Proceedings of the Standing Senate Committee on
Social Affairs, Science and Technology
Issue No. 22 - Evidence - May 3, 2017
OTTAWA, Wednesday, May 3, 2017
The Standing Senate Committee on Social Affairs, Science and Technology met this day at 4:15 p.m. to continue its study on the role of robotics, 3D printing and artificial intelligence in the healthcare system.
Senator Kelvin Kenneth Ogilvie (Chair) in the chair.
[Translation]
The Chair: Welcome to the Standing Senate Committee on Social Affairs, Science and Technology.
[English]
I'm Kelvin Ogilvie from Nova Scotia, chair of the committee. I'm going to start by inviting my colleagues to introduce themselves, starting on my left.
Senator Eggleton: Art Eggleton, senator from Toronto and deputy chair of the committee.
Senator Dean: Tony Dean, senator from Ontario.
[Translation]
Senator Petitclerc: Chantal Petitclerc from Quebec.
[English]
Senator Unger: Betty Unger from Alberta, Edmonton.
Senator Stewart Olsen: Carolyn Stewart Olsen, New Brunswick.
Senator Seidman: Judith Seidman, from Montreal, Quebec.
The Chair: We continue our study on the role of robotics, 3-D printing and artificial intelligence in the health care system. We are very pleased to have two witnesses today. I invite Dr. Bengio, who is Director and Professor at the University of Montreal with the Montreal Institute for Learning Algorithms, to present.
So Dr. Bengio, please proceed.
Yoshua Bengio, Director, Professor, University of Montreal, Montreal Institute for Learning Algorithms: Thank you. It's a pleasure to be here. The Université de Montréal has been a pioneer in artificial intelligence, in particular in the area called deep learning, which I have seen you have heard about in previous meetings. In the last 10 years, since this breakthrough in deep learning happened, the lab grew a lot, and we now have the largest concentration of deep learning researchers and students in academia across the world.
Let me say a few words about the progress that has happened in AI. Computers are now much better able to understand at least some aspects of the world, and take good decisions that will help us in many different ways, but that's probably just a beginning of approaching human level intelligence. We don't know how far this is in the future, but even with the progress that's been achieved up to now scientifically, there will be major economic and social impact of these advances in the next decade.
For example, there has been a breakthrough in the last few years using deep learning to allow computers to recognize the content of images, to understand speech — that is to translate the sounds to words, that doesn't mean that they understand what they mean — to translate from one language to another, to control robots, to control cars, to plan sophisticated sequences of actions, for example playing complex games like the game of Go, and beating the world champion last year.
It's the same underlying science, the same technology, which is behind all of these advances and more that I won't have time to tell you. It's fairly horizontal and general purpose, which is why it can have such a big impact on the economy in general and many different kinds of applications.
What's interesting about this for Canada is that Canada's been a leader in the science of AI, in particular deep learning, with a lot of the progress having happened in Toronto and Montreal and to some extent in Edmonton in the area of reinforcement learning especially.
These Canadian academics have been really leading the world in this area, but unfortunately most of the business investment in this area has been elsewhere, mostly in the U.S.A. and to some extent maybe even more now in China, and to some extent in the U.K.
A lot of Canadian academics and former graduates have moved out of the country, so we have had this brain drain for a number of years.
However, the tide is turning. The Canadian and provincial governments have realized the strategic importance of AI and are starting to invest both for continuing to advance the fundamental research and to promote innovation connecting universities and companies in Canada.
For example, in Montreal, if we add up the different grants and promises that governments and companies have made over the last year for the next five years, and with what we are expecting with high probability in the coming year, we're talking about more than half a billion dollars for just Montreal. And Toronto is about the same.
Something is really changing in the Canadian scene, both in universities and in the makeup of the ecosystems of small and large companies that are investing in AI.
For example, in Montreal in the last six months, there have been announcements from major multinationals to create research labs in deep learning. So it started with Google, then Microsoft, now just last week Facebook, Huawei and IBM and others.
This is important to note because the economic impact of these advances is expected by economists and scientists to be major, probably comparable or even more important than the impact that electrification had at the beginning of the 20th century, but probably happening faster, which is something that we should be concerned about. For example, it has been estimated that more than half of the jobs — this is a U.S. study, but we can imagine the same thing happening in Canada — will be partially or completely automated, bringing new products and services and increasing productivity greatly, but also leading to potential negative impacts in terms of unemployment. There will also be new jobs created, but they will be in different places.
One thing we care a lot about is creating a lot of these technology jobs, but of course the people who are losing their jobs are not the same ones who have a PhD or masters and are going to be creating these technologies. However, there will also be new jobs in non-routine work, and new jobs that involve more work for people who are in jobs involving a strong relational or human aspect. I don't want to have a robot take care of my children, for example.
There will be a transition, and this transition could be painful and involve a lot of human suffering. So it is important to start thinking about it and about how our social system, social security and education systems may need to be adapted to these changes in the next decade, roughly.
Financially speaking, one thing that is important is that a lot of wealth will be created with AI, but that wealth will mostly be in countries where that technology is being developed and where the companies creating it are. We would like Canada to be a producer of AI and not just a consumer. Every country in the world will be a consumer of these things and may suffer some of those negative impacts. We would like to have that wealth created in our tax base, at least sufficiently so that we can pay for those transitions that are coming.
Hence the efforts of governments to stimulate innovation and technology transfer in this area are very important. In particular, a crucial element of this is talent, the expertise that we have or that we can attract. For example, in Montreal, in Toronto and Edmonton we are creating AI institutes that will allow us to provide the best-in-the-world facilities for researchers doing fundamental research in AI, as well as surround these people with groups of applied researchers and innovation technologists who will connect that talent with local industry and stimulate the creation of small and large companies.
Another important thing that is being discussed in the community, and also with governments, is that it's important not just to think about the profitable applications of AI, but also to think about those applications that have a socially positive impact may not be profitable but may be important for many people. For example, as it is important for this committee, applications in the medical areas. However, it could be in other areas, say providing free services to the poorest. That could be useful for a lot of people.
In MILA, my institute, we already are working with dozens of companies, and about 100 companies are knocking on our door to work with us. Many new companies are created, sometimes by the students in the lab and sometimes by outsiders who come and ask us for technical help. In particular, we also work with companies in the medical area, and also with medical researchers and hospitals, to explore the numerous medical applications of deep learning and AI.
One example that I'm connected to is a company called Imagia, which has developed a cancer detection system for a particular kind of intestinal cancer from medical images to identify the parts of the image that contain cancerous cells. According to their tests, they can do that slightly better than the best doctors and much better than the average doctors.
In addition to all this investment in basic science and innovation, one thing that I and others believe is that we must start thinking about the social and ethical implications of these changes and begin a discussion in our society involving not just the scientists and the engineers but also the experts in the social sciences, humanities and ethics who have something important to say about how to handle those changes.
As an analogy, and as a motivation for doing this kind of work, think about how much human misery and political turmoil could have been avoided if, at the end of the 19th century, people in power had realized that the Industrial Revolution would do a lot of damage before it would bring a lot of wealth. If we had put the social security net that ended up happening in the 1940s and 1950s and 1960s earlier, say at the beginning of the 20th century, a lot of that misery would have been avoided.
