Tag Archives: machine learning

Health technology and the Canadian Broadcasting Corporation’s (CBC) two-tier health system ‘Viewpoint’

There’s a lot of talk and handwringing about Canada’s health care system, which ebbs and flows in almost predictable cycles. Jesse Hirsh in a May 16, 2017 ‘Viewpoints’ segment (an occasional series run as part the of the CBC’s [Canadian Broadcasting Corporation] flagship, daily news programme, The National) dared to reframe the discussion as one about technology and ‘those who get it’  [the technologically literate] and ‘those who don’t’,  a state Hirsh described as being illiterate as you can see and hear in the following video.

I don’t know about you but I’m getting tired of being called illiterate when I don’t know something. To be illiterate means you can’t read and write and as it turns out I do both of those things on a daily basis (sometimes even in two languages). Despite my efforts, I’m ignorant about any number of things and those numbers keep increasing day by day. BTW, Is there anyone who isn’t having trouble keeping up?

Moving on from my rhetorical question, Hirsh has a point about the tech divide and about the need for discussion. It’s a point that hadn’t occurred to me (although I think he’s taking it in the wrong direction). In fact, this business of a tech divide already exists if you consider that people who live in rural environments and need the latest lifesaving techniques or complex procedures or access to highly specialized experts have to travel to urban centres. I gather that Hirsh feels that this divide isn’t necessarily going to be an urban/rural split so much as an issue of how technically literate you and your doctor are.  That’s intriguing but then his argumentation gets muddled. Confusingly, he seems to be suggesting that the key to the split is your access (not your technical literacy) to artificial intelligence (AI) and algorithms (presumably he’s referring to big data and data analytics). I expect access will come down more to money than technological literacy.

For example, money is likely to be a key issue when you consider his big pitch is for access to IBM’s Watson computer. (My Feb. 28, 2011 posting titled: Engineering, entertainment, IBM’s Watson, and product placement focuses largely on Watson, its winning appearances on the US television game show, Jeopardy, and its subsequent adoption into the University of Maryland’s School of Medicine in a project to bring Watson into the examining room with patients.)

Hirsh’s choice of IBM’s Watson is particularly interesting for a number of reasons. (1) Presumably there are companies other than IBM in this sector. Why do they not rate a mention?  (2) Given the current situation with IBM and the Canadian federal government’s introduction of the Phoenix payroll system (a PeopleSoft product customized by IBM), which is  a failure of monumental proportions (a Feb. 23, 2017 article by David Reevely for the Ottawa Citizen and a May 25, 2017 article by Jordan Press for the National Post), there may be a little hesitation, if not downright resistance, to a large scale implementation of any IBM product or service, regardless of where the blame lies. (3) Hirsh notes on the home page for his eponymous website,

I’m presently spending time at the IBM Innovation Space in Toronto Canada, investigating the impact of artificial intelligence and cognitive computing on all sectors and industries.

Yes, it would seem he has some sort of relationship with IBM not referenced in his Viewpoints segment on The National. Also, his description of the relationship isn’t especially illuminating but perhaps it.s this? (from the IBM Innovation Space  – Toronto Incubator Application webpage),

Our incubator

The IBM Innovation Space is a Toronto-based incubator that provides startups with a collaborative space to innovate and disrupt the market. Our goal is to provide you with the tools needed to take your idea to the next level, introduce you to the right networks and help you acquire new clients. Our unique approach, specifically around client engagement, positions your company for optimal growth and revenue at an accelerated pace.

OUR SERVICES

IBM Bluemix
IBM Global Entrepreneur
Softlayer – an IBM Company
Watson

Startups partnered with the IBM Innovation Space can receive up to $120,000 in IBM credits at no charge for up to 12 months through the Global Entrepreneurship Program (GEP). These credits can be used in our products such our IBM Bluemix developer platform, Softlayer cloud services, and our world-renowned IBM Watson ‘cognitive thinking’ APIs. We provide you with enterprise grade technology to meet your clients’ needs, large or small.

Collaborative workspace in the heart of Downtown Toronto
Mentorship opportunities available with leading experts
Access to large clients to scale your startup quickly and effectively
Weekly programming ranging from guest speakers to collaborative activities
Help with funding and access to local VCs and investors​

Final comments

While I have some issues with Hirsh’s presentation, I agree that we should be discussing the issues around increased automation of our health care system. A friend of mine’s husband is a doctor and according to him those prescriptions and orders you get when leaving the hospital? They are not made up by a doctor so much as they are spit up by a computer based on the data that the doctors and nurses have supplied.

GIGO, bias, and de-skilling

Leaving aside the wonders that Hirsh describes, there’s an oldish saying in the computer business, garbage in/garbage out (gigo). At its simplest, who’s going to catch a mistake? (There are lots of mistakes made in hospitals and other health care settings.)

There are also issues around the quality of research. Are all the research papers included in the data used by the algorithms going to be considered equal? There’s more than one case where a piece of problematic research has been accepted uncritically, even if it get through peer review, and subsequently cited many times over. One of the ways to measure impact, i.e., importance, is to track the number of citations. There’s also the matter of where the research is published. A ‘high impact’ journal, such as Nature, Science, or Cell, automatically gives a piece of research a boost.

There are other kinds of bias as well. Increasingly, there’s discussion about algorithms being biased and about how machine learning (AI) can become biased. (See my May 24, 2017 posting: Machine learning programs learn bias, which highlights the issues and cites other FrogHeart posts on that and other related topics.)

These problems are to a large extent already present. Doctors have biases and research can be wrong and it can take a long time before there are corrections. However, the advent of an automated health diagnosis and treatment system is likely to exacerbate the problems. For example, if you don’t agree with your doctor’s diagnosis or treatment, you can search other opinions. What happens when your diagnosis and treatment have become data? Will the system give you another opinion? Who will you talk to? The doctor who got an answer from ‘Watson”? Is she or he going to debate Watson? Are you?

This leads to another issue and that’s automated systems getting more credit than they deserve. Futurists such as Hirsh tend to underestimate people and overestimate the positive impact that automation will have. A computer, data analystics, or an AI system are tools not gods. You’ll have as much luck petitioning one of those tools as you would Zeus.

The unasked question is how will your doctor or other health professional gain experience and skills if they never have to practice the basic, boring aspects of health care (asking questions for a history, reading medical journals to keep up with the research, etc.) and leave them to the computers? There had to be  a reason for calling it a medical ‘practice’.

There are definitely going to be advantages to these technological innovations but thoughtful adoption of these practices (pun intended) should be our goal.

Who owns your data?

Another issue which is increasingly making itself felt is ownership of data. Jacob Brogan has written a provocative May 23, 2017 piece for slate.com asking that question about the data Ancestry.com gathers for DNA testing (Note: Links have been removed),

AncestryDNA’s pitch to consumers is simple enough. For $99 (US), the company will analyze a sample of your saliva and then send back information about your “ethnic mix.” While that promise may be scientifically dubious, it’s a relatively clear-cut proposal. Some, however, worry that the service might raise significant privacy concerns.

After surveying AncestryDNA’s terms and conditions, consumer protection attorney Joel Winston found a few issues that troubled him. As he noted in a Medium post last week, the agreement asserts that it grants the company “a perpetual, royalty-free, world-wide, transferable license to use your DNA.” (The actual clause is considerably longer.) According to Winston, “With this single contractual provision, customers are granting Ancestry.com the broadest possible rights to own and exploit their genetic information.”

