s increasingly being driven to a large extent by nonhuman automation.

But despite the dreams of science-fiction writers, few people enter a customer-service encounter hoping to talk to a robot. And when the robot malfunctions, as they so often do, it’s a human who is left to calm angry customers. It’s understandable that after navigating a string of automated phone menus and being put on hold for 20 minutes, a customer might take her frustration out on a customer service representative. Even if you know it’s not the customer service agent’s fault, there’s really no one else to get mad at. It’s not like a robot cares if you’re angry.

When human beings need help with something, says Madeleine Elish, an anthropologist and researcher at the Data and Society Institute who studies how humans interact with machines, they’re not only looking for the most efficient solution to a problem. They’re often looking for a kind of validation that a robot can’t give. “Usually you don’t just want the answer,” Elish explained. “You want sympathy, understanding, and to be heard”—none of which are things robots are particularly good at delivering. In a 2015 survey of over 1,300 people conducted by researchers at Boston University, over 90 percent of respondents said they start their customer service interaction hoping to speak to a real person, and 83 percent admitted that in their last customer service call they trotted through phone menus only to make their way to a human on the line at the end.

“People can get so angry that they have to go through all those automated messages,” said Brian Gnerer, a call center representative with AT&T in Bloomington, Minnesota. “They’ve been misrouted or been on hold forever or they pressed one, then two, then zero to speak to somebody, and they are not getting where they want.” And when people do finally get a human on the phone, “they just sigh and are like, ‘Thank God, finally there’s somebody I can speak to.’ ”

Even if robots don’t always make customers happy, more and more companies are making the leap to bring in machines to take over jobs that used to specifically necessitate human interaction. McDonald’s and Wendy’s both reportedly plan to add touchscreen self-ordering machines to restaurants this year. Facebook is saturated with thousands of customer service chatbots that can do anything from hail an Uber, retrieve movie times, to order flowers for loved ones. And of course, corporations prefer automated labor. As Andy Puzder, CEO of the fast-food chains Carl’s Jr. and Hardee’s and former Trump pick for labor secretary, bluntly put it in an interview with Business Insider last year, robots are “always polite, they always upsell, they never take a vacation, they never show up late, there’s never a slip-and-fall, or an age, sex, or race discrimination case.”

But those robots are backstopped by human beings. How does interacting with more automated technology affect the way we treat each other? …

“We know that people treat artificial entities like they’re alive, even when they’re aware of their inanimacy,” writes Kate Darling, a researcher at MIT who studies ethical relationships between humans and robots, in a recent paper on anthropomorphism in human-robot interaction. Sure, robots don’t have feelings and don’t feel pain (not yet, anyway). But as more robots rely on interaction that resembles human interaction, like voice assistants, the way we treat those machines will increasingly bleed into the way we treat each other.

It took me a while to realize that what Glaser is talking about are AI systems and not robots as such. (sigh) It’s so easy to conflate the concepts.

AI ethics (Toby Walsh and Suzanne Gildert)

Jack Stilgoe of the Guardian published a brief Oct. 9, 2017 introduction to his more substantive (30 mins.?) podcast interview with Dr. Toby Walsh where they discuss stupid AI amongst other topics (Note: A link has been removed),

Professor Toby Walsh has recently published a book – Android Dreams – giving a researcher’s perspective on the uncertainties and opportunities of artificial intelligence. Here, he explains to Jack Stilgoe that we should worry more about the short-term risks of stupid AI in self-driving cars and smartphones than the speculative risks of super-intelligence.

Professor Walsh discusses the effects that AI could have on our jobs, the shapes of our cities and our understandings of ourselves. As someone developing AI, he questions the hype surrounding the technology. He is scared by some drivers’ real-world experimentation with their not-quite-self-driving Teslas. And he thinks that Siri needs to start owning up to being a computer.

I found this discussion to cast a decidedly different light on the future of robotics and AI. Walsh is much more interested in discussing immediate issues like the problems posed by ‘self-driving’ cars. (Aside: Should we be calling them robot cars?)

One ethical issue Walsh raises is with data regarding accidents. He compares what’s happening with accident data from self-driving (robot) cars to how the aviation industry handles accidents. Hint: accident data involving air planes is shared. Would you like to guess who does not share their data?