This is something to inspire us. The last thing I want to mention regarding the social aspects and the ethical aspects is one of the concerns that people in my community have — and we are having a lot of discussions already around the world about this — about how those advances in AI could be used in ethically wrong ways. One of the most salient examples of this is the use of AI in autonomously lethal weapons, weapons that can kill without a human involved in the decision. There were letters two years ago that I signed, and many others signed, to ask governments to work on this, and there is a committee of the UN currently working on this question.
I will stop here for my little speech.
The Chair: Thank you very much. I will now turn to Dr. Ferguson-Pell who is a professor at the University of Alberta but is appearing before us as an individual. You have the floor.
Martin Ferguson-Pell, Professor, University of Alberta, as an individual: Thank you very much. I'm part of the Faculty of Rehabilitation Medicine at the University of Alberta, so a lot of the focus of what I'm going to talk about is the use of what I would describe as a tool kit that can help us to address problems in rehabilitation and chronic disease management.
I will touch on three areas, although the scope can be much larger. One of them is the delivery of care to people in remote rural communities. The second is in the area of prosthetics, and the third is in the area of enhancing the effectiveness of education.
The tool kit really comprises a whole series of different technologies. Some of them we link closely to robotics, some of them more loosely. For example, virtual and augmented reality are rapidly growing areas which I think have enormous potential in the three areas that I have identified. I'd like to use my time mainly talking about the virtual reality and augmented reality side of things.
The problem we have for people who live in remote rural communities is the enormous expense of either getting clinicians to them or getting them to clinicians, or both. In the area of chronic disease management, one of the challenges is that a lot of the care these patients need is, in a way, serial or longitudinal care. It's not a single face-to- face interaction where a diagnosis is made and maybe a surgery performed and it's all over and the problem is solved. It is long-term care over many years.
This becomes extremely difficult for people living in remote rural communities, indigenous reserves and in the military, as well.
When we think of tele-medicine, we tend to think, I think, of a web cam sitting on top of a flat screen and maybe a group of people in a room talking together. In a way that's the way tele-medicine tends to be delivered in that it tends to be more of a discussion rather than a physical interaction.
But at the University of Alberta Faculty of Rehab Medicine, for the last five or six years we have been delivering a master's degree for professional physical therapist and occupational therapist remotely, at two separate sites simultaneously. So we have our hub in Edmonton, and then we have two satellite programs, one in Camrose and the other one in Calgary.
The reason we established this was because we were finding that the clinical professionals tended to settle where they were trained, and that meant in a province like Alberta, we had very patchy access to these professionals. So what we put in place was a very advanced way of delivering what we call "synchronous teaching'' to these students, where we would have a generalist instructor at the remote site and a specialist instructor at the hub site.
What the generalist would then do was follow the instructions of the specialist instructor and take the students through whatever the topic of the class would be.
Our sense is, well, if we can do that and instruct students, then why can't we do similar things in a very hands-on way in the delivery of care to patients in remote communities? One of the challenges is we need information that we take for granted when it's a face-to-face interaction with patients but isn't available to us when we have someone in a remote location. For example, the sense of touch, or the sense of a force being exerted by the patient, or certain nuances in terms of the anatomy of the patient that we need to be able to monitor while the patient performs a certain series of activities.
These are all part of how a typical assessment would be done, and so what we're now able to do, and what's interesting at very affordable prices, is essentially to put together a tool kit, at less than $10,000, that would be located at a remote clinic with a generalist clinician who would be able to follow the instructions of a specialist clinician who would be based at a hub urban centre.
The Internet of things is essentially a whole tool kit of different types of sensors that can be plugged into the Internet and then the data that they are generating received in another location.
What we're doing is beginning to put together the necessary components to undertake a series of different kinds of clinical assessment, in many cases longitudinal over a period of weeks or months. This can include the sense of force, the ability to see, in detail, certain features of the patient and the ability to measure movement of different parts of the body in a very accurate and quantitative way.
We can use it, for example, to look at things like the stability of the patient. So if they have a concussion or they may have some other form of impairment that may affect their ability to maintain an upright stance, then we're able to monitor that. In someone who's had a stroke, we want to be able to monitor what may be happening, for example, with visual neglect, all of these things can be done using these relatively inexpensive sensors.
The thing is that when you're doing this you're generating many channels of data at the same time, and so the concept that we are developing and have demonstrated in our lab is that we send raw data down the Internet pipe to the urban centre and then build a series of apps that can then interpret that data and present it to the specialist clinician, using a display system. It could be VR goggles or it could be something like the HoloLens augmented reality display.
The advantage that this gives us is the ability to port to the data that we're interested in. If you can imagine you're sitting there, and in front of you is virtually displayed a series of dashboards, each dashboard representing a different sensor. If you want to look at what's happening to muscle activity, you glance across it, and then you move close up in the virtual world to that display.
Then maybe you want to look at something else — you may want to look at a force measurement, so then you look in the other direction and then move across to that.
In the package I sent you is what is inspiring us, and this is some work that Microsoft is doing, using a technique called Holoportation. What Holoportation does is truly Star Trek in that what it enables us to do is to take a three- dimensional representation of a person in a remote location, live, and port that person's image in 3-D over to a second location which, in our case, is the urban centre.
If you have a look at the handout that I gave you, there's a link, and you can play a little YouTube video of Microsoft demonstrating Holoportation.
I believe that within the next three to five years, this will become a mainstream technology and will revolutionize our ability to interact with patients in remote locations. I think what we're beginning to do is to put in place the building blocks, and then this technology will enable us to do some remarkable clinical things.
Now, why is this important to Canada? I would say Canada is in the perfect position to be the living lab that develops these technologies. Just look at Alberta alone. You have two outstanding urban centres in terms of the quality of medical care being provided. In Alberta, we have a single health care system, Alberta Health Services, and we have a huge geographic footprint that we're responsible for.
And so if we wanted the perfect environment to start to demonstrate these technologies, Canada is in a great position to do that. Why is that economically important? Because when you start to put those apps and those technologies together, they can then be applied to many other remote locations around the world, such as China, Africa, Australia, Scotland and many other places that struggle with this problem of delivering care to people in remote locations.
I wanted to bring to your attention what I think is a very exciting opportunity, clinically, to patients the federal government is responsible for, such as indigenous people and people in the Canadian Armed Forces, and also to look at the opportunity to develop these technologies for economic benefit as well. Thank you very much.
The Chair: Thank you both very much. I'm now going to open up the floor to my colleagues, and we will begin with Senator Eggleton.
Senator Eggleton: Thank you very much, both of you, for your presentations. They were both quite informative for the purpose of this meeting.
Dr. Bengio, I'll start with you. Thank you for raising the issues of the social and ethical implications of these changes. You talked about disruption in terms of employment for many people. Yes, there will be new jobs created, but ultimately there will be a lot that will be lost. I think those are important issues to raise.