Winston also noted a handful of other issues that further complicate the question of ownership. Since we share much of our DNA with our relatives, he warned, “Even if you’ve never used Ancestry.com, but one of your genetic relatives has, the company may already own identifiable portions of your DNA.” [emphasis mine] Theoretically, that means information about your genetic makeup could make its way into the hands of insurers or other interested parties, whether or not you’ve sent the company your spit. (Maryam Zaringhalam explored some related risks in a recent Slate article.) Further, Winston notes that Ancestry’s customers waive their legal rights, meaning that they cannot sue the company if their information gets used against them in some way.

Over the weekend, Eric Heath, Ancestry’s chief privacy officer, responded to these concerns on the company’s own site. He claims that the transferable license is necessary for the company to provide its customers with the service that they’re paying for: “We need that license in order to move your data through our systems, render it around the globe, and to provide you with the results of our analysis work.” In other words, it allows them to send genetic samples to labs (Ancestry uses outside vendors), store the resulting data on servers, and furnish the company’s customers with the results of the study they’ve requested.

Speaking to me over the phone, Heath suggested that this license was akin to the ones that companies such as YouTube employ when users upload original content. It grants them the right to shift that data around and manipulate it in various ways, but isn’t an assertion of ownership. “We have committed to our users that their DNA data is theirs. They own their DNA,” he said.

I’m glad to see the company’s representatives are open to discussion and, later in the article, you’ll see there’ve already been some changes made. Still, there is no guarantee that the situation won’t again change, for ill this time.

What data do they have and what can they do with it?

It’s not everybody who thinks data collection and data analytics constitute problems. While some people might balk at the thought of their genetic data being traded around and possibly used against them, e.g., while hunting for a job, or turned into a source of revenue, there tends to be a more laissez-faire attitude to other types of data. Andrew MacLeod’s May 24, 2017 article for thetyee.ca highlights political implications and privacy issues (Note: Links have been removed),

After a small Victoria [British Columbia, Canada] company played an outsized role in the Brexit vote, government information and privacy watchdogs in British Columbia and Britain have been consulting each other about the use of social media to target voters based on their personal data.

The U.K.’s information commissioner, Elizabeth Denham [Note: Denham was formerly B.C.’s Office of the Information and Privacy Commissioner], announced last week [May 17, 2017] that she is launching an investigation into “the use of data analytics for political purposes.”

The investigation will look at whether political parties or advocacy groups are gathering personal information from Facebook and other social media and using it to target individuals with messages, Denham said.

B.C.’s Office of the Information and Privacy Commissioner confirmed it has been contacted by Denham.

Macleod’s March 6, 2017 article for thetyee.ca provides more details about the company’s role (note: Links have been removed),

The “tiny” and “secretive” British Columbia technology company [AggregateIQ; AIQ] that played a key role in the Brexit referendum was until recently listed as the Canadian office of a much larger firm that has 25 years of experience using behavioural research to shape public opinion around the world.

The larger firm, SCL Group, says it has worked to influence election outcomes in 19 countries. Its associated company in the U.S., Cambridge Analytica, has worked on a wide range of campaigns, including Donald Trump’s presidential bid.

In late February [2017], the Telegraph reported that campaign disclosures showed that Vote Leave campaigners had spent £3.5 million — about C$5.75 million [emphasis mine] — with a company called AggregateIQ, run by CEO Zack Massingham in downtown Victoria.

That was more than the Leave side paid any other company or individual during the campaign and about 40 per cent of its spending ahead of the June referendum that saw Britons narrowly vote to exit the European Union.

According to media reports, Aggregate develops advertising to be used on sites including Facebook, Twitter and YouTube, then targets messages to audiences who are likely to be receptive.

The Telegraph story described Victoria as “provincial” and “picturesque” and AggregateIQ as “secretive” and “low-profile.”

Canadian media also expressed surprise at AggregateIQ’s outsized role in the Brexit vote.

The Globe and Mail’s Paul Waldie wrote “It’s quite a coup for Mr. Massingham, who has only been involved in politics for six years and started AggregateIQ in 2013.”

Victoria Times Colonist columnist Jack Knox wrote “If you have never heard of AIQ, join the club.”

The Victoria company, however, appears to be connected to the much larger SCL Group, which describes itself on its website as “the global leader in data-driven communications.”

In the United States it works through related company Cambridge Analytica and has been involved in elections since 2012. Politico reported in 2015 that the firm was working on Ted Cruz’s presidential primary campaign.

And NBC and other media outlets reported that the Trump campaign paid Cambridge Analytica millions to crunch data on 230 million U.S. adults, using information from loyalty cards, club and gym memberships and charity donations [emphasis mine] to predict how an individual might vote and to shape targeted political messages.

That’s quite a chunk of change and I don’t believe that gym memberships, charity donations, etc. were the only sources of information (in the US, there’s voter registration, credit card information, and more) but the list did raise my eyebrows. It would seem we are under surveillance at all times, even in the gym.

In any event, I hope that Hirsh’s call for discussion is successful and that the discussion includes more critical thinking about the implications of Hirsh’s ‘Brave New World’.

Machine learning programs learn bias

The notion of bias in artificial intelligence (AI)/algorithms/robots is gaining prominence (links to other posts featuring algorithms and bias are at the end of this post). The latest research concerns machine learning where an artificial intelligence system trains itself with ordinary human language from the internet. From an April 13, 2017 American Association for the Advancement of Science (AAAS) news release on EurekAlert,

As artificial intelligence systems “learn” language from existing texts, they exhibit the same biases that humans do, a new study reveals. The results not only provide a tool for studying prejudicial attitudes and behavior in humans, but also emphasize how language is intimately intertwined with historical biases and cultural stereotypes. A common way to measure biases in humans is the Implicit Association Test (IAT), where subjects are asked to pair two concepts they find similar, in contrast to two concepts they find different; their response times can vary greatly, indicating how well they associated one word with another (for example, people are more likely to associate “flowers” with “pleasant,” and “insects” with “unpleasant”). Here, Aylin Caliskan and colleagues developed a similar way to measure biases in AI systems that acquire language from human texts; rather than measuring lag time, however, they used the statistical number of associations between words, analyzing roughly 2.2 million words in total. Their results demonstrate that AI systems retain biases seen in humans. For example, studies of human behavior show that the exact same resume is 50% more likely to result in an opportunity for an interview if the candidate’s name is European American rather than African-American. Indeed, the AI system was more likely to associate European American names with “pleasant” stimuli (e.g. “gift,” or “happy”). In terms of gender, the AI system also reflected human biases, where female words (e.g., “woman” and “girl”) were more associated than male words with the arts, compared to mathematics. In a related Perspective, Anthony G. Greenwald discusses these findings and how they could be used to further analyze biases in the real world.

There are more details about the research in this April 13, 2017 Princeton University news release on EurekAlert (also on ScienceDaily),

In debates over the future of artificial intelligence, many experts think of the new systems as coldly logical and objectively rational. But in a new study, researchers have demonstrated how machines can be reflections of us, their creators, in potentially problematic ways. Common machine learning programs, when trained with ordinary human language available online, can acquire cultural biases embedded in the patterns of wording, the researchers found. These biases range from the morally neutral, like a preference for flowers over insects, to the objectionable views of race and gender.

Identifying and addressing possible bias in machine learning will be critically important as we increasingly turn to computers for processing the natural language humans use to communicate, for instance in doing online text searches, image categorization and automated translations.

“Questions about fairness and bias in machine learning are tremendously important for our society,” said researcher Arvind Narayanan, an assistant professor of computer science and an affiliated faculty member at the Center for Information Technology Policy (CITP) at Princeton University, as well as an affiliate scholar at Stanford Law School’s Center for Internet and Society. “We have a situation where these artificial intelligence systems may be perpetuating historical patterns of bias that we might find socially unacceptable and which we might be trying to move away from.”