Sharing and analyzing data and developing new safety techniques based on that data has made flying a remarkably safe transportation technology.. Walsh argues the same could be done for self-driving cars if companies like Tesla took the attitude that safety is in everyone’s best interests and shared their accident data in a scheme similar to the aviation industry’s.

In an Oct. 12, 2017 article by Matthew Braga for Canadian Broadcasting Corporation (CBC) news online another ethical issue is raised by Suzanne Gildert (a participant in the Canadian Robotics Roadmap/Strategy meetings mentioned earlier here), Note: Links have been removed,

… Suzanne Gildert, the co-founder and chief science officer of Vancouver-based robotics company Kindred. Since 2014, her company has been developing intelligent robots [emphasis mine] that can be taught by humans to perform automated tasks — for example, handling and sorting products in a warehouse.

The idea is that when one of Kindred’s robots encounters a scenario it can’t handle, a human pilot can take control. The human can see, feel and hear the same things the robot does, and the robot can learn from how the human pilot handles the problematic task.

This process, called teleoperation, is one way to fast-track learning by manually showing the robot examples of what its trainers want it to do. But it also poses a potential moral and ethical quandary that will only grow more serious as robots become more intelligent.

“That AI is also learning my values,” Gildert explained during a talk on robot ethics at the Singularity University Canada Summit in Toronto on Wednesday [Oct. 11, 2017]. “Everything — my mannerisms, my behaviours — is all going into the AI.”

At its worst, everything from algorithms used in the U.S. to sentence criminals to image-recognition software has been found to inherit the racist and sexist biases of the data on which it was trained.

But just as bad habits can be learned, good habits can be learned too. The question is, if you’re building a warehouse robot like Kindred is, is it more effective to train those robots’ algorithms to reflect the personalities and behaviours of the humans who will be working alongside it? Or do you try to blend all the data from all the humans who might eventually train Kindred robots around the world into something that reflects the best strengths of all?

I notice Gildert distinguishes her robots as “intelligent robots” and then focuses on AI and issues with bias which have already arisen with regard to algorithms (see my May 24, 2017 posting about bias in machine learning, AI, and .Note: if you’re in Vancouver on Oct. 26, 2017 and interested in algorithms and bias), there’s a talk being given by Dr. Cathy O’Neil, author the Weapons of Math Destruction, on the topic of Gender and Bias in Algorithms. It’s not free but  tickets are here.)

Final comments

There is one more aspect I want to mention. Even as someone who usually deals with nanobots, it’s easy to start discussing robots as if the humanoid ones are the only ones that exist. To recapitulate, there are humanoid robots, utilitarian robots, intelligent robots, AI, nanobots, ‘microscopic bots, and more all of which raise questions about ethics and social impacts.

However, there is one more category I want to add to this list: cyborgs. They live amongst us now. Anyone who’s had a hip or knee replacement or a pacemaker or a deep brain stimulator or other such implanted device qualifies as a cyborg. Increasingly too, prosthetics are being introduced and made part of the body. My April 24, 2017 posting features this story,

This Case Western Reserve University (CRWU) video accompanies a March 28, 2017 CRWU news release, (h/t ScienceDaily March 28, 2017 news item)

Bill Kochevar grabbed a mug of water, drew it to his lips and drank through the straw.

His motions were slow and deliberate, but then Kochevar hadn’t moved his right arm or hand for eight years.

And it took some practice to reach and grasp just by thinking about it.

Kochevar, who was paralyzed below his shoulders in a bicycling accident, is believed to be the first person with quadriplegia in the world to have arm and hand movements restored with the help of two temporarily implanted technologies. [emphasis mine]

A brain-computer interface with recording electrodes under his skull, and a functional electrical stimulation (FES) system* activating his arm and hand, reconnect his brain to paralyzed muscles.

Does a brain-computer interface have an effect on human brain and, if so, what might that be?

In any discussion (assuming there is funding for it) about ethics and social impact, we might want to invite the broadest range of people possible at an ‘earlyish’ stage (although we’re already pretty far down the ‘automation road’) stage or as Jack Stilgoe and Toby Walsh note, technological determinism holds sway.