The other question that comes to mind is: How far should we be going? Where should we draw the lines in terms of artificial intelligence and automated systems? You have drawn one line yourself. You said you wouldn't want your kids looked after by a robot. Well, with some of these things, particularly as you get into deep learning, as I understand it, you can have machines that could be quite powerful and quite influential and potentially could be harmful in some respects. They could be very manipulative of a person in exercising undue influence.
I'm talking about this in the medical field. You have already said, well, there's one example of something we shouldn't allow, and that's autonomous lethal weapons. I understand and agree with that.
Let me take it into the health care area. Where should we be drawing lines here? We need to think in advance of where a lot of this may be going so that we don't go down a path that develops this in a harmful way. Can you comment on where we might draw the line?
Mr. Bengio: For now I think of the kinds of tools that are going to be developed in the next few years, with deep learning, for medical applications. Mostly, it is providing additional elements of information to doctors, at least in countries like Canada. I don't see these things as replacing doctors in any way. We still want doctors to interact with people.
I don't have a specific example of where I could see this going wrong in the medical applications. One area, which is related, is the question of private data. There is a social dilemma between, on one hand, the fact that I would not like my personal information to be used for anything that would put it in danger of being available to people who shouldn't have access to it. At the same time, for deep learning and AI to really succeed, we need to pool together the data from millions of people. There is a balance here that needs to be struck.
I don't see this as a danger rather than trying to find the right balance between the individual need for privacy and the collective need for better treatment.
I don't know if you have something in mind, perhaps?
Senator Eggleton: Let me give you one example because it has been raised here before. That is what I think we are calling Dr. Watson, Watson at IBM, and its ability to perhaps have a better record of diagnosis than some physicians.
The concern here is over the influence that this will have on the medical community, that the physician will feel that they have to bow to the machine when their instincts suggest otherwise simply because there is a better statistical analysis. When dealing with your patient, looking for the best odds is probably — let me give you that as an example.
Mr. Bengio: I think it is an issue, but I am not too concerned. Right now, maybe some doctors, and probably many doctors, are afraid and would say, "I'm not going to use this thing, this robot.'' Really, once you get used to having machines do things and compute quantities for you and you get to know about their reliability from experience, it's not very different from a doctor using a test that has been done with a robot that looks at the blood of a patient and comes up with numbers. The doctor didn't actually do the tests herself. She trusts that the machine counting the cells is doing it right. So I'm not that concerned about this aspect.
Senator Eggleton: Okay. If I may ask Dr. Ferguson-Pell a question, what you are describing here is fascinating. I think there has been a lot of concern about being able to provide good medical treatment to people in rural areas. Your telemedicine program would appear to be quite promising in that regard.
How is it funded? Do you get to most remote communities? It would still, I would imagine, involve a fair expense. This is very high-tech equipment. Are you able to have an extensive reach into Alberta?
Mr. Ferguson-Pell: This is at proof-of-concept stage at this point. So we are not delivering care right now, but we have been very carefully considering that question from several standpoints. One of them is the funding; the other one is the speed at which you can develop this technology so that you can stay ahead of the competition.
One of the challenges with any telemedicine system, when you are dealing with a primary health-care system, is the privacy management that is introduced by that health-care system, which slows down your ability to develop.
We are looking at this in a three-pronged way. We are looking at it from the standpoint that many of the services that we are getting started with are ones that can be provided through the typical physical therapy clinic, which is a private pay clinic. Many of the services that people receive for physical therapy is out-of-pocket cost. That enables us to get started without getting caught up in the complexity of how we get a policy change or a change in the funding model within the health care system that would deal with the fact that we have multiple clinicians now — generalists, specialists. Then, also, maybe there are others involved in the team, such as doctors and physical therapists, with different roles. That is one way of expediting the development from a technological standpoint.
Senator Eggleton: Hopefully, eventually, you'll get into the health care system?
Mr. Ferguson-Pell: Absolutely, we intend to. Also, that is why the third prong is the opportunity to work with the Canadian Forces.
Senator Stewart Olsen: Thank you both for being here. I have one question for each of you.
Professor Bengio, one of the things that we really see with these technologies is the changing world. Have you given thought to the preparation of our youth? It almost has to go right back to beginning high school or even grade school, when they start to think about a career path. What should they be thinking about? These are going to be the jobs. Other jobs are going to be changing. How can they move forward with that? What should they be thinking about?
Mr. Bengio: Because the changes are going to be very fast, I believe that it is really important that our education system prepares our youth to be fairly broadly educated. If they specialize in something and then that something gets to be automated, that is pretty bad for them. That is one aspect.
The other aspect is pretty obvious that we want to encourage youth who are going to contribute to the development of these technologies, so computer scientists especially and engineers. There is already an effect of all of the discussions in the media about AI and the breakthroughs that have happened recently. I can see colleges calling us and asking if they can visit and students in undergrad and high school being interested. We need more of that, obviously, because Canada needs to have as many of these high-paying salaries that are going to create that technology here as possible.
We need to continue carefully evaluating where things are moving. If you look at what economists are writing or just your good sense, you can make fairly good guesses. The people in charge of curricula in schools and colleges need to pay attention to that to try to tell the students about what is going on so that they make the right choices.
Senator Stewart Olsen: Dr. Ferguson-Pell, your demonstrations on what Alberta is doing are fantastic. I come from New Brunswick; we don't have a medical school. Are these done in silos? Would you do your telemedicine strictly for Alberta, remote Alberta, or can you envision a Canada where a school in British Columbia can provide this kind of learning for someone in New Brunswick, for a school in New Brunswick or a class in New Brunswick?
Mr. Ferguson-Pell: Absolutely. One example of this at the moment, in our faculty, is around stuttering. We have an institute called ISTAR, which is to help people who stutter to overcome this disability. We are delivering stuttering therapy, essentially, to countries in the Middle East. This becomes simply a matter, I think, of, in a way, financial logistics in terms of how you manage the resources that are needed in order to deliver it.
To some degree, some harmonization in the methodology is needed so that we agree on some basic protocols as to how to deliver a particular kind of assessment. Take someone who's got an injured rotator cuff. We would have a standard operating procedure that we would want to follow using this technology so that we go through a systematic assessment of that person's rotator cuff. We would want that to be harmonized with other collaborating provinces or other organizations so that we can get it done smoothly and quickly.
With all of these things, a lot of the cost is around how long it takes to do the assessment. That means the technology must be seamless and reliable; you plug it in and off we go. We have all had that experience with video conferencing. Second, the protocols can help us do that.
To give you confidence that this is doable, the experience we have had in teaching the physical and occupational therapists in remote locations — not so remote, Calgary and Camrose — co-located groups — we have demonstrated that those students, when they graduate and sit the national exam, come out with equivalent or better marks to those who have face-to-face. I am confident this will work and give good quality care, but we need some degree of harmonization.
Senator Seidman: Thank you both very much for being here.
As a Montrealer, I have to direct my questions to you, Professor Bengio. I'd like to start out by congratulating you sincerely for the enormous efforts you have put in to creating this hub in Montreal. I have collected the newspaper clippings, been watching Google and Microsoft, and been feeling energized and excited about that.