The paper, “Semantics derived automatically from language corpora contain human-like biases,” published April 14  [2017] in Science. Its lead author is Aylin Caliskan, a postdoctoral research associate and a CITP fellow at Princeton; Joanna Bryson, a reader at University of Bath, and CITP affiliate, is a coauthor.

As a touchstone for documented human biases, the study turned to the Implicit Association Test, used in numerous social psychology studies since its development at the University of Washington in the late 1990s. The test measures response times (in milliseconds) by human subjects asked to pair word concepts displayed on a computer screen. Response times are far shorter, the Implicit Association Test has repeatedly shown, when subjects are asked to pair two concepts they find similar, versus two concepts they find dissimilar.

Take flower types, like “rose” and “daisy,” and insects like “ant” and “moth.” These words can be paired with pleasant concepts, like “caress” and “love,” or unpleasant notions, like “filth” and “ugly.” People more quickly associate the flower words with pleasant concepts, and the insect terms with unpleasant ideas.

The Princeton team devised an experiment with a program where it essentially functioned like a machine learning version of the Implicit Association Test. Called GloVe, and developed by Stanford University researchers, the popular, open-source program is of the sort that a startup machine learning company might use at the heart of its product. The GloVe algorithm can represent the co-occurrence statistics of words in, say, a 10-word window of text. Words that often appear near one another have a stronger association than those words that seldom do.

The Stanford researchers turned GloVe loose on a huge trawl of contents from the World Wide Web, containing 840 billion words. Within this large sample of written human culture, Narayanan and colleagues then examined sets of so-called target words, like “programmer, engineer, scientist” and “nurse, teacher, librarian” alongside two sets of attribute words, such as “man, male” and “woman, female,” looking for evidence of the kinds of biases humans can unwittingly possess.

In the results, innocent, inoffensive biases, like for flowers over bugs, showed up, but so did examples along lines of gender and race. As it turned out, the Princeton machine learning experiment managed to replicate the broad substantiations of bias found in select Implicit Association Test studies over the years that have relied on live, human subjects.

For instance, the machine learning program associated female names more with familial attribute words, like “parents” and “wedding,” than male names. In turn, male names had stronger associations with career attributes, like “professional” and “salary.” Of course, results such as these are often just objective reflections of the true, unequal distributions of occupation types with respect to gender–like how 77 percent of computer programmers are male, according to the U.S. Bureau of Labor Statistics.

Yet this correctly distinguished bias about occupations can end up having pernicious, sexist effects. An example: when foreign languages are naively processed by machine learning programs, leading to gender-stereotyped sentences. The Turkish language uses a gender-neutral, third person pronoun, “o.” Plugged into the well-known, online translation service Google Translate, however, the Turkish sentences “o bir doktor” and “o bir hem?ire” with this gender-neutral pronoun are translated into English as “he is a doctor” and “she is a nurse.”

“This paper reiterates the important point that machine learning methods are not ‘objective’ or ‘unbiased’ just because they rely on mathematics and algorithms,” said Hanna Wallach, a senior researcher at Microsoft Research New York City, who was not involved in the study. “Rather, as long as they are trained using data from society and as long as society exhibits biases, these methods will likely reproduce these biases.”

Another objectionable example harkens back to a well-known 2004 paper by Marianne Bertrand of the University of Chicago Booth School of Business and Sendhil Mullainathan of Harvard University. The economists sent out close to 5,000 identical resumes to 1,300 job advertisements, changing only the applicants’ names to be either traditionally European American or African American. The former group was 50 percent more likely to be offered an interview than the latter. In an apparent corroboration of this bias, the new Princeton study demonstrated that a set of African American names had more unpleasantness associations than a European American set.

Computer programmers might hope to prevent cultural stereotype perpetuation through the development of explicit, mathematics-based instructions for the machine learning programs underlying AI systems. Not unlike how parents and mentors try to instill concepts of fairness and equality in children and students, coders could endeavor to make machines reflect the better angels of human nature.

“The biases that we studied in the paper are easy to overlook when designers are creating systems,” said Narayanan. “The biases and stereotypes in our society reflected in our language are complex and longstanding. Rather than trying to sanitize or eliminate them, we should treat biases as part of the language and establish an explicit way in machine learning of determining what we consider acceptable and unacceptable.”

Here’s a link to and a citation for the Princeton paper,

Semantics derived automatically from language corpora contain human-like biases by Aylin Caliskan, Joanna J. Bryson, Arvind Narayanan. Science  14 Apr 2017: Vol. 356, Issue 6334, pp. 183-186 DOI: 10.1126/science.aal4230

This paper appears to be open access.

Links to more cautionary posts about AI,

Aug 5, 2009: Autonomous algorithms; intelligent windows; pretty nano pictures

June 14, 2016:  Accountability for artificial intelligence decision-making

Oct. 25, 2016 Removing gender-based stereotypes from algorithms

March 1, 2017: Algorithms in decision-making: a government inquiry in the UK

There’s also a book which makes some of the current use of AI programmes and big data quite accessible reading: Cathy O’Neal’s ‘Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy’.

Evolution of literature as seen by a classicist, a biologist and a computer scientist

Studying intertextuality shows how books are related in various ways and are reorganized and recombined over time. Image courtesy of Elena Poiata.

I find the image more instructive when I read it from the bottom up. For those who prefer to prefer to read from the top down, there’s this April 5, 2017 University of Texas at Austin news release (also on EurekAlert),

A classicist, biologist and computer scientist all walk into a room — what comes next isn’t the punchline but a new method to analyze relationships among ancient Latin and Greek texts, developed in part by researchers from The University of Texas at Austin.

Their work, referred to as quantitative criticism, is highlighted in a study published in the Proceedings of the National Academy of Sciences. The paper identifies subtle literary patterns in order to map relationships between texts and more broadly to trace the cultural evolution of literature.

“As scholars of the humanities well know, literature is a system within which texts bear a multitude of relationships to one another. Understanding what is distinctive about one text entails knowing how it fits within that system,” said Pramit Chaudhuri, associate professor in the Department of Classics at UT Austin. “Our work seeks to harness the power of quantification and computation to describe those relationships at macro and micro levels not easily achieved by conventional reading alone.”

In the study, the researchers create literary profiles based on stylometric features, such as word usage, punctuation and sentence structure, and use techniques from machine learning to understand these complex datasets. Taking a computational approach enables the discovery of small but important characteristics that distinguish one work from another — a process that could require years using manual counting methods.

“One aspect of the technical novelty of our work lies in the unusual types of literary features studied,” Chaudhuri said. “Much computational text analysis focuses on words, but there are many other important hallmarks of style, such as sound, rhythm and syntax.”

Another component of their work builds on Matthew Jockers’ literary “macroanalysis,” which uses machine learning to identify stylistic signatures of particular genres within a large body of English literature. Implementing related approaches, Chaudhuri and his colleagues have begun to trace the evolution of Latin prose style, providing new, quantitative evidence for the sweeping impact of writers such as Caesar and Livy on the subsequent development of Roman prose literature.

“There is a growing appreciation that culture evolves and that language can be studied as a cultural artifact, but there has been less research focused specifically on the cultural evolution of literature,” said the study’s lead author Joseph Dexter, a Ph.D. candidate in systems biology at Harvard University. “Working in the area of classics offers two advantages: the literary tradition is a long and influential one well served by digital resources, and classical scholarship maintains a strong interest in close linguistic study of literature.”