Once again here are links for the articles and information mentioned in this double posting,

That’s it!

ETA Oct. 16, 2017: Well, I guess that wasn’t quite ‘it’. BBC’s (British Broadcasting Corporation) Magazine published a thoughtful Oct. 15, 2017 piece titled: Can we teach robots ethics?

Simon Fraser University (Vancouver, Canada) and its president’s (Andrew Petter) dream colloquium: women in technology

I’m a little late with this event news (sadly,. I only received the information yesterday, Sept. 20, 2017) but even with two event dates already past (happily, videos for the two events have been posted), there are still several “Women in Technology” events to attend or view live according to the Simon Fraser University (SFU) President’s Dream Colloquium: Women in Technology; Attaining, Retaining, and Promoting Diverse Talent’s webpage text by Wan Yee Lok,

Women in Technology: Attracting, Retaining and Promoting Diverse Talent is a seven-part public [emphasis mine] lecture series beginning on Sept. 13. Key experts from around the world will identify challenges to gender equity and discover solutions for improving recruitment, retention and leadership options for women.

Diversity and inclusion are critical to high-tech corporate success. Yet statistics reveal that less than 25 per cent of those working in the science, technology, engineering and math sectors (STEM) are women, and that they earn seven-and-a-half per cent less than men.

“There is a crucial need to achieve gender equality in the tech sector, especially at a time when it is growing faster than ever,” says colloquium organizer Lesley Shannon, an SFU engineering science professor. She holds the Natural Sciences and Engineering Research Council (NSERC) Chair for Women in Science and Engineering for the B.C. and Yukon region.

“We hope the colloquium will help people engage in a multidisciplinary dialogue about the value of creating more space in technology for women and other under-represented groups.”

Six of the lectures are free, except for Cathy O’Neil’s lecture on Oct. 26.

The President’s Dream Colloquium schedule is as follows:

Sept. 13: SFU KEY presents: We the Data
Juliette Powell, founder, Turing AI and WeTheData.org, author of 33 Million People in the Room

Sept. 14: Diversity 101: The Case for Diversity in Technology
Maria Klawe, president, Harvey Mudd College

Sept. 21: Women in Media and Advertising
Shari Graydon, catalyst, Informed Opinions

Oct. 12: Social Psychological Phenomena
Steven Spencer, the Robert K. and Dale J. Weary Chair in Social Psychology, Ohio State University

Oct. 26: Gender and Bias in Algorithmic Design
Cathy O’Neil, author, Weapons of Math Destruction [tickets are $5 for students; $15 for the rest of us; go here to buy tickets, click on green button in the upper right, below the banner; the event will be held at SFU’s Harbour Centre Vancouver location]

Nov 9: Gendered Language
Danielle Gaucher, associate professor, Department of Psychology, University of Winnipeg

Nov. 23: Women as Leaders and Innovators
Jo Miller, founder, Be Leaderly

Lectures will be webcast live and available on the President’s Dream Colloquium website, www.sfu.ca/womenintech.

SFU engineering science professor Lesley Shannon is the colloquium organizer as well as the Natural Sciences and Engineering Research Council (NSERC) Chair for Women in Science and Engineering for the B.C. and Yukon region.

 

As a part of the colloquium, students can enroll in a graduate course covering a broad range of topics related to diversity in the technology sector. Shannon says the course will focus on women and their role in technology as well as issues that affect other under‐represented groups.

“I hope the course will establish a foundation for future managers, supervisors, sponsors, mentors and others wanting to pursue leadership roles to work towards creating a level playing field in technology and other industries,” says Shannon.

The colloquium course (SAR 897) is still accepting students. Visit go.sfu.ca to enroll.

A reminder after the last few paragraphs of the event text, you don’t actually have to be a student to attend the lectures although for anyone who doesn’t want to make the trek up the hill (SFU is located on a hill in Burnaby, BC) for the majority of the events, there is the livestream video. For those who can’t make the scheduled times, given that both the Sept. 13 and Sept. 14, 2017 event videos have been posted, they are being pretty quick about uploading the videos afterwards.