In your presentation to us today, you have said clearly that something is changing and we are moving in new directions. You have said that to keep our talent and attract more, we need to provide the best in the world facilities. You want Canada to be a producer of AI products, not just a consumer. You are already demonstrating that in Montreal.
If you use Montreal as a model for other potential hubs in the country, which I hope is what happens —
Mr. Bengio: This is what happens.
Senator Seidman: What were the positive assets? What is the model that you will transmit to other parts of the country? What were the positive assets, and what were the particular challenges that you overcame?
Mr. Bengio: The core asset is a critical mass of talent. I mean the top-level talent: people, researchers, professors, PhD students, postdocs who are doing the best research in the world, concentrated in one place, working together and attracting the attention of not just the scientific community but also the business community.
This is what happened.
Another thing that happened is a shift in the attitudes of investors and businesspeople — this is just the beginning — that is making it possible for AI startups to begin here immediately and not be sucked to Silicon Valley. It used to be that Canadians would start high-tech companies, and as soon as they reached a certain level, they would not necessarily be bought, but investors who were sufficient risk-takers — there would not be enough of them in Canada. They would be essentially people from California, and they would say, "Come to the valley, because we want to be close to you.'' A lot of companies left like that.
That has now changed. Canadian investors are starting to understand the asset we have here, and foreign investors also understand that if they want to be part of this game, they have to accept that Canadian companies will stay in Canada. The competition effect is moving things in the right direction.
In general, to come back to my first answer about critical mass, it's a virtuous circle. That is why the intervention of government is so important. It is pumping the motor to get it started, and then it is self-growing in the right direction. We probably need to keep pumping to some extent, but now big business is coming in.
There are still challenges; you asked me about challenges. In general, there is a lot to change in the mentality of investors in Canada. They are very conservative. Even more conservative are the big corporations in Canada. They are not used to making these big, bold investments into the future. They want to be reassured that whatever they buy will be safe. Somehow we need to change that. I know some companies are starting to open up, but this is something that needs to change.
Senator Seidman: I would like to ask you the same question, Mr. Ferguson-Pell, as long as I can have a second round to ask Professor Bengio another question.
The Chair: We will put you on a second round, but I can't guarantee it.
Senator Seidman: I understand you are a huge innovator as well. I must say the idea of Holoportation is really something. It is "Star Trek'' — I can visualize it. I would like to ask you the same question I asked Mr. Bengio.
Mr. Ferguson-Pell: To ensure I remember your question, the first part of the question really was: What are some of the drivers for this innovation, and what resources or investment are necessary to make it successful?
That is a pretty challenging question in this context, because there are a lot of stakeholders. One of the stakeholders, in a way, is our moral conscience, and that is the importance of providing equity in the access to health care. That is a very powerful principle of Canadian health care. There is a lot of data to show the inequities that people who live in remote and rural settings experience simply because of where they are located.
I can't promise that this technology will necessarily directly reduce costs in that you could do simplistic calculations and say that instead of that person having to travel from a remote location to Edmonton, which costs $5,000, that will pay for half of this toolkit that could maybe be helpful to 100 patients. That is a simplistic way of looking at it, because they probably would never have made that trip in the first place for a lot of the conditions we are talking about.
This is an investment in part in fairness as well as an investment in developing a technology that has the potential for economic benefit to Canada more widely.
Second, the stakeholders are pretty diffuse. As we were saying a few minutes ago, this technology can be beneficial to all provinces and beyond. Federal support to help develop these technologies, from an innovation standpoint, is important, and it has already provided that support. We have had CFI and Mitacs funding. They are exactly the kind of funding we need to do this kind of work. It is more difficult to position this kind of work in the CIHR context.
Those programs are critically important to enable this sort of thing to happen.
With industry investment, I think there are opportunities. The Mitacs funding we received, for example, was a consequence of an investment the TELUS Health made. Companies involved in communications can see benefits; they may be collateral benefits in that they are seeing increased utilization of their IT infrastructure. So they are not actually investing directly because they are obtaining a new industrial growth around a health technology, but they are getting secondary benefits.
These are the kinds of complexities we are working within this area.
Senator Seidman: And this contributes — I am just thinking of Professor Bengio's point that Canada must be a producer. We can then export that technology and benefit from that aspect, not only within our country but outside our country. Is that the point you're making as well?
Mr. Ferguson-Pell: Yes, and there are some technologies where it's more difficult to do that than others, simply because there are many other players in the field. So in order to be truly successful, nimble and get to market fast, you have to have the kind of venture investment and angel investment with the confidence behind it that Dr. Bengio was talking about being the first to market, and we are frequently not first to market because we don't have that culture.
In this particular case, what I'm suggesting is we have geography on our side, and I think that's our opportunity maybe to get ahead of the game, because we have the medical centres and the excellence in health care. We have a publicly funded health care system that is nicely positioned to be able to do this sort of thing. We have the geography and we have the technology, so we have a sweet spot to work with.
Senator Dean: Thank you, both. This is terrific and exciting stuff. I want to talk a little bit about the role of government. We have heard about government priming the well or pumping, and that funding is important. We have heard both of you say that.
When we think about success factors in this terrific, decade-long success story, are there other things that government can do, other than providing funding? Or is it just simply a matter of providing the funding and then stay out of the way? That is my first question.
Second, we know about the virtuous, positive social and economic impacts. I want to return to the negative ones, and without the particulars, because many of the negative aspects of this, by the very nature of the technology, will be unpredictable by virtue of its disruptive nature.
Where does responsibility lie in terms of governance? Who has the lead on thinking about and looking at and having responsibility for mapping and attempting to foresee some of the downside consequences of this? We have heard that health inequity is one. The exportation of benefits to others, while they may not be available to some home countries, obviously another. Any thoughts on those questions?
Mr. Bengio: About the role of governments, besides initiating the kind of innovation that we are seeing in AI, as I mentioned, our ability to keep and actually bring to Canada the best people in the area, scientists but also entrepreneurs, for example, is closely connected. One important aspect of this is our immigration process.
In my own life as a professor, I encounter a lot of issues with students having trouble, especially the students coming on short-term internships and things like that, but even grad students coming and being delayed one session and then deciding to go to another university elsewhere in the world. I believe that we could do a better job to make the process of getting visas and work permits and so on seamless and faster, especially for these areas that are critical like this, but I think ideally for everyone. So that's one place I think where it matters in terms of policy.
The other place which is difficult is regarding sharing of data. Health is mostly dealt with in the provinces. The provinces, for example Quebec and Ontario, are doing pretty nice work trying to centralize their medical data. This will become really important for AI, but each province is doing it separately.
I'm not sure how we will be able to deal with this, but as a country we would be able not only to provide better health care but export our systems and be much more competitive if we found ways for the different provinces to allow researchers and companies involved in the development of these products to have access to shared data. Of course with all the techniques that are used, we must ensure the data isn't leaking in any way. It's a jurisdiction problem where maybe the federal government could have a role in bringing to the table the difference provinces.
Senator Dean: Thank you.
Mr. Ferguson-Pell: First of all, I would absolutely agree on the data point, and I would add an extra dimension to that in that data can be used in a number of different ways. As academics, our interest is in using data on the whole for research purposes. We need to be able to access that data supported by appropriate ethics approval. Ethics approval becomes complex because that means individual patients need to have given their permission for that data to be used for the purpose that we would like to put it to as researchers.