Unusually for a publication in a science journal, the paper contains several examples of the types of more speculative literary reading enabled by the quantitative methods introduced. The authors discuss the poetic use of rhyming sounds for emphasis and of particular vocabulary to evoke mood, among other literary features.

“Computation has long been employed for attribution and dating of literary works, problems that are unambiguous in scope and invite binary or numerical answers,” Dexter said. “The recent explosion of interest in the digital humanities, however, has led to the key insight that similar computational methods can be repurposed to address questions of literary significance and style, which are often more ambiguous and open ended. For our group, this humanist work of criticism is just as important as quantitative methods and data.”

The paper is the work of the Quantitative Criticism Lab (www.qcrit.org), co-directed by Chaudhuri and Dexter in collaboration with researchers from several other institutions. It is funded in part by a 2016 National Endowment for the Humanities grant and the Andrew W. Mellon Foundation New Directions Fellowship, awarded in 2016 to Chaudhuri to further his education in statistics and biology. Chaudhuri was one of 12 scholars selected for the award, which provides humanities researchers the opportunity to train outside of their own area of special interest with a larger goal of bridging the humanities and social sciences.

Here’s another link to the paper along with a citation,

Quantitative criticism of literary relationships by Joseph P. Dexter, Theodore Katz, Nilesh Tripuraneni, Tathagata Dasgupta, Ajay Kannan, James A. Brofos, Jorge A. Bonilla Lopez, Lea A. Schroeder, Adriana Casarez, Maxim Rabinovich, Ayelet Haimson Lushkov, and Pramit Chaudhuri. PNAS Published online before print April 3, 2017, doi: 10.1073/pnas.1611910114

This paper appears to be open access.

Vector Institute and Canada’s artificial intelligence sector

On the heels of the March 22, 2017 federal budget announcement of $125M for a Pan-Canadian Artificial Intelligence Strategy, the University of Toronto (U of T) has announced the inception of the Vector Institute for Artificial Intelligence in a March 28, 2017 news release by Jennifer Robinson (Note: Links have been removed),

A team of globally renowned researchers at the University of Toronto is driving the planning of a new institute staking Toronto’s and Canada’s claim as the global leader in AI.

Geoffrey Hinton, a University Professor Emeritus in computer science at U of T and vice-president engineering fellow at Google, will serve as the chief scientific adviser of the newly created Vector Institute based in downtown Toronto.

“The University of Toronto has long been considered a global leader in artificial intelligence research,” said U of T President Meric Gertler. “It’s wonderful to see that expertise act as an anchor to bring together researchers, government and private sector actors through the Vector Institute, enabling them to aim even higher in leading advancements in this fast-growing, critical field.”

As part of the Government of Canada’s Pan-Canadian Artificial Intelligence Strategy, Vector will share $125 million in federal funding with fellow institutes in Montreal and Edmonton. All three will conduct research and secure talent to cement Canada’s position as a world leader in AI.

In addition, Vector is expected to receive funding from the Province of Ontario and more than 30 top Canadian and global companies eager to tap this pool of talent to grow their businesses. The institute will also work closely with other Ontario universities with AI talent.

(See my March 24, 2017 posting; scroll down about 25% for the science part, including the Pan-Canadian Artificial Intelligence Strategy of the budget.)

Not obvious in last week’s coverage of the Pan-Canadian Artificial Intelligence Strategy is that the much lauded Hinton has been living in the US and working for Google. These latest announcements (Pan-Canadian AI Strategy and Vector Institute) mean that he’s moving back.

A March 28, 2017 article by Kate Allen for TorontoStar.com provides more details about the Vector Institute, Hinton, and the Canadian ‘brain drain’ as it applies to artificial intelligence, (Note:  A link has been removed)

Toronto will host a new institute devoted to artificial intelligence, a major gambit to bolster a field of research pioneered in Canada but consistently drained of talent by major U.S. technology companies like Google, Facebook and Microsoft.

The Vector Institute, an independent non-profit affiliated with the University of Toronto, will hire about 25 new faculty and research scientists. It will be backed by more than $150 million in public and corporate funding in an unusual hybridization of pure research and business-minded commercial goals.

The province will spend $50 million over five years, while the federal government, which announced a $125-million Pan-Canadian Artificial Intelligence Strategy in last week’s budget, is providing at least $40 million, backers say. More than two dozen companies have committed millions more over 10 years, including $5 million each from sponsors including Google, Air Canada, Loblaws, and Canada’s five biggest banks [Bank of Montreal (BMO). Canadian Imperial Bank of Commerce ({CIBC} President’s Choice Financial},  Royal Bank of Canada (RBC), Scotiabank (Tangerine), Toronto-Dominion Bank (TD Canada Trust)].

The mode of artificial intelligence that the Vector Institute will focus on, deep learning, has seen remarkable results in recent years, particularly in image and speech recognition. Geoffrey Hinton, considered the “godfather” of deep learning for the breakthroughs he made while a professor at U of T, has worked for Google since 2013 in California and Toronto.

Hinton will move back to Canada to lead a research team based at the tech giant’s Toronto offices and act as chief scientific adviser of the new institute.

Researchers trained in Canadian artificial intelligence labs fill the ranks of major technology companies, working on tools like instant language translation, facial recognition, and recommendation services. Academic institutions and startups in Toronto, Waterloo, Montreal and Edmonton boast leaders in the field, but other researchers have left for U.S. universities and corporate labs.

The goals of the Vector Institute are to retain, repatriate and attract AI talent, to create more trained experts, and to feed that expertise into existing Canadian companies and startups.

Hospitals are expected to be a major partner, since health care is an intriguing application for AI. Last month, researchers from Stanford University announced they had trained a deep learning algorithm to identify potentially cancerous skin lesions with accuracy comparable to human dermatologists. The Toronto company Deep Genomics is using deep learning to read genomes and identify mutations that may lead to disease, among other things.

Intelligent algorithms can also be applied to tasks that might seem less virtuous, like reading private data to better target advertising. Zemel [Richard Zemel, the institute’s research director and a professor of computer science at U of T] says the centre is creating an ethics working group [emphasis mine] and maintaining ties with organizations that promote fairness and transparency in machine learning. As for privacy concerns, “that’s something we are well aware of. We don’t have a well-formed policy yet but we will fairly soon.”

The institute’s annual funding pales in comparison to the revenues of the American tech giants, which are measured in tens of billions. The risk the institute’s backers are taking is simply creating an even more robust machine learning PhD mill for the U.S.

“They obviously won’t all stay in Canada, but Toronto industry is very keen to get them,” Hinton said. “I think Trump might help there.” Two researchers on Hinton’s new Toronto-based team are Iranian, one of the countries targeted by U.S. President Donald Trump’s travel bans.

Ethics do seem to be a bit of an afterthought. Presumably the Vector Institute’s ‘ethics working group’ won’t include any regular folks. Is there any thought to what the rest of us think about these developments? As there will also be some collaboration with other proposed AI institutes including ones at the University of Montreal (Université de Montréal) and the University of Alberta (Kate McGillivray’s article coming up shortly mentions them), might the ethics group be centered in either Edmonton or Montreal? Interestingly, two Canadians (Timothy Caulfield at the University of Alberta and Eric Racine at Université de Montréa) testified at the US Commission for the Study of Bioethical Issues Feb. 10 – 11, 2014 meeting, the Brain research, ethics, and nanotechnology. Still speculating here but I imagine Caulfield and/or Racine could be persuaded to extend their expertise in ethics and the human brain to AI and its neural networks.

Getting back to the topic at hand the ‘AI sceneCanada’, Allen’s article is worth reading in its entirety if you have the time.

