Amir Feizpour is a Senior Manager at the Enterprise Data Science team at the Royal Bank of Canada. He is also a Scientific Advisor at SEMA Lab. At RBC, Amir manages a small team with a focus on enabling internal clients with data science tools and practices. Prior to this, he held a post-doctoral position at the University of Oxford conducting research on experimental quantum computing. Amir holds a Ph.D. in Physics from the University of Toronto. He shared his thoughts with RDP Blog on the state of Canadian AI.
_______________________

You’ve stated that practitioners of data science are puzzled by their fuzzy job descriptions and that there is a sense of frustration in the industry about the lack of standard definition of who a data scientist is. Can you explain why?

There is a big hype around data science right now, both from the viewpoint of the practitioners and businesses. On one hand, businesses are trying to become data-driven without taking the appropriate steps in most cases. On the other hand, practitioners with various backgrounds take a few online courses and think that they have become data scientists. This is a very young field and most people entering it are not specifically trained to be data scientists, but rather are good problem solvers from other fields who are looking for new challenges.

A significant number of people who seek my advice for entering the field ask me about the job ads they have seen pointing out that the ads are asking for every possible thing in IT and computer science. They feel discouraged and confused about what is really important. And then, when they are hired as data scientists, they are mistaken with many other data-related or even IT roles, especially since the businesses haven’t taken the steps to be ready to onboard data scientists.

Harvard and the MIT are jointly offering courses on the ethics and regulation of artificial intelligence. As machines are taught to mimic human thinking and carry out various tasks, can ethics and morality be injected into these tasks, and do you think everyone out there really cares about these issues?

That is a very difficult question to answer because I’m not convinced that humans are that great at morality and ethics. Regardless, I guess humans are the best examples that we have to work with. The foremost action is really to define objectively what morality is and what can be considered fair and just. It doesn’t take too long of a research to realize how difficult this task is.

Other than some serious main issues, most of what we consider moral behavior are completely context dependent, so defining morality metrics that are measurable might prove to be daunting. I know that there are some people who are doing academic research on this topic, but a quantifiable objective measure of ethics is what we need before even being able to teach machines to be moral. One might try to rely on humans to achieve this by crowdsourcing moral decisions to create datasets that machines can be trained on. But experience has shown that removing biases from those datasets might not be exactly straightforward, but perhaps the best we can do right now.

Finally, it’s very important to emphasize the use of transparent decision-making algorithms. Having black-box algorithms that make important decisions can obfuscate our ability to verify the logic and therefore could be a recipe for potentially unfair situations.

Massive supplies of data appear to be the lifeblood of AI applications. How can regulations around access to data impact business viability and profitability for Canadian AI companies, whereas jurisdictions with weak regulations offer an abundant supply of data to their AI companies?

The challenge is that both strong and weak regulations are damaging. Strong ones could hinder progress, and weak ones could provide environments for low-quality and shady activities. However, there is a huge gray area in between the two extremes that we are still exploring and trying to figure out where the right “line in the sand” falls.

Both industry and academia are trying to understand this and concepts like algorithmic bias and accountability, differential privacy, and mathematical definitions of fairness are popping out which are still in their infancy stage, but certainly good starting points.

At the end of the day, what is important is how smart our regulations are rather than how many of them we have or how strong they are.

Because entrepreneurs and researchers will have the opportunity to find out ways to address the concerns and still make good progress. This could also encourage collecting and using higher quality data that is more important than the quantity of data. Finally, consumers are likely to pick companies for their needs that are (at least perceived to be) more reliable in terms of morality, ethics and transparency as we have seen in other industries.

You work as a data scientist in one of Canada’s major financial institutions. How do you see AI, machine learning (ML) and the onslaught of fintech impacting financial services in Canada? How well-prepared are incumbent banks?

There is some good work going on in large organizations like RBC. It is very clear at this point that financial institutions cannot continue to be only banks, and they need to transform themselves into technology companies and ecosystem platforms. The danger is not coming only from fintech companies, but also from Amazons and Alibabas of the world. The leaders of large financial institutions realize the importance of this and are investing money and effort in transforming the way they do business. Acquisition of Layer 6 by TD, investment in fundamental deep learning research at RBC’s Borealis AI, and many more examples show that financial institutions are serious about the transformation and the competition.

How can the Superclusters Initiative—Canada’s latest policy blueprint for innovation—help Canadian AI and ML companies with commercializing their products?

This is a very valuable step for the AI ecosystem in Canada. There are so many parties including research institutes, public sector and government, and companies of various sizes in private sector coming together to accelerate our progress in AI. This could lead to the development of a very fertile ground for cutting-edge work that is done in academia to find its way into the industry.

The current focus of the supercluster (SCALE.AI) is specifically on retail, manufacturing, and infrastructure, which comprise a large part of the Canadian economy. The goal is going to be digital transformation and enhancement of supply chain in those areas using AI-powered solutions. This is a great starting point and I am looking forward to seeing how it will pan out in future and in practice.

There are industry observers and practitioners who believe Canada needs billion-plus dollar companies to become globally relevant in AI and to attract top-notch talent. Do you concur with them?

I could see that as something that can be quite attractive, though I don’t think that would be sufficient. I think the whole ecosystem, beyond academia, needs to be improved. Maybe having billion-plus-dollar companies would also be an outcome of such an improvement and therefore correlated with being globally relevant, and subsequently retaining AI talent.

From an academic point of view, Canada is the birthplace of many revolutionary ideas in modern AI. However, what’s really missing is the translation of these academic achievements into successful large tech enterprises. We do produce a great deal of top-notch talent in academic settings. But then we lose them to American companies because of a combination of having more options and freedom, significantly better compensations, and perhaps more importantly, the ability to work for companies that are at the cutting edge of technology when it comes to AI.

Do you think government incentives, such as grants, tax credits for R&D, or hiring, are essential to encourage more Canadian companies to implement AI and ML on a larger scale and with greater confidence?

Definitely. The government can play a big role in this. Although we are losing a significant number of our AI talent, we are still in no shortage of talent in the market. There are people who are very driven and want to make a difference. More importantly, a lot of them would really prefer to pursue a career in AI here in Canada and have no choice but to go to the US because of the abundance of funding among other considerations. So, anything that we can do financially to support the existing ecosystem would significantly impact the market and help us move forward more confidently.

This interview was conducted by Reza Akhlaghi, a content marketing manager with RDP Associates.