Creating a culture — which I do not think patients have a lot of problem with — where patients more or less opt out, where their data is essentially freed up for use for research purposes, as long as appropriate anonymization is put in place, could be a real enabler. I think this is something we really struggle with in Alberta in particular because I think Alberta tends to have a very protective view relative to other provinces in Canada.
We certainly struggle with this as academics in Alberta, but I think it's a relatively common problem across the country.
The second point I would make is to do with joined-up-ness whether it's at a provincial or a federal level. When we start talking about acute and chronic care, we suffer from really challenging problems in joining up our recognition of what holistic care really is, and this occurs in the field of people who are aging in place where appropriate, relatively inexpensive interventions at the community level can save enormous amounts of money at the acute level. But the ability to move money across the divide, move policy across the divide and move administrative responsibility across the divide is a perplexing problem, and it doesn't make common sense, but it's the nature of the way things are put together.
That becomes a real barrier I think on the chronic disease management side of things because it's difficult in isolation to demonstrate that you are saving money. But when you look at interventions and innovation holistically and you include what the benefits are at the acute side, then it's a slam dunk. I think that's the challenge we have both at a provincial level and a federal level.
Mr. Bengio: I agree very much with what you said. One connected element regarding the opt-in of patients to make sure their data could be used for research is we have to ensure that patients feel secure about that data; in particular, it has come to the Canadian government, the question of what insurance companies could be doing and asking from the data that has been measured on patients.
I don't want to go into that debate, but obviously from the point of view of developing AI with patients' data, it's much, much better if patients feel confident that the data is not going to be used against them in one way or another.
Senator Unger: Thank you both very much for extremely interesting presentations. Dr. Bengio, just a comment: I really like your message that if you want to do business with us, come to Canada. I really applaud that. I recently read in one of our national papers about a company in California that packed up and moved to Toronto specifically regarding AI. That's a great message.
You talked about your work with dozens of companies and hundreds more knocking on your door. I'm wondering how you would handle all of this potential business.
Mr. Bengio: Well, precisely what we will be doing with a lot of the money that we will be receiving from both the provincial government and the federal government in the next few years is to make it possible for academics in my group to recruit applied researchers who will actually do the work of the technology transfer and working with companies.
The professors and the grad students for the most part will continue to do their basic research, their long-term research. We don't want to stop that; otherwise, we will lose eventually to other places in the world where they are continuing to do this. We won't have the attractive power that we currently have for talent.
Instead, we need to bring together in the same places — that's what these institutes are supposed to be about, the ones in Montreal, Toronto and Edmonton — both the fundamental researchers, who are already there, but recruit a large number, potentially more than the number of students and professors that we currently have, and they will be former grad students, engineers, people who we recruit from overseas who will be the interface between those companies and the research.
For example, in my group, we are telling those people when we recruit them that your main mission will be to work with those companies, but that will be like 80 per cent of your time, and 20 per cent of your time will be essentially free and you will be able to collaborate with the grad students and the post-docs doing fundamental research, publish papers, so they get to stay connected with the advances in science, and they are motivated to be here rather than working in industry and maybe have better salaries.
Senator Unger: Thank you.
For both of you I had questions that have already been asked, but Dr. Ferguson-Pell, you mentioned funding because Alberta has that kind of system. Do you see a role in the private sector or for the private sector to get involved with your work?
Mr. Ferguson-Pell: Absolutely, and in fact I think the private sector might be the early adopter, and it might be one way that we can get, if you like, soft venture funding in order to get some of this off the ground at an early stage. If you could imagine, there are a number of interventions for a typical physical therapy clinic that delivers fee-for-service care that is out-of-pocket paid, where expanding the market for them, by being able to reach out more effectively into rural communities, would be a very appealing thing to do. It relies, of course, upon those more remote communities to be able to afford to buy that. So there is an underlying inequity around the delivery of, for example, physical therapy services.
In Alberta, for example, if you have a hip replacement or a knee replacement, you get a certain number of physical therapy visits associated with that. So that would be a very good starting point.
The other thing we're doing now, which I think is another opportunity where it's at the interface between the health system funding and the private funding is an intervention called GLA:D which came out of Denmark, and this is a physical therapy intervention that is intended to delay the amount of time needed for you to have your knee or hip surgery. In other words, the physical therapy program essentially extends the period of time that you can live comfortably and functionally with your normal hip or knee before you have to have an artificial replacement.
This has economic benefit to the health care system and to the patient as well as a quality of life benefit to the patient and is an opportunity for a private clinic to be able to deliver that service. There are these sorts of overlap areas that I think are interesting areas for us to look at at the beginning because it gets kind of everybody on board, but we don't get into complex conversations about where is new money going to come from in order to deliver this service, which is always challenging, especially in difficult economic times.
Senator Unger: Yes. I had one more short question.
The Chair: All right.
Senator Unger: I'm wondering also about security and privacy of data.
Mr. Ferguson-Pell: Yes. The data that relates to things like force measurement, movement measurements and things like that is very easily encrypted, and usually that would not carry much in the way of an identifier because the data is being streamed across in real time. Then it is being stored in a relatively normal way.
The tricky part is the video information, and there are protocols for doing that already in telemedicine, and they will need to be enhanced as we go forward, as we use this technology more and more, but this technology is in place to be able to do that. It's a matter of just, in a way, getting the approvals and the acceptance that the encryption techniques are sufficiently robust for the system to accept.
Senator Unger: Thank you both very much.
[Translation]
Senator Petitclerc: Thank you both for joining us and for your presentations. My question is for both of you.
Mr. Bengio, since I too am from Montreal, let me start with you.
[English]
I'm trying to get a sense of how democratic, if I may say, your sector and area are or will be. I get that this is very exciting, and I guess maybe there are a few parts.
I'm trying to get a sense of where it's driven from. Is it led by the businesses? Is it led by the academics or a combination of both? The reason I want to know is I see that there is a lot of investment and it's high technology. It seems to me a lot of this would be very expensive research and products.
How do you get that to the most Canadians? How will it help as many Canadians as possible? I see that it could remain a very privileged access or it could save the world — I don't know — maybe give me your opinion on that and if you can give me a sense of where it stands.
Mr. Bengio: Maybe one part of the answer is something really nice is happening in the machine learning scientific community, which is that it's currently and has been for many years, as far as I remember, very open. So not only do scientists publish their work in the usual ways, but they publish their work even before it's been reviewed in what's called "archive.''
As soon as somebody has a good idea that's sufficiently fleshed, it's posted on the web, and this is not just the academics doing this. The researchers in industry, at least in the leading AI industry, so that's companies like Google, Microsoft, Facebook, IBM, do the same thing. There is a lot of movement of ideas.
Students spend some time in one place and then another place, so everybody knows what everybody is doing very fast. The reason I'm saying this is that for anyone who has the training, and, of course, there are thousands of people around the world who do, you can very quickly know if there's something new happening. This is not like what you see in movies where there are some isolated scientists maybe working for a company discovering something big that no one knows about. This is not at all how it's working. That science is moving by actually pretty small steps, even though the impact could be important, but the science itself is moving by small steps on top of what others are doing, and everybody knows what is going on.