Kate McGillivray’s March 29, 2017 article for the Canadian Broadcasting Corporation’s (CBC) news online provides more details about the Canadian AI situation and the new strategies,

With artificial intelligence set to transform our world, a new institute is putting Toronto to the front of the line to lead the charge.

The Vector Institute for Artificial Intelligence, made possible by funding from the federal government revealed in the 2017 budget, will move into new digs in the MaRS Discovery District by the end of the year.

Vector’s funding comes partially from a $125 million investment announced in last Wednesday’s federal budget to launch a pan-Canadian artificial intelligence strategy, with similar institutes being established in Montreal and Edmonton.

“[A.I.] cuts across pretty well every sector of the economy,” said Dr. Alan Bernstein, CEO and president of the Canadian Institute for Advanced Research, the organization tasked with administering the federal program.

“Silicon Valley and England and other places really jumped on it, so we kind of lost the lead a little bit. I think the Canadian federal government has now realized that,” he said.

Stopping up the brain drain

Critical to the strategy’s success is building a homegrown base of A.I. experts and innovators — a problem in the last decade, despite pioneering work on so-called “Deep Learning” by Canadian scholars such as Yoshua Bengio and Geoffrey Hinton, a former University of Toronto professor who will now serve as Vector’s chief scientific advisor.

With few university faculty positions in Canada and with many innovative companies headquartered elsewhere, it has been tough to keep the few graduates specializing in A.I. in town.

“We were paying to educate people and shipping them south,” explained Ed Clark, chair of the Vector Institute and business advisor to Ontario Premier Kathleen Wynne.

The existence of that “fantastic science” will lean heavily on how much buy-in Vector and Canada’s other two A.I. centres get.

Toronto’s portion of the $125 million is a “great start,” said Bernstein, but taken alone, “it’s not enough money.”

“My estimate of the right amount of money to make a difference is a half a billion or so, and I think we will get there,” he said.

Jessica Murphy’s March 29, 2017 article for the British Broadcasting Corporation’s (BBC) news online offers some intriguing detail about the Canadian AI scene,

Canadian researchers have been behind some recent major breakthroughs in artificial intelligence. Now, the country is betting on becoming a big player in one of the hottest fields in technology, with help from the likes of Google and RBC [Royal Bank of Canada].

In an unassuming building on the University of Toronto’s downtown campus, Geoff Hinton laboured for years on the “lunatic fringe” of academia and artificial intelligence, pursuing research in an area of AI called neural networks.

Also known as “deep learning”, neural networks are computer programs that learn in similar way to human brains. The field showed early promise in the 1980s, but the tech sector turned its attention to other AI methods after that promise seemed slow to develop.

“The approaches that I thought were silly were in the ascendancy and the approach that I thought was the right approach was regarded as silly,” says the British-born [emphasis mine] professor, who splits his time between the university and Google, where he is a vice-president of engineering fellow.

Neural networks are used by the likes of Netflix to recommend what you should binge watch and smartphones with voice assistance tools. Google DeepMind’s AlphaGo AI used them to win against a human in the ancient game of Go in 2016.

Foteini Agrafioti, who heads up the new RBC Research in Machine Learning lab at the University of Toronto, said those recent innovations made AI attractive to researchers and the tech industry.

“Anything that’s powering Google’s engines right now is powered by deep learning,” she says.

Developments in the field helped jumpstart innovation and paved the way for the technology’s commercialisation. They also captured the attention of Google, IBM and Microsoft, and kicked off a hiring race in the field.

The renewed focus on neural networks has boosted the careers of early Canadian AI machine learning pioneers like Hinton, the University of Montreal’s Yoshua Bengio, and University of Alberta’s Richard Sutton.

Money from big tech is coming north, along with investments by domestic corporations like banking multinational RBC and auto parts giant Magna, and millions of dollars in government funding.

Former banking executive Ed Clark will head the institute, and says the goal is to make Toronto, which has the largest concentration of AI-related industries in Canada, one of the top five places in the world for AI innovation and business.

The founders also want it to serve as a magnet and retention tool for top talent aggressively head-hunted by US firms.

Clark says they want to “wake up” Canadian industry to the possibilities of AI, which is expected to have a massive impact on fields like healthcare, banking, manufacturing and transportation.

Google invested C$4.5m (US$3.4m/£2.7m) last November [2016] in the University of Montreal’s Montreal Institute for Learning Algorithms.

Microsoft is funding a Montreal startup, Element AI. The Seattle-based company also announced it would acquire Montreal-based Maluuba and help fund AI research at the University of Montreal and McGill University.

Thomson Reuters and General Motors both recently moved AI labs to Toronto.

RBC is also investing in the future of AI in Canada, including opening a machine learning lab headed by Agrafioti, co-funding a program to bring global AI talent and entrepreneurs to Toronto, and collaborating with Sutton and the University of Alberta’s Machine Intelligence Institute.

Canadian tech also sees the travel uncertainty created by the Trump administration in the US as making Canada more attractive to foreign talent. (One of Clark’s the selling points is that Toronto as an “open and diverse” city).

This may reverse the ‘brain drain’ but it appears Canada’s role as a ‘branch plant economy’ for foreign (usually US) companies could become an important discussion once more. From the ‘Foreign ownership of companies of Canada’ Wikipedia entry (Note: Links have been removed),

Historically, foreign ownership was a political issue in Canada in the late 1960s and early 1970s, when it was believed by some that U.S. investment had reached new heights (though its levels had actually remained stable for decades), and then in the 1980s, during debates over the Free Trade Agreement.

But the situation has changed, since in the interim period Canada itself became a major investor and owner of foreign corporations. Since the 1980s, Canada’s levels of investment and ownership in foreign companies have been larger than foreign investment and ownership in Canada. In some smaller countries, such as Montenegro, Canadian investment is sizable enough to make up a major portion of the economy. In Northern Ireland, for example, Canada is the largest foreign investor. By becoming foreign owners themselves, Canadians have become far less politically concerned about investment within Canada.

Of note is that Canada’s largest companies by value, and largest employers, tend to be foreign-owned in a way that is more typical of a developing nation than a G8 member. The best example is the automotive sector, one of Canada’s most important industries. It is dominated by American, German, and Japanese giants. Although this situation is not unique to Canada in the global context, it is unique among G-8 nations, and many other relatively small nations also have national automotive companies.

It’s interesting to note that sometimes Canadian companies are the big investors but that doesn’t change our basic position. And, as I’ve noted in other postings (including the March 24, 2017 posting), these government investments in science and technology won’t necessarily lead to a move away from our ‘branch plant economy’ towards an innovative Canada.

You can find out more about the Vector Institute for Artificial Intelligence here.

BTW, I noted that reference to Hinton as ‘British-born’ in the BBC article. He was educated in the UK and subsidized by UK taxpayers (from his Wikipedia entry; Note: Links have been removed),

Hinton was educated at King’s College, Cambridge graduating in 1970, with a Bachelor of Arts in experimental psychology.[1] He continued his study at the University of Edinburgh where he was awarded a PhD in artificial intelligence in 1977 for research supervised by H. Christopher Longuet-Higgins.[3][12]

It seems Canadians are not the only ones to experience  ‘brain drains’.

Finally, I wrote at length about a recent initiative taking place between the University of British Columbia (Vancouver, Canada) and the University of Washington (Seattle, Washington), the Cascadia Urban Analytics Cooperative in a Feb. 28, 2017 posting noting that the initiative is being funded by Microsoft to the tune $1M and is part of a larger cooperative effort between the province of British Columbia and the state of Washington. Artificial intelligence is not the only area where US technology companies are hedging their bets (against Trump’s administration which seems determined to terrify people from crossing US borders) by investing in Canada.

