Sunday, 30 April 2017

Alberta is Realizing the Value from Two Long-range Research Investments

In 2000, the Government of Alberta created Alberta Ingenuity, an organization to stimulate research in science and engineering (today it is called Alberta Innovates). One of their first initiatives was to create a program to fund research centers, areas where there was critical mass of world-class expertise in Alberta. In 2002, they announced their first two centers – in glycomics (studying the affects of sugar on the body’s chemistry) and machine learning (turning data into knowledge). These two centers, both housed in the Faculty of Science at the University of Alberta, continue to this day under the names of the Alberta Glycomics Centre (AGC, www.glycomicscentre.ca) and the Alberta Machine Intelligence Institute (Amii, www.amii.ca).

From 2002 to the present – 15 years – the Government of Alberta has continued to invest in both centers. This has happened despite many vicissitudes, including difficult financial times and changes in political direction.  This type of long-term support is unusual in academia. Far more common is one-time term-limited funding. Alberta Ingenuity/Innovates had faith in the quality of the people involved, and in the progress the centers were making towards achieving important new results and then turning this new technology into economic value.

Now in 2017 we can take stock of what has been achieved. AGC researchers have numerous patents, technology licensing agreements, and local spinoff companies. Further, AGC was the catalyst that united all academic glycomics research in Canada into a national organization, GlycoNet. GlycoNet is funded by Ottawa at roughly the $5M per year level, excluding substantial funding from numerous provinces and companies. The University of Alberta leads this National Centre of Excellence and much of the money stays in Alberta. We are proud of the truly world-class stature that this group has.

Amii has been doing groundbreaking research for years. For over 25 years, the University of Alberta has ranked 2nd or 3rd in the world in artificial intelligence (AI) research and its prominent sub-area of machine learning (ML). In the past year AI has suddenly become “hot”. The global demand for AI/ML experts is huge. In their recent budget, Ottawa decided to create a national AI initiative, of which $25M is now coming to the University of Alberta to grow Amii. Further major Canadian and international companies are lined up to partner with Amii. In January the Royal Bank announced they are setting up a research office in Edmonton. Soon you will hear more announcements of major international companies doing the same. The economic impact for Edmonton will be huge, including attracting companies to town, retaining superb scientists in Edmonton, and creating new spinoff companies. The Government wants to diversity the economy. Amii’s world-class stature is going to help achieve that goal.

This is a perfect example of the importance of long-term fundamental research. All too often, we see governments investing in “get payback quick” schemes that look good in the short-term (i.e., in time for the next election) but do not necessarily have long-term impact. Both AGC and Amii started out doing fundamental (“basic”) research, but now we are seeing the benefits of this foundational work. By applying these ideas to solve industrial applications (“applied” research) and creating new products, Albertans will now realize a major economic return from the Government’s investment.

It is important to give credit where credit is due. As academics we sometimes are guilty of taking research funding for granted and not being suitably appreciative of the faith that the funders are putting in our ability to deliver value. On behalf of the Faculty of Science at the University of Alberta, I want to express my deepest heart-felt thank you to the Government of Alberta for their vision, initiative, and stick-with-it-ness – for believing in the AGC and Amii through thick and thin. In my 33 years of academic experience, this kind of investment is the exception. I am delighted that the AGC and Amii research programs are realizing the potential seen in 2002 and delivering major returns to the Alberta taxpayer.

Monday, 24 April 2017

Multitasking Taxi Driver

Last month I was in Hong Kong and had a unique taxi experience. The driver had the vehicle traveling at high speed, darting left and right through busy traffic. But that's not the whole story. Look closely at the picture below. Mounted on the dashboard in front of him are seven (!) cell phones. While the vehicle was in motion, the driver's hands would dart forward (leaving the steering wheel) to push a button or type some text on the phones. As well he had earphones on and was conducting a phone call. Extreme multitasking? Or foolhardy time management?


We arrived safe, sound, and early. Whew!

Monday, 6 March 2017

Canada Must Focus Its AI Vision If It Wants to Lead the World

The following appeared in the Globe and Mail newspaper on March 3, 2017. Although I do not have time to do much research and write academic papers, I seem to be finding time to write editorial and opinion pieces for newspapers.