Worse than that, the companies that are trying to be secretive are losing. They are losing ground. That's why, for example, a company like Apple, which used to be extremely secretive, decided that — at least as far as deep learning is concerned — they would open up and start publishing. For them, it was the only way they could recruit strong researchers in the area. Otherwise they would not even be able to recruit them, because those researchers care more about being part of the community, being able to talk to their peers and getting their feedback, than having a 10 per cent or whatever better salary.
That means that, in a sense, anyone with the training can build their own thing. Even regarding patents, there has been a movement in the last few years by those major companies to say that they would only use their patents in a defensive way. In other words, they would not prevent other companies from using the science they are producing, or just not even patent the things that are more on the science side of things that they are doing. So that's very good.
Also, even though it's true that it may take some capital to develop, mostly it's people. There is computing equipment, but it's mostly people, so it's not actually that capital-intensive. A lot of the things that are being developed are already mass distributed. If you use almost any of the Google services or a large part of the Facebook services, you are already using deep learning and you are getting it almost for free. You are getting advertisements, but that's another problem.
That technology is being used and is sort of mass distributed. Everybody will be a user of these things, and is already a user to some extent.
There is a question of democratization at the level of business. I co-founded a company called Element AI, which is trying to make that kind of technology accessible to the other companies besides those big IT companies, those companies which don't have the expertise. This is true of many Canadian companies, even large ones, and even large multinationals. They just don't have the expertise or the talent, so they don't know how to jump into this. So there's an issue there, but companies like Element AI and IBM are making that kind of expertise available to other companies.
I'm not sure if I answered your question or if you had something else in mind.
Senator Petitclerc: We want to know the impact on health. Is it your perception that it will have a positive impact on the health of Canadians? Because, in the end, that would be the desired result, right?
Mr. Bengio: Yes. In a sense, this is related to what Martin talked about earlier. There are cases where it's fairly obvious that it's profitable for a company to build a product that will be used by hospitals, for example. The example I gave earlier of medical images, this is an area that's booming because it's a very straightforward path from building the technology to making money out of it.
However, I'm sure there will be other cases where it's not so clear from a business point of view, yet there could be important positive social impacts. That's where having a strong role for governments — for example, those institutes that are being created — is good.
Academics are very motivated by the idea that what they are going to be doing could positively impact a lot of people. In my group, we are ready to tackle problems where there is no company that is interested, but there are medical doctors or hospitals that think this is an important problem that could be useful for many people, and then we just go and do it. Our salaries, as professors, are paid by the government, so we are doing things for the common good. It wouldn't necessarily work like this if you were in a country where it was all driven by the immediate profit motive.
Senator Petitclerc: Thank you. I don't know if you had anything to add to that.
Mr. Ferguson-Pell: No. I completely agree. I think it was covered very nicely.
The Chair: Before I move to the second round, I want to ask about a couple of things that have arisen.
With regard to the confidence and the ethics of using data that you and Senator Eggleton began to discuss, Dr. Bengio, I thought you gave a very good example when you referred to the counting of cells. In my opinion, it was a very good example because it focuses us on reality versus hypothesis. We quickly come to accept technology that gives us information that is beneficial to making a diagnosis or helping out with issues, and we quickly cease worrying about the ethics of the technology.
Now let's move up one scale to 3-D imaging, brain imaging. How do we know that the interpretation of the brain image of a particular individual hasn't been influenced by the data that went into telling the machine how to look for a brain and deal with it? Well, we just don't. If we have a brain injury, we want you to image us immediately. If we think we have a deterioration of mental capability, the first thing we want is an image. In fact, today it's getting so that people almost ask for an MRI, whatever they are going into the hospital with.
I think that one of the things we have to do as we look at these ethical issues around data concentration is the reality of how they're used and how they may benefit, and weigh the potential benefit against any potential negative issue.
We've had discussion here about this issue. As we get further into deep learning, we know how the learning is structured. We have been informed about how it has gone from simple stacking of photocopies into modelling on the actual brain and how it works, and Canada has been a leader. Part of the reason we're a leader is because we continued our neurological research beyond that of a number of other countries.
We now get into a situation where, in terms of diagnosis of disease, we know that many diseases are culturally or ethnically related. They are statistically higher and more pronounced in certain ethnicities than in others. So the issue then becomes: Is there sufficient data moving into it? In my opinion, your simple example of the cell counting is a good basis for us to reflect on, in that as we get sufficient data it no longer matters, because if you get a total ethnic screen of a good, solid sample of a population, then there's a much higher probability than any factors that will be considered. In fact, there's probably more diversity within one ethnicity than there may be across ethnicities. We know that the genetics, particularly at the microbiome level of individuals, even within any given ethnicity, is considerably varied across that population. Those factors have to be taken into consideration now when we look at the microbe and so on.
I don't want to go on at length, because I do have a question for Dr. Ferguson-Pell. Is there anything in what I have summarized that causes you to focus in on a further answer to Senator Eggleton's question?
Mr. Bengio: An issue that you raised which is well studied now is the issue of bias due to the particular choice of data that we are using to train the machine. The machine is only imitating the things that it sees in the training data, and so if the training data reflects the biases of the people or the lack of diversity from which data came, that will also be in the machine, at least if we don't do anything special about it. But we can do something about it.
For a few years — and some of the papers came from Canada — we know some techniques that can help us to reduce that issue considerably. The idea is if we also measure, say, the ethnicity as the variable we would like the system to be insensitive to, because we would like the system to take a decision irrespective of gender, race or whatever the variable is, we can do that. We can simply train it to predict those variables and then become insensitive to them in its other judgments. It's not necessarily perfect, but we have knowledge of the mechanisms to do that.
I can imagine that, in some situations, it could be something that's legally required, just as a form of respect of the values of a particular country, so that some systems in medical or in legal areas have those safeguards. Scientifically, we know how it could be done. There is extra effort when someone delivers a product to enforce these kinds of ways to make the system unbiased, and it may be at the expense of the accuracy of the system. This may be something companies will not do unless they are asked to do so.
The Chair: You said something I was hoping you would say and that is the implication that if we get sufficient data across sufficient ethnicities, we not only have the data for a first scan to give us a higher likelihood of a successful scan, but we have the ability to ask the second question.
That is: Now that you have given us your data, if you take your subset of a particular ethnicity and apply it to the sample you are dealing with when we are talking to our robotic instrument here, does your interpretation of what you have just scanned change? Ultimately, we have the potential capacity, if we scan sufficient data into the system, not only to get a general quick answer, which may be absolutely correct the first time through, but to refine questions based on different body structures, ethnicity and so on.
I will leave it at that.
Mr. Bengio: Sure.
The Chair: Dr. Ferguson-Pell, I was struck by the Holoport example that you gave which, if I understood correctly in the first instance, is the ability to bring the global three-dimensional physical structure into view, but at first glance it doesn't go internally within the individual. I want to take that next step.