For anyone interested in a little more information about AI in the US and China, there’s today’s (March 31, 2017)earlier posting: China, US, and the race for artificial intelligence research domination.

Ishiguro’s robots and Swiss scientist question artificial intelligence at SXSW (South by Southwest) 2017

It seems unexpected to stumble across presentations on robots and on artificial intelligence at an entertainment conference such as South by South West (SXSW). Here’s why I thought so, from the SXSW Wikipedia entry (Note: Links have been removed),

South by Southwest (abbreviated as SXSW) is an annual conglomerate of film, interactive media, and music festivals and conferences that take place in mid-March in Austin, Texas, United States. It began in 1987, and has continued to grow in both scope and size every year. In 2011, the conference lasted for 10 days with SXSW Interactive lasting for 5 days, Music for 6 days, and Film running concurrently for 9 days.

Lifelike robots

The 2017 SXSW Interactive featured separate presentations by Japanese roboticist, Hiroshi Ishiguro (mentioned here a few times), and EPFL (École Polytechnique Fédérale de Lausanne; Switzerland) artificial intelligence expert, Marcel Salathé.

Ishiguro’s work is the subject of Harry McCracken’s March 14, 2017 article for Fast Company (Note: Links have been removed),

I’m sitting in the Japan Factory pavilion at SXSW in Austin, Texas, talking to two other attendees about whether human beings are more valuable than robots. I say that I believe human life to be uniquely precious, whereupon one of the others rebuts me by stating that humans allow cars to exist even though they kill humans.

It’s a reasonable point. But my fellow conventioneer has a bias: It’s a robot itself, with an ivory-colored, mask-like face and visible innards. So is the third participant in the conversation, a much more human automaton modeled on a Japanese woman and wearing a black-and-white blouse and a blue scarf.

We’re chatting as part of a demo of technologies developed by the robotics lab of Hiroshi Ishiguro, based at Osaka University, and Japanese telecommunications company NTT. Ishiguro has gained fame in the field by creating increasingly humanlike robots—that is, androids—with the ultimate goal of eliminating the uncanny valley that exists between people and robotic people.

I also caught up with Ishiguro himself at the conference—his second SXSW—to talk about his work. He’s a champion of the notion that people will respond best to robots who simulate humanity, thereby creating “a feeling of presence,” as he describes it. That gives him and his researchers a challenge that encompasses everything from technology to psychology. “Our approach is quite interdisciplinary,” he says, which is what prompted him to bring his work to SXSW.

A SXSW attendee talks about robots with two robots.

If you have the time, do read McCracken’t piece in its entirety.

You can find out more about the ‘uncanny valley’ in my March 10, 2011 posting about Ishiguro’s work if you scroll down about 70% of the way to find the ‘uncanny valley’ diagram and Masahiro Mori’s description of the concept he developed.

You can read more about Ishiguro and his colleague, Ryuichiro Higashinaka, on their SXSW biography page.

Artificial intelligence (AI)

In a March 15, 2017 EPFL press release by Hilary Sanctuary, scientist Marcel Salathé poses the question: Is Reliable Artificial Intelligence Possible?,

In the quest for reliable artificial intelligence, EPFL scientist Marcel Salathé argues that AI technology should be openly available. He will be discussing the topic at this year’s edition of South by South West on March 14th in Austin, Texas.

Will artificial intelligence (AI) change the nature of work? For EPFL theoretical biologist Marcel Salathé, the answer is invariably yes. To him, a more fundamental question that needs to be addressed is who owns that artificial intelligence?

“We have to hold AI accountable, and the only way to do this is to verify it for biases and make sure there is no deliberate misinformation,” says Salathé. “This is not possible if the AI is privatized.”

AI is both the algorithm and the data

So what exactly is AI? It is generally regarded as “intelligence exhibited by machines”. Today, it is highly task specific, specially designed to beat humans at strategic games like Chess and Go, or diagnose skin disease on par with doctors’ skills.

On a practical level, AI is implemented through what scientists call “machine learning”, which means using a computer to run specifically designed software that can be “trained”, i.e. process data with the help of algorithms and to correctly identify certain features from that data set. Like human cognition, AI learns by trial and error. Unlike humans, however, AI can process and recall large quantities of data, giving it a tremendous advantage over us.

Crucial to AI learning, therefore, is the underlying data. For Salathé, AI is defined by both the algorithm and the data, and as such, both should be publicly available.

Deep learning algorithms can be perturbed

Last year, Salathé created an algorithm to recognize plant diseases. With more than 50,000 photos of healthy and diseased plants in the database, the algorithm uses artificial intelligence to diagnose plant diseases with the help of your smartphone. As for human disease, a recent study by a Stanford Group on cancer showed that AI can be trained to recognize skin cancer slightly better than a group of doctors. The consequences are far-reaching: AI may one day diagnose our diseases instead of doctors. If so, will we really be able to trust its diagnosis?

These diagnostic tools use data sets of images to train and learn. But visual data sets can be perturbed that prevent deep learning algorithms from correctly classifying images. Deep neural networks are highly vulnerable to visual perturbations that are practically impossible to detect with the naked eye, yet causing the AI to misclassify images.

In future implementations of AI-assisted medical diagnostic tools, these perturbations pose a serious threat. More generally, the perturbations are real and may already be affecting the filtered information that reaches us every day. These vulnerabilities underscore the importance of certifying AI technology and monitoring its reliability.

h/t phys.org March 15, 2017 news item

As I noted earlier, these are not the kind of presentations you’d expect at an ‘entertainment’ festival.

High-performance, low-energy artificial synapse for neural network computing

This artificial synapse is apparently an improvement on the standard memristor-based artificial synapse but that doesn’t become clear until reading the abstract for the paper. First, there’s a Feb. 20, 2017 Stanford University news release by Taylor Kubota (dated Feb. 21, 2017 on EurekAlert), Note: Links have been removed,

For all the improvements in computer technology over the years, we still struggle to recreate the low-energy, elegant processing of the human brain. Now, researchers at Stanford University and Sandia National Laboratories have made an advance that could help computers mimic one piece of the brain’s efficient design – an artificial version of the space over which neurons communicate, called a synapse.

“It works like a real synapse but it’s an organic electronic device that can be engineered,” said Alberto Salleo, associate professor of materials science and engineering at Stanford and senior author of the paper. “It’s an entirely new family of devices because this type of architecture has not been shown before. For many key metrics, it also performs better than anything that’s been done before with inorganics.”

The new artificial synapse, reported in the Feb. 20 issue of Nature Materials, mimics the way synapses in the brain learn through the signals that cross them. This is a significant energy savings over traditional computing, which involves separately processing information and then storing it into memory. Here, the processing creates the memory.

This synapse may one day be part of a more brain-like computer, which could be especially beneficial for computing that works with visual and auditory signals. Examples of this are seen in voice-controlled interfaces and driverless cars. Past efforts in this field have produced high-performance neural networks supported by artificially intelligent algorithms but these are still distant imitators of the brain that depend on energy-consuming traditional computer hardware.

Building a brain

When we learn, electrical signals are sent between neurons in our brain. The most energy is needed the first time a synapse is traversed. Every time afterward, the connection requires less energy. This is how synapses efficiently facilitate both learning something new and remembering what we’ve learned. The artificial synapse, unlike most other versions of brain-like computing, also fulfills these two tasks simultaneously, and does so with substantial energy savings.