In McKinsey & Company’s Disruptive Technology, technologies containing artificial intelligence (AI) are expected to create tens of trillions of dollars in economic impact by the year 2025. Similarly, Gartner, Inc. lists machine learning (a subset of AI) as one of the top 10 strategic technology trends for 2017. Yet another technology fad? Perhaps, but it is hard to imagine any industry that won’t be affected by software applications with “smart” capabilities.

For several decades, Canada has been a global powerhouse in artificial intelligence research. The Universities of Alberta and Toronto and the Université de Montréal have world-class AI research centers, and several other universities have strong programs. With all the media attention that AI is suddenly getting, it is critical that Canada leverage this enormous technology asset.

AI technology has been widely deployed for over 20 years – but it is invisible to most people. Whether it is being used for credit card fraud detection (banking), online product recommendations (sales), or in computer games (entertainment), AI is already impacting the world. But the potential of new machine learning technology is so much greater. Soon we will see safe driverless cars, life-saving medical advances, and the Internet of things  – all powered by AI technology. But unless we do something about it, foreign companies will supply most of this technology to Canadians, even though key parts of it will have been invented in Canada.

Whereas AI in Canadian academia is world class, what can we say about the role of AI in Canadian industry? My AI research colleagues and I are challenged to identify even a few Canadian companies that employ ten or more AI specialists, yet it is easy to name numerous international companies that have hundreds of such employees, many of them graduates of our excellent AI programs. What does the rest of the world know that Canadian companies do not?

Montreal-based Maluuba is a rare example of a commercially successful Canadian AI company. It was recently purchased by Microsoft. This is reflective of the current state of our high-tech economy and a warning as to where we are going.

I have graduated 75 Masters and PhD students in my 33 years as a Professor at the University of Alberta. Less than half remain in Canada; few are employed in jobs that take advantage of their AI background. My colleagues have similar statistics. We train the best and the brightest only to watch them leave our country to enrich the rest of the world, or stay in Canada and likely work in a non-AI field of computing.

Now is the time for Canada to think big and act boldly. Canada has few major high-tech success stories to brag about. With AI, we have most of the ingredients for changing the world; now we need the winning recipe:
  • Retain and grow the academic expertise. Non-Canadian companies and universities are increasingly targeting Canada’s AI talent, and with them the most valuable asset – intellectual property (IP).
  • Grow the AI workforce. Our universities can feed ideas and graduates to industry, but Canadian industry has limited capacity to exploit this and innovate. We must create a larger and strategic receptor capacity in industry for AI expertise. With few innovative companies and high-quality job opportunities, our graduates leave.
  • Promote entrepreneurship. Create an environment that encourages faculty and students to turn their research into new companies of products and services. Critical to success in this area is putting in place the local support and early-stage funding necessary to maximize the chances of market success. Make it easier to quickly launch, acquire funding, and build a successful business.
  • Diversify the economy. AI technology will touch almost every aspect of industry. We can develop new products and services and export them to the world, but we need existing Canadian companies to invest in R&D areas that employ our AI talent. By doing so, we can help erase the 20th century stigma of our economy as one of commodity extraction and exportation. Our highly skilled workforce is not a resource to be extracted and then exported to the world. Move our resource-based economy to an idea-based economy
  • Aspire to be the global leader in the area of artificial intelligence. Such bravado may seem un-Canadian. There’s nothing wrong with having a bold vision, as long as the country follows through with a bold implementation that’s oriented towards capturing wealth from the AI expertise we create for Canada.
None of this is hard; it just takes time, a mind-set change, and focused funding to make this happen. There is nothing novel or specific to artificial intelligence in the above recipe. Commercializing IP is how the new wealth is created in the 21st century global economy. Canada needs to be doing that across all sectors and industries.

It is tempting to throw money at this Canadian AI opportunity and hope that everything works out. By all means, start the process of putting the necessary funding in place (through federal, provincial, and industrial initiatives). However, reflection is needed to come up with a vision that maximizes the chances of achieving major economic outcomes for Canada.

History tells us that it a top down political approach is unlikely to succeed. Conversely, a purely academic-led initiative will not produce the commercialization to scale or marketable products and services that will have economic impact. The solution has to be one that creates a continuum between pure research (solutions looking for problems; long-term view) and applied research (problems looking for solutions; short-/medium-term view), and introduces policy strategies that support our best companies who are aiming to scale up globally.