The question that I have been asking with regard to our abilities with optics and deep learning — that is, having a reference background — is surgeries in difficult areas. Two that are in close proximity involving two very different disease issues but are complicated because of their location are anal cancer and prostate cancer. We know that errors are made in both because the surgeon is unable to see all he wants behind an organ in the intricacy of the region and so on, and a slight nick in those areas can do long-term damage.
There are reports that the system is advancing rapidly in giving a three-dimensional, on-screen view of the area in which the surgery has to occur and to reveal details about the physical structure and the proximity of various systems — I will leave it at that — that are in both those areas, all of which are enormously important and tightly packed.
From your experience with the Holoport, do you see this being able to move internally and help with that presentation, taking the two examples that I gave?
Mr. Ferguson-Pell: Yes. I think we are already seeing it, but not in real time, and that is in surgical planning. If you take images that have been created, for example, CT or MRI images, and you know a procedure that will be followed, the example we have at the U of A that is compelling when you see it, is replacing a heart valve that was previously implanted and now needs to be replaced. This is a catheter procedure. It is not open-heart surgery. However, it is tricky to feed the instruments through the veins to get to the right place, remove the old valve and put a new one in, especially if it is not located ideally. To be able to rehearse that operation before it is undertaken has the potential to significantly reduce the risks of the kinds of problems you are describing.
I would say the low-hanging fruit in this area would be in surgical planning and rehearsing and simulating a situation, but doing it with the patient's own data. What we have been used to in the past is generic models that are produced that enable people to practice. Recently, in some areas individual customized models are made of the patient themselves where you can practice the procedure, but that requires producing a physical model, which is where 3-D printing can be helpful.
But the opportunity in virtual and augmented reality is that you can take that MRI image for that individual patient. You can start to rehearse the procedure but you can also change its scale. Unlike the physical device, where you are fixed with the scale, with this you can actually get inside and take a look around from the perspective of the blood vessel or from another part of the heart, for example, where you are wanting to see exactly what the shape is. That is where I think the immediate opportunities lie in that kind of surgery.
The Chair: Thank you very much. That was great.
I want to come to the issue of confidentiality of information. It seems to me that, in Canada, we have been substantially held back in the health area because of our paranoia over privacy of information. For example, we still do not have electronic health records for Canadians. In looking at that, we have heard testimony from Canada Health Infoway and others here and we have delved into this in a number of areas. I am increasingly of the view that the requirements put on providers, where they have confidentiality protection as the highest requirement in the requests for proposals for these things, is such that it makes it virtually impossible to make a practical electronic health record.
If we can't even do an electronic health record, we will not get to the things you are talking about here, using data in Canada on which to base decisions for instrumentation and technological developments, and so on. We are already at a disadvantage even on our national population, which is smaller than one very advanced state, in particular, south of us. We have to be able to deal with that.
I am always amazed that we have this paranoia in this particular area, whereas people are jumping over themselves to get an electronic submission of their income tax; I have heard numbers as high as 80 per cent this year may be in an electronic format. I suspect in terms of immediate injury to people, there is nothing more potentially sensitive than their private financial information, yet we have been able to deal with that in these kinds of areas.
I want to come back to the issue that you have raised. I believe the access to large data is critical for our companies moving forward to be able to help Canadians, let alone to be able to compete with the products in the end. We have to be able to deal with this issue. The real issue that I think you touched on, in going through it, is the protection against using an individual's data against their opportunity for employment. Those are very real, but they are different than the issue of total data in the system analyzed for things that appear to arise in a given disease symptom over a wide range of patients in a wide range of different things.
If we are to be as advanced as Mr. Bengio wants us to be with regard to artificial intelligence and its benefits to Canadians, we have to be capable of dealing with a realistic problem. Otherwise, we will not get there.
I will turn to the second round now.
Senator Eggleton: Gentlemen, in Bill C-43, the government provided for $125 million to the Canadian Institute for Advanced Research, CIFAR, to support a pan-Canadian artificial intelligence strategy. I understand, Dr. Bengio, you are involved with that.
What are the components that you see of this strategy and what is the timetable you see in the development of it?
Mr. Bengio: There are a number of components. The bulk of the money is going to the three institutes. As far as I know, for Montreal and Toronto, the plan would be to have something like $40 million over five years. Out of that $40 million, $30 million would be used for chairs to attract and retain the best professors in the area and pay for students and post-doctorates and so on and $10 million would be for the operation of the institute. That is over five years, so it is $2 million per year. I don't remember the numbers for Edmonton but I don't think these have been made official. This is from the discussions I have been part of.
That comes to about $100 million. The rest goes to more pan-Canadian activities managed by CIFAR. One aspect is to promote the discussions around the ethical and social aspects of AI. Another is to encourage collaboration between the different centres. Finally, there is $10 million to help other places in the three major cities. For example, UBC used to have a pretty important group but, unfortunately, due to the brain drain, most of them went away. They could recruit more and that money could be used to help them attract professors in this area. That is roughly the division as far as I know.
Your other question was?
Senator Eggleton: Time frame.
Mr. Bengio: The three institutes are being set up. The Vector Institute in Toronto was created a few weeks ago. Hopefully the one in Montreal will happen in the next few weeks. The plan is basically that all three will start operations within this year.
Senator Eggleton: Will this strategy unfold in increments? Will there be reports from time to time as to its progress?
Mr. Bengio: Yes. CIFAR has been mandated as the organization that will monitor progress. They are already doing this for their usual programs that are fundamental research programs. There is a secretariat or a committee that will be put in place to evaluate and to take global decisions, for example, and put the bar on the quality of the applicants for those chairs in a way that will be consistent across the country.
Senator Eggleton: Do you think the amount of money being invested by the federal government — it is always nice to have more — is sufficient to make a difference?
Mr. Bengio: That money is really to help with the basic research aspects. We are working toward getting some money from the federal government for the innovation aspect.
You will remember that I told you we want to build institutes that have both the basic research groups and the applied technology transfer groups. That extra funding will be needed for making that happen. The provinces are putting money in for both, but having the extra money from the federal government for this would be important. If we only invest in the university research and we don't also help the transfer, then we are losing a big opportunity to make the whole thing succeed.
Senator Eggleton: Let me ask Dr. Ferguson-Pell whether there is any federal support into the program or is it entirely the Province of Alberta?
Mr. Ferguson-Pell: The one I am describing at the moment?
Senator Eggleton: Yes.
Mr. Ferguson-Pell: Yes, it is mainly in our area. Apart from the CFI and Mitacs funding we have received, they are small amounts of funding but they can have a good impact.
The interesting thing about the field that we are working in is that it is pretty eclectic. There are many secondary benefits. For example, the funding going into machine learning indirectly benefits us, too. Imagine that data being sent from a rural location into an urban hub and put into a repository. You do that nationwide and start to mine that and look at trajectories of care, and being able to predict what the outcome for a patient might be with a complex chronic disease that needs to be managed with a lot of resources and careful planning.
Machine learning is probably the only way that we will really be able to get to the bottom of what those trajectories are.