“Deep learning algorithms are very powerful but they rely on processors to calculate and simulate the electrical states and store them somewhere else, which is inefficient in terms of energy and time,” said Yoeri van de Burgt, former postdoctoral scholar in the Salleo lab and lead author of the paper. “Instead of simulating a neural network, our work is trying to make a neural network.”

The artificial synapse is based off a battery design. It consists of two thin, flexible films with three terminals, connected by an electrolyte of salty water. The device works as a transistor, with one of the terminals controlling the flow of electricity between the other two.

Like a neural path in a brain being reinforced through learning, the researchers program the artificial synapse by discharging and recharging it repeatedly. Through this training, they have been able to predict within 1 percent of uncertainly what voltage will be required to get the synapse to a specific electrical state and, once there, it remains at that state. In other words, unlike a common computer, where you save your work to the hard drive before you turn it off, the artificial synapse can recall its programming without any additional actions or parts.

Testing a network of artificial synapses

Only one artificial synapse has been produced but researchers at Sandia used 15,000 measurements from experiments on that synapse to simulate how an array of them would work in a neural network. They tested the simulated network’s ability to recognize handwriting of digits 0 through 9. Tested on three datasets, the simulated array was able to identify the handwritten digits with an accuracy between 93 to 97 percent.

Although this task would be relatively simple for a person, traditional computers have a difficult time interpreting visual and auditory signals.

“More and more, the kinds of tasks that we expect our computing devices to do require computing that mimics the brain because using traditional computing to perform these tasks is becoming really power hungry,” said A. Alec Talin, distinguished member of technical staff at Sandia National Laboratories in Livermore, California, and senior author of the paper. “We’ve demonstrated a device that’s ideal for running these type of algorithms and that consumes a lot less power.”

This device is extremely well suited for the kind of signal identification and classification that traditional computers struggle to perform. Whereas digital transistors can be in only two states, such as 0 and 1, the researchers successfully programmed 500 states in the artificial synapse, which is useful for neuron-type computation models. In switching from one state to another they used about one-tenth as much energy as a state-of-the-art computing system needs in order to move data from the processing unit to the memory.

This, however, means they are still using about 10,000 times as much energy as the minimum a biological synapse needs in order to fire. The researchers are hopeful that they can attain neuron-level energy efficiency once they test the artificial synapse in smaller devices.

Organic potential

Every part of the device is made of inexpensive organic materials. These aren’t found in nature but they are largely composed of hydrogen and carbon and are compatible with the brain’s chemistry. Cells have been grown on these materials and they have even been used to make artificial pumps for neural transmitters. The voltages applied to train the artificial synapse are also the same as those that move through human neurons.

All this means it’s possible that the artificial synapse could communicate with live neurons, leading to improved brain-machine interfaces. The softness and flexibility of the device also lends itself to being used in biological environments. Before any applications to biology, however, the team plans to build an actual array of artificial synapses for further research and testing.

Additional Stanford co-authors of this work include co-lead author Ewout Lubberman, also of the University of Groningen in the Netherlands, Scott T. Keene and Grégorio C. Faria, also of Universidade de São Paulo, in Brazil. Sandia National Laboratories co-authors include Elliot J. Fuller and Sapan Agarwal in Livermore and Matthew J. Marinella in Albuquerque, New Mexico. Salleo is an affiliate of the Stanford Precourt Institute for Energy and the Stanford Neurosciences Institute. Van de Burgt is now an assistant professor in microsystems and an affiliate of the Institute for Complex Molecular Studies (ICMS) at Eindhoven University of Technology in the Netherlands.

This research was funded by the National Science Foundation, the Keck Faculty Scholar Funds, the Neurofab at Stanford, the Stanford Graduate Fellowship, Sandia’s Laboratory-Directed Research and Development Program, the U.S. Department of Energy, the Holland Scholarship, the University of Groningen Scholarship for Excellent Students, the Hendrik Muller National Fund, the Schuurman Schimmel-van Outeren Foundation, the Foundation of Renswoude (The Hague and Delft), the Marco Polo Fund, the Instituto Nacional de Ciência e Tecnologia/Instituto Nacional de Eletrônica Orgânica in Brazil, the Fundação de Amparo à Pesquisa do Estado de São Paulo and the Brazilian National Council.

Here’s an abstract for the researchers’ paper (link to paper provided after abstract) and it’s where you’ll find the memristor connection explained,

The brain is capable of massively parallel information processing while consuming only ~1–100fJ per synaptic event1, 2. Inspired by the efficiency of the brain, CMOS-based neural architectures3 and memristors4, 5 are being developed for pattern recognition and machine learning. However, the volatility, design complexity and high supply voltages for CMOS architectures, and the stochastic and energy-costly switching of memristors complicate the path to achieve the interconnectivity, information density, and energy efficiency of the brain using either approach. Here we describe an electrochemical neuromorphic organic device (ENODe) operating with a fundamentally different mechanism from existing memristors. ENODe switches at low voltage and energy (<10pJ for 103μm2 devices), displays >500 distinct, non-volatile conductance states within a ~1V range, and achieves high classification accuracy when implemented in neural network simulations. Plastic ENODes are also fabricated on flexible substrates enabling the integration of neuromorphic functionality in stretchable electronic systems6, 7. Mechanical flexibility makes ENODes compatible with three-dimensional architectures, opening a path towards extreme interconnectivity comparable to the human brain.

Here’s a link to and a citation for the paper,

A non-volatile organic electrochemical device as a low-voltage artificial synapse for neuromorphic computing by Yoeri van de Burgt, Ewout Lubberman, Elliot J. Fuller, Scott T. Keene, Grégorio C. Faria, Sapan Agarwal, Matthew J. Marinella, A. Alec Talin, & Alberto Salleo. Nature Materials (2017) doi:10.1038/nmat4856 Published online 20 February 2017

This paper is behind a paywall.

ETA March 8, 2017 10:28 PST: You may find this this piece on ferroelectricity and neuromorphic engineering of interest (March 7, 2017 posting titled: Ferroelectric roadmap to neuromorphic computing).

New principles for AI (artificial intelligence) research along with some history and a plea for a democratic discussion

For almost a month I’ve been meaning to get to this Feb. 1, 2017 essay by Andrew Maynard (director of Risk Innovation Lab at Arizona State University) and Jack Stilgoe (science policy lecturer at University College London [UCL]) on the topic of artificial intelligence and principles (Note: Links have been removed). First, a walk down memory lane,

Today [Feb. 1, 2017] in Washington DC, leading US and UK scientists are meeting to share dispatches from the frontiers of machine learning – an area of research that is creating new breakthroughs in artificial intelligence (AI). Their meeting follows the publication of a set of principles for beneficial AI that emerged from a conference earlier this year at a place with an important history.

In February 1975, 140 people – mostly scientists, with a few assorted lawyers, journalists and others – gathered at a conference centre on the California coast. A magazine article from the time by Michael Rogers, one of the few journalists allowed in, reported that most of the four days’ discussion was about the scientific possibilities of genetic modification. Two years earlier, scientists had begun using recombinant DNA to genetically modify viruses. The Promethean nature of this new tool prompted scientists to impose a moratorium on such experiments until they had worked out the risks. By the time of the Asilomar conference, the pent-up excitement was ready to burst. It was only towards the end of the conference when a lawyer stood up to raise the possibility of a multimillion-dollar lawsuit that the scientists focussed on the task at hand – creating a set of principles to govern their experiments.