Experience also tells us that a single national initiative is unlikely to be effective (centralized). Conversely, a combination of purely local solutions will lack vision and coordination (decentralized). The solution must be to work together under a national umbrella but have hyperlocal organization.

The end result must be to create, grow, and scale up Canadian companies globally and in the process bring significant private and public wealth to Canada. With innovative ideas and excellent graduates coming from our universities, the opportunities for industry innovation in this area will multiply. We can reset Canada’s future in artificial intelligence technology from extraction to attraction.

Jonathan Schaeffer
Fellow of the Associate for the Advancement of Artificial Intelligence
Dean, Faculty of Science, University of Alberta

Saturday, 14 January 2017

Artificial Intelligence in Canada

On Saturday January 14, Canada's National Newspaper, the Globe and Mail, published an opinion piece arguing that the areas of artificial intelligence (AI) and machine learning (ML) were important technologies today, and that Canada should create a national AI Institute. All well and good, except that the article was Toronto-centric and ignored the reality that there are several outstanding AI/ML groups in Canada. Needless to say, the article generated negative feedback from the major AI groups in Canada, including the University de Montreal, University of British Columbia, and, of course, the University of Alberta.

When I first read the opinion piece, I was livid. I decided to write a rebuttal to the Globe and Mail. After writing a longer piece that expressed what I really thought (for the therapeutic value), I then edited it down into a short letter that was more politically correct. Below is the article that appeared today.

Please look at the graphic that follows (referenced in the article but not printed). It shows that the University of Alberta is proudly ranked third in the world in this area. Note also that the size of our research team is smaller, and in some cases considerably smaller, than our peers.

I am proud that we have been able to build a world class AI/ML research team at the University of Alberta.

Globe and Mail, Saturday January 14.
AI Institute? Think national

Re AI Is The Future, And Canada Must Seize It (Report on Business, Jan. 7):

Canada has a rich history of research into artificial intelligence (AI), going back more than 40 years. Globally, we punch well above our weight. For example, in the areas of artificial intelligence and machine learning, the Computer Science Rankings site places the University of Alberta third, the University of Toronto seventh, and includes three other Canadian universities in the top 50.

Two machine-learning areas that are generating the most excitement today, deep learning (Geoffrey Hinton) and reinforcement learning (Richard Sutton), were pioneered by Canadian academics.

The article’s authors assert that “We must build a world-leading AI Institute in Toronto.” Why Toronto? The authors, who extol the virtues of Toronto to the exclusion of the excellence elsewhere in Canada, call for a “very significant funding commitment” to build the AI Institute. What about the recent federal government investment of $93-million directed to the Université de Montréal for machine-learning research?

As a Torontonian now working at the University of Alberta, I am acutely sensitive to the “Toronto-is-the-centre-of-the-world” syndrome. A Toronto-based AI Institute would be a way to solve a University of Toronto problem that its “machine-learning researchers are spread across many departments in disparate buildings already at capacity.” There is not much that is of national benefit in that.

Jonathan Schaeffer, dean, Faculty of Science, University of Alberta; Fellow of the Association for the Advancement of Artificial Intelligence

Artificial Intelligence / Machine Learning rankings from csrankings.org

Sunday, 18 December 2016

Hacking the U.S. Election


A shorter version of the following article was published in the Edmonton Journal on Saturday November 26, 2016. Several other media outlets also picked up the article.

----------

At the heart of the principle of democracy is the election process. If the integrity of that process is in question, then every citizen who values freedom must be concerned.

On Wednesday [November 23, 2016], New York Magazine came out with a report that “Hillary Clinton is being urged by a group of prominent computer scientists and election lawyers to call for a recount in three swing states won by Donald Trump.” The report stated that, “The group… believes they’ve found persuasive evidence that results in Wisconsin, Michigan, and Pennsylvania may have been manipulated or hacked.” In other words, precisely what Donald Trump repeatedly stated during the run-up to November 8: “This election is rigged.”

As a Canadian, I have no say in who the United States chooses to lead their country. But as someone who looks to the United States as the flag-bearer for democracy around the world, this report raises alarm bells. Nothing has been proven, but I believe the allegations must be taken seriously.