We have to expand our concept of what we mean by "data.'' Earlier we were talking about genetic profiles, for example, and metabolites and other measurements put into a data pool. They lend themselves very nicely to machine learning. Let me give you another scenario: A chronic wound. A huge problem and extremely expensive problem for our health system. What data are we generating around chronic wounds? What are we measuring? What do we look for? What characterizes a chronic wound? What helps us understand what the treatment is for a chronic wound? How can we optimize the rate of repair for that chronic wound and how do we decide where it is treated with conservative management or plastic surgery?
That is one condition that is a very major expense within our system. It is not a particularly pleasant topic to talk about. It doesn't attract the intelligentsia of academia to work on because it is messy. Yet it is an interesting area that could be looked at and could be one that would benefit from using machine learning techniques.
I go back to the point that when this basic research is done, the benefits to society rely heavily upon our imagination and ability to apply it to interesting, challenging and important areas, including chronic wounds.
Senator Eggleton: What are you calling a chronic wound?
Mr. Ferguson-Pell: A pressure ulcer, a diabetic ulcer, a surgical site that hasn't repaired properly, and so on.
Senator Eggleton: I see. Thank you very much.
Senator Seidman: Thank you both very much for your continuing offering to us of stimulating materials.
Professor Bengio, you mentioned Element AI, but we didn't hear any real description of how that fits into this whole puzzle that you have presented and this hub. There are pan-Canadian aspects and a lot of moving parts here.
Mr. Bengio: Yes.
Senator Seidman: I know it is always that way when we are in the midst of major innovation.
How does Element AI fit into this whole picture?
Mr. Bengio: We are trying to build an ecosystem, including academics, small companies and incubators, start-ups, large existing corporate, and Element AI started as a small start-up last fall and is now becoming a medium company very quickly. They will probably break the record for the largest round A funding investment. That will probably come in the next few weeks.
There is a huge interest for these kinds of companies in Canada and elsewhere in the world.
In a sense, they will be complementary to the kinds of things we are doing with the institutes. The institutes are academic-led and not-for-profit. Element AI is a company, so it is for profit. Element AI has already recruited a number of high-level researchers, so it's doing both basic and fairly applied research. One could think, "Why do we have the institutes and companies like this?'' Because it is a different way of operating. We attract different kinds of people, and we solve different kinds of problems.
There are all of these different actors, and Element AI is playing a different role, at least for the Montreal ecosystem, in having a fairly large group that has a lot of international visibility to attract the kinds of people who are currently going to, say, DeepMind in London, which is the largest company doing AI these days — they can compete with those kinds of companies that are specializing in AI and are large enough to attract really strong people.
Senator Seidman: Part of this is what you have described before, Mila, for example; it's private-public, so there is private and public funding, correct?
Mr. Bengio: Yes, but it is really a nonprofit. Although we are getting contracts from companies, first, we choose what contracts we take. What guides our decisions are how much it will impact the ecosystem and how good it will be in terms of social impact. These considerations are things we can have the luxury to consider if we are a government- funded organization, but a company like Element AI will look for profit, which addresses a different objective.
Senator Seidman: You have already anticipated where I am going with my line of questioning here, because that is exactly what I was trying to understand; namely, how you determine priorities and, if you even do determine priorities, on what basis. With Mila, for example, there is a social good element.
Mr. Bengio: Yes, exactly.
Senator Seidman: Given our committee's particular interest in health care, I will take it one step further.
Mr. Bengio: We are putting a lot of emphasis on the health care applications, irrespective of whether they will be profitable. We are talking to a lot of medical researchers, and we will get our funding in part from what the government already gives us but also from existing other government funding like CIHR, NSERC and things like this — and potentially companies.
It will be a mix, but at the end of the day, we are not limited by purely whether this investment is profitable in two years or something like that.
Senator Seidman: Is there some indication you can give us about what the best investments could be in the health field, where it could have the most impact and what your brainpower tells you are the real investments for the future?
Mr. Bengio: Medical images is a no-brainer. It is the low-hanging fruit. That's why there are already companies involved there.
Another area that depends on our ability to collect data is helping to, for example, process medical reports that are natural language. One thing that is happening with deep learning is that we can now use computers to extract information from pure text; it doesn't have to be structured data. Doctors' reports are even hard for humans to make sense of. This is a place where we can extract a lot of information. In fact, this could be used as a complement to medical images or to train the machine about what the doctor thought about the problem with that person based on the medical image.
There are a lot of places around medical data where we have the usual numbers, tests and so on, but the new thing, beside our ability to take advantage of large quantities of data, is our ability to exploit textual data. We have to make sure that we collect that data; usually that is the case, at least in Quebec.
That opens up a lot of potential applications where those reports could be used to help train other systems or to reduce the workload of the people who process those reports and so on.
The Chair: We have probably exhausted this issue in a general sense today, but certainly not in terms of the actual applications and implications.
Dr. Ferguson-Pell, we were fascinated a number of years ago when we were asked to review the health accord. We looked at health delivery across the country of some of the interesting telemedicine examples that were occurring in the West, centred in Winnipeg, Saskatoon and so on. Your illustration has added to the tremendous potential that exists there.
Ultimately, we know from a lot of observations that no matter where you are dealing with health, the best place to recover more quickly is in the home or at least in your local area. We know that technologies that are evolving have the enormous potential to help deal with that — to take the example you gave and to take it to the nurse practitioner dealing with it at the home, based on a diagnosis that has already been provided on the basis of information gathered and transported through a robotic tool or to a real practitioner who is interpreting that data as it is coming in.
I don't want you to respond to this today, but if you have any thoughts as you go forward, these will have enormous impacts. They will be very disruptive to a health care system that is already overburdened, largely because of lack of innovation and delivery of health care.
Particularly to you, Dr. Ferguson-Pell, if you have any thoughts about how we don't really have a system in Canada; we have a collection of Balkanized systems. How will we innovate to actually respond to the demand from citizens who will be able to have a pretty good idea that they need to have a response on the medical condition that their iPhone has told them they have to get to a centre based on that sort of thing.
I will leave it with you to think about.
Dr. Bengio, we would like to have any thoughts you might have with regard to the large-impact issues you have been dealing with on AI.
I think we have finally moved to something where — we are a laggard in international competition is true centres of excellence that bring in a broad-spectrum of interrelated activities. In the biotech area, we are not able to advance as rapidly because we were not able to get a collection of expertise in a centre where the applications would occur immediately. In order for the researchers to know exactly how they can benefit, they have to be in dialogue with people who have need of the use of knowledge. Again, to use the biotechnology example, in San Diego alone, there are more biotechnology companies than in all of Canada. You go out to lunch and you two are sitting at adjacent tables and spot one another. You say to Dr. Ferguson-Pell, "Look, we just observed this and we heard you give a talk the other day.'' That is how it spreads.
So I am hopeful in the AI area where we are putting serious money to bring collections together that we will move forward with true centres that will have the huge magnitude of impact and benefit for Canadians — but also in the larger sense, the economic and knowledge-based sense, from delivering products to the world benefiting Canada.
I want to thank you both very much for being here. If you have any further thoughts to communicate to our clerk, we would be delighted to hear from you.
Once again, thank you, and to my colleagues, I will see you tomorrow.
(The committee adjourned.)