The 1975 Asilomar meeting is still held up as a beacon of scientific responsibility. However, the story told by Rogers, and subsequently by historians, is of scientists motivated by a desire to head-off top down regulation with a promise of self-governance. Geneticist Stanley Cohen said at the time, ‘If the collected wisdom of this group doesn’t result in recommendations, the recommendations may come from other groups less well qualified’. The mayor of Cambridge, Massachusetts was a prominent critic of the biotechnology experiments then taking place in his city. He said, ‘I don’t think these scientists are thinking about mankind at all. I think that they’re getting the thrills and the excitement and the passion to dig in and keep digging to see what the hell they can do’.

The concern in 1975 was with safety and containment in research, not with the futures that biotechnology might bring about. A year after Asilomar, Cohen’s colleague Herbert Boyer founded Genentech, one of the first biotechnology companies. Corporate interests barely figured in the conversations of the mainly university scientists.

Fast-forward 42 years and it is clear that machine learning, natural language processing and other technologies that come under the AI umbrella are becoming big business. The cast list of the 2017 Asilomar meeting included corporate wunderkinds from Google, Facebook and Tesla as well as researchers, philosophers, and other academics. The group was more intellectually diverse than their 1975 equivalents, but there were some notable absences – no public and their concerns, no journalists, and few experts in the responsible development of new technologies.

Maynard and Stilgoe offer a critique of the latest principles,

The principles that came out of the meeting are, at least at first glance, a comforting affirmation that AI should be ‘for the people’, and not to be developed in ways that could cause harm. They promote the idea of beneficial and secure AI, development for the common good, and the importance of upholding human values and shared prosperity.

This is good stuff. But it’s all rather Motherhood and Apple Pie: comforting and hard to argue against, but lacking substance. The principles are short on accountability, and there are notable absences, including the need to engage with a broader set of stakeholders and the public. At the early stages of developing new technologies, public concerns are often seen as an inconvenience. In a world in which populism appears to be trampling expertise into the dirt, it is easy to understand why scientists may be defensive.

I encourage you to read this thoughtful essay in its entirety although I do have one nit to pick:  Why only US and UK scientists? I imagine the answer may lie in funding and logistics issues but I find it surprising that the critique makes no mention of the international community as a nod to inclusion.

For anyone interested in the Asolimar AI principles (2017), you can find them here. You can also find videos of the two-day workshop (Jan. 31 – Feb. 1, 2017 workshop titled The Frontiers of Machine Learning (a Raymond and Beverly Sackler USA-UK Scientific Forum [US National Academy of Sciences]) here (videos for each session are available on Youtube).

Artificial intelligence and industrial applications

This is take on artificial intelligence that I haven’t encountered before. Sean Captain’s Nov. 15, 2016 article for Fast Company profiles industry giant GE (General Electric) and its foray into that world (Note: Links have been removed),

When you hear the term “artificial intelligence,” you may think of tech giants Amazon, Google, IBM, Microsoft, or Facebook. Industrial powerhouse General Electric is now aiming to be included on that short list. It may not have a chipper digital assistant like Cortana or Alexa. It won’t sort through selfies, but it will look through X-rays. It won’t recommend movies, but it will suggest how to care for a diesel locomotive. Today, GE announced a pair of acquisitions and new services that will bring machine learning AI to the kinds of products it’s known for, including planes, trains, X-ray machines, and power plants.

The effort started in 2015 when GE announced Predix Cloud—an online platform to network and collect data from sensors on industrial machinery such as gas turbines or windmills. At the time, GE touted the benefits of using machine learning to find patterns in sensor data that could lead to energy savings or preventative maintenance before a breakdown. Predix Cloud opened up to customers in February [2016?], but GE is still building up the AI capabilities to fulfill the promise. “We were using machine learning, but I would call it in a custom way,” says Bill Ruh, GE’s chief digital officer and CEO of its GE Digital business (GE calls its division heads CEOs). “And we hadn’t gotten to a general-purpose framework in machine learning.”

Today [Nov. 15, 2016] GE revealed the purchase of two AI companies that Ruh says will get them there. Bit Stew Systems, founded in 2005, was already doing much of what Predix Cloud promises—collecting and analyzing sensor data from power utilities, oil and gas companies, aviation, and factories. (GE Ventures has funded the company.) Customers include BC Hydro, Pacific Gas & Electric, and Scottish & Southern Energy.

The second purchase, Wise.io is a less obvious purchase. Founded by astrophysics and AI experts using machine learning to study the heavens, the company reapplied the tech to streamlining a company’s customer support systems, picking up clients like Pinterest, Twilio, and TaskRabbit. GE believes the technology will transfer yet again, to managing industrial machines. “I think by the middle of next year we will have a full machine learning stack,” says Ruh.

Though young, Predix is growing fast, with 270 partner companies using the platform, according to GE, which expects revenue on software and services to grow over 25% this year, to more than $7 billion. Ruh calls Predix a “significant part” of that extra money. And he’s ready to brag, taking a jab at IBM Watson for being a “general-purpose” machine-learning provider without the deep knowledge of the industries it serves. “We have domain algorithms, on machine learning, that’ll know what a power plant is and all the depth of that, that a general-purpose machine learning will never really understand,” he says.

One especially dull-sounding new Predix service—Predictive Corrosion Management—touches on a very hot political issue: giant oil and gas pipeline projects. Over 400 people have been arrested in months of protests against the Dakota Access Pipeline, which would carry crude oil from North Dakota to Illinois. The issue is very complicated, but one concern of protestors is that a pipeline rupture would contaminate drinking water for the Standing Rock Sioux reservation.

“I think absolutely this is aimed at that problem. If you look at why pipelines spill, it’s corrosion,” says Ruh. “We believe that 10 years from now, we can detect a leak before it occurs and fix it before you see it happen.” Given how political battles over pipelines drag on, 10 years might not be so long to wait.

I recommend reading the article in its entirety if you have the time. And, for those of us in British Columbia, Canada, it was a surprise to see BC Hydro on the list of customers for one of GE’s new acquisitions. As well, that business about the pipelines hits home hard given the current debates (Enbridge Northern Gateway Pipelines) here. *ETA Dec. 27, 2016: This was originally edited just prior to publication to include information about the announcement by the Trudeau cabinet approving two pipelines for TransMountain  and Enbridge respectively while rejecting the Northern Gateway pipeline (Canadian Broadcasting Corporation [CBC] online news Nov. 29, 2016).  I trust this second edit will stick.*

It seems GE is splashing out in a big way. There’s a second piece on Fast Company, a Nov. 16, 2016 article by Sean Captain (again) this time featuring a chat between an engineer and a robotic power plant,

We are entering the era of talking machines—and it’s about more than just asking Amazon’s Alexa to turn down the music. General Electric has built a digital assistant into its cloud service for managing power plants, jet engines, locomotives, and the other heavy equipment it builds. Over the internet, an engineer can ask a machine—even one hundreds of miles away—how it’s doing and what it needs. …

Voice controls are built on top of GE’s Digital Twin program, which uses sensor readings from machinery to create virtual models in cyberspace. “That model is constantly getting a stream of data, both operational and environmental,” says Colin Parris, VP at GE Software Research. “So it’s adapting itself to that type of data.” The machines live virtual lives online, allowing engineers to see how efficiently each is running and if they are wearing down.

GE partnered with Microsoft on the interface, using the Bing Speech API (the same tech powering the Cortana digital assistant), with special training on key terms like “rotor.” The twin had little trouble understanding the Mandarin Chinese accent of Bo Yu, one of the researchers who built the system; nor did it stumble on Parris’s Trinidad accent. Digital Twin will also work with Microsoft’s HoloLens mixed reality goggles, allowing someone to step into a 3D image of the equipment.

I can’t help wondering if there are some jobs that were eliminated with this technology.

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.