I have mentioned Cathy O’Neil here a couple of times, more substantively in a Feb. 28, 2017 posting about a major’ big data’ collaboration between the province of BC and the state of Washington (for Cathy O’Neil, scroll down to the subsection titled: Algorithms and big data) and briefly at the end in a May 24, 2017 posting that was chiefly concerned with bias in algorithms.

Removing gender-based stereotypes from algorithms

Most people don’t think of algorithms as having biases and stereotypes but Michael Zou in his Sept. 26, 2016 essay for The Conversation (h/t phys.org Sept. 26, 2016 news item) says different, Note: Links have been removed,

Machine learning is ubiquitous in our daily lives. Every time we talk to our smartphones, search for images or ask for restaurant recommendations, we are interacting with machine learning algorithms. They take as input large amounts of raw data, like the entire text of an encyclopedia, or the entire archives of a newspaper, and analyze the information to extract patterns that might not be visible to human analysts. But when these large data sets include social bias, the machines learn that too.

A machine learning algorithm is like a newborn baby that has been given millions of books to read without being taught the alphabet or knowing any words or grammar. The power of this type of information processing is impressive, but there is a problem. When it takes in the text data, a computer observes relationships between words based on various factors, including how often they are used together.

We can test how well the word relationships are identified by using analogy puzzles. Suppose I ask the system to complete the analogy “He is to King as She is to X.” If the system comes back with “Queen,” then we would say it is successful, because it returns the same answer a human would.

Our research group trained the system on Google News articles, and then asked it to complete a different analogy: “Man is to Computer Programmer as Woman is to X.” The answer came back: “Homemaker.”

Zou explains how a machine (algorithm) learns and then notes this,

Not only can the algorithm reflect society’s biases – demonstrating how much those biases are contained in the input data – but the system can potentially amplify gender stereotypes. Suppose I search for “computer programmer” and the search program uses a gender-biased database that associates that term more closely with a man than a woman.

The search results could come back flawed by the bias. Because “John” as a male name is more closely related to “computer programmer” than the female name “Mary” in the biased data set, the search program could evaluate John’s website as more relevant to the search than Mary’s – even if the two websites are identical except for the names and gender pronouns.

It’s true that the biased data set could actually reflect factual reality – perhaps there are more “Johns” who are programmers than there are “Marys” – and the algorithms simply capture these biases. This does not absolve the responsibility of machine learning in combating potentially harmful stereotypes. The biased results would not just repeat but could even boost the statistical bias that most programmers are male, by moving the few female programmers lower in the search results. It’s useful and important to have an alternative that’s not biased.

There is a way according to Zou that stereotypes can be removed,

Our debiasing system uses real people to identify examples of the types of connections that are appropriate (brother/sister, king/queen) and those that should be removed. Then, using these human-generated distinctions, we quantified the degree to which gender was a factor in those word choices – as opposed to, say, family relationships or words relating to royalty.

Next we told our machine-learning algorithm to remove the gender factor from the connections in the embedding. This removes the biased stereotypes without reducing the overall usefulness of the embedding.

When that is done, we found that the machine learning algorithm no longer exhibits blatant gender stereotypes. We are investigating applying related ideas to remove other types of biases in the embedding, such as racial or cultural stereotypes.

If you have time, I encourage you to read the essay in its entirety and this June 14, 2016 posting about research into algorithms and how they make decisions for you about credit, medical diagnoses, job opportunities and more.

There’s also an Oct. 24, 2016 article by Michael Light on Salon.com on the topic (Note: Links have been removed),

In a recent book that was longlisted for the National Book Award, Cathy O’Neil, a data scientist, blogger and former hedge-fund quant, details a number of flawed algorithms to which we have given incredible power — she calls them “Weapons of Math Destruction.” We have entrusted these WMDs to make important, potentially life-altering decisions, yet in many cases, they embed human race and class biases; in other cases, they don’t function at all.
Among other examples, O’Neil examines a “value-added” model New York City used to decide which teachers to fire, even though, she writes, the algorithm was useless, functioning essentially as a random number generator, arbitrarily ending careers. She looks at models put to use by judges to assign recidivism scores to inmates that ended up having a racist inclination. And she looks at how algorithms are contributing to American partisanship, allowing political operatives to target voters with information that plays to their existing biases and fears.

I recommend reading Light’s article in its entirety.