The issue at hand is the credibility of electronic voting. According to New York Magazine, there is data to show that, “in Wisconsin, Clinton received 7 percent fewer votes in counties that relied on electronic-voting machines compared with counties that used optical scanners and paper ballots.” The three states in question represent the margin between having President-elect Trump or President-elect Clinton.

There are two parts to this investigation, both of which hinge on understanding the principles of mathematics and computer science. First, is there data to suggest that something is amiss? Nothing has yet been revealed by the scientists, other than through the news media. Without seeing the data, it is impossible to pass judgement. However, the source is J. Alex Halderman, a Professor of Computer Science and the director of the University of Michigan’s Center for Computer Security and Society. Dr. Halderman has an impressive track record and established credibility.

Here is a simple way of thinking about the claim being made (pardon my mathematical imprecision). Pretend you have two coins and you flip each of them 100 times. Assuming that the coins have no imperfections, you would expect each coin to get around 50 heads and 50 tails. The first coin yields the expected result, 50-50. But flipping the second coin results in 55 heads and 45 tails. This result is possible, but mathematically the chances are small (roughly 1 in 20). What if now you flip the second coin 1000 times and get 550 heads? This is still the same percentage of heads, but now the chances of this occurring drop to around 1 in 6000. With virtual certainty something is amiss.

Loosely speaking, the above analogy describes what is being alleged. Data from one set of counties, those that did not use the voting machines, matches expectations (possibly even matching exit polling results). Data from the set of counties using voting machines does not. We have not yet seen the data analysis from the scientists to support their claims, but if the New York Magazine report is accurate, then there is an anomaly that must be investigated. It is not something that one can write off as an everyday possibility. It is like asking the question “What are my chances of winning a big lottery prize today?” Possible, but unlikely to say the least.

The second part of the investigation is proof. If something untoward has happened, can it be proven? Sadly, this may be difficult to do. The electronic voting machines are run by computer software. They could have been programmed to ignore some votes, something small enough that it might escape attention. That should be easy to prove: just examine the machine’s software. That won’t necessarily reveal anything, as clever hackers often leave no trace of what they have done. The program code would be simple: “Wait until November 8, 2016. On that date ignore every 10th vote for the Democrats. At 6AM on November 9, 2016 erase this code.” Malicious programs do precisely this to cover their tracks.

If the software shows no sign of tampering, can you prove that somehow malicious software was put on the machines? Maybe. It depends on many factors, including the machine, the hardware/software safeguards built into it, whether it was connected to a network, and who might have had access to it. At one extreme, if all the electronic voting machines were connected to the Internet, a hacker might be able to break into the machines and download a “special” version of the software. At the other extreme, none of the machines are connected to the network, and someone had to go to each machine individually and put new software on it. Showing that someone could access the machine without permission would be cause for concern, but still does not prove that the vote counts are wrong.

There is another way that you can prove that the machines were flawed. It may be that some counties that used electronic voting machines have an independent record of the voting. For example, what if it was discovered that these counties averaged 1,000 people casting votes (as recorded at the polling station) but the machine only registered 900 actual votes? If there is independent consistent evidence such as this, it will be compelling proof that the results are flawed. Of course, even this would not tell us who was responsible or who was the legitimate winner of the election.

Where does this leave us? The evidence for voting irregularities must be made public and assessed. If there is cause for concern, then an investigation must be launched. There may be no physical evidence to support the contention that the voting machines were compromised. What do you do, however, if you can prove that the voting pattern recorded by these machines was so unusual that the result had only a one-in-one-thousand chance of occurring? Does that constitute reasonable doubt in the legal sense? Does that meet the bar for casting doubt on the result of the U.S. election?

The U.S. election was a bitter contest, perhaps the most partisan election in that nation’s history. Now a credible source may have data to suggest there are widespread voting anomalies. Every American citizen – Republicans and Democrats – must be very concerned and insist on a thorough bipartisan investigation. Going forward, the United States must put in place a process that allows all methods of vote counting to be audited. As a Canadian, it is important to me that the American election result is above reproach. Anything less than that is an affront to democracy. Canada and the world are watching.