The Black Box Problem (and beyond)
Welcome to the 22nd issue of the Digital Policy Salon weekly briefing.
What is the difference between AI R&D and successful AI application? This week, we dive into the complex world of effective, ethical, and practical applications of artificial intelligence, with featured research on AI in the forestry sector and its vulnerability to aggressive patenting, as well as two “what we’re reading” pieces on AI applications in medicine and predicting individual life outcomes. One of the organizations at the forefront of industry coalitions on ethical AI, Silicon Valley’s Partnership on AI, is featured in this issue’s Tech & Human Rights series interview.
In addition, our weekly perspective piece provides an overview of a timely and in-demand workforce: experienced cybersecurity professionals remain difficult to find in Canada, but workforce development solutions are waiting in the wings. Finally, our policy update highlights a high-priority topic for many Canadians: how is each province handling “back to school” plans this fall?
Thanks for joining us, and we encourage you to leave us feedback on what you’d like to see more of in the Digital Policy Salon at the bottom of this issue.
- Faun, Khiran, and Tyler
COVID-19 Policy Updates 🇨🇦
Just weeks away from the 2020–2021 school year, what do post-COVID-19 education plans look like in Canada’s four biggest provinces?
While all provinces have adopted new health safety rules focused on class size, social distancing, and PPE, plans for online and in-person course delivery vary significantly across jurisdictions.
Ontario
Ontario schools will be moving forward with a hybrid approach. Conventional, in-person classes will be offered to students kindergarten to grade 8, however parents may choose to opt out of in-person delivery to pursue online learning instead. For secondary students, at least 50% of courses will be delivered through in-person instruction, supplemented by up to 50% online courses.
Parents have expressed concern about the quality of online learning and the potential for additional labour for parents who choose that route. A recent survey of the Ottawa district school board showed that approximately 24% of Ottawa students have opted to learn online.
Québec
In Québec, students from Preschool to Secondary III will be required to return to school for in-person learning; while students Secondary IV to V will attend at-least 50% of their courses in-person, with complementary online learning taking place at home.
British Columbia
Much like other years, students registered in “brick and mortar” schools across British Columbia will be expected to attend their classes in person. Elementary and middle school students will only be provided daily online learning options if they are specifically registered in an online or distributed learning school, and secondary students will only learn online when its needed to accommodate new class sizes.
The province has also announced an additional $3 million in education funding to acquire new remote learning tools and other tech.
Alberta
With perhaps the least number of changes to its regular school programming, Alberta will move forward with a “near normal” school year. All students aged Kindergarten to grade 12 will attend in-person classes while following new COVID-19 health measures. - Mairead Matthews | email
Our Perspective
Searching for Hidden Talent | Overview
Searching for Hidden Talent: Experience and Expertise in New Brunswick’s Cybersecurity Community explores the cybersecurity ecosystem in New Brunswick — a significant centre of cybersecurity activity in Canada — through the lens of cybersecurity labour demand and supply.
Study Context
Cybersecurity talent is in short supply in the global digital economy. By 2022, a worldwide cybersecurity labour shortage of 1.8 million workers is expected, with a corresponding North American shortfall of 265,000 people.
The demand for cybersecurity talent in New Brunswick is confirmed by the study’s key informant interviews and job postings:
- 67% of the study respondents expected to expand their cybersecurity workforce in the next year
- New Brunswick has a high volume of cybersecurity job openings compared to its population
General Study Findings
The cybersecurity industry in New Brunswick continues to gain international recognition as a well-networked and collaborative ecosystem, which is expected to continue its expansion.
The province’s cybersecurity ecosystem offers the following advantages and disadvantages:
Advantages:
- Public sector investment
- Large industry players
- High quality of life
Disadvantages:
- Remoteness
- High unemployment
In the study, industry experts praised New Brunswick’s dedicated organizations that play a hands-on role in workforce development:
- Provincial colleges are receptive to industry feedback and agile in creating relevant digital tech curricula and innovative training, including internships and co-op student work placements.
- Current academic efforts are graduating sufficient entry-level cybersecurity workers.
- However, senior-level cybersecurity roles in the province are difficult to fill.
Interviews in the Field
The Partnership on AI: A Multi-Stakeholder Coalition for Responsible Technology
Katya Klinova is Program Lead at the Partnership on AI (PAI), a coalition of over 100 organizations from civil society, industry, and academia, where her work focuses on AI, Labour, and the Economy. Prior to joining PAI, Katya was at the Harvard Kennedy School of Government, researching the potential impact of AI advancement on economic growth trajectories of developing countries. Previously, she worked at the United Nations Executive Office of the Secretary-General (SG) to prepare the launch of the SG’s Strategy for New Technology, and at Google in a variety of roles.
Kiera:
AI, as it advances, will influence the nature of work in Canada and globally. AI advances can inject great value into the economy, but they can also cause disruptions as new kinds of work are created and others become less needed. Can you highlight the most pressing disruptions and/or issues relating to AI and labour at the moment in Canada, the US, and/or globally?
Katya:
I group the impact uncertainties into two buckets of questions: one question around labour demand and another question around quality of jobs. On the former, the question is whether the labour demand is going to go down for certain groups, and for whom? From historical experience, technology usually automates some tasks but also creates new tasks, but the question is: do these processes balance out?
Research by Acemoglu and Restrepo for the US shows that in the decades following World War II, automation and task-creation balanced out nicely. But that balance has tipped toward automation that has been accelerating, while the creation of new tasks has slowed down in the past three to four decades. But we also need to ask: “Who are these tasks for? What tasks are we automating, and whom are we taking them away from?” If we are automating tasks that don’t require college degrees and only creating tasks that require select college or graduate degrees, then we are creating a skills bias. This is the kind of technological change that advantages the highly skilled and those with a lot of educational attainment while disadvantaging people who did not have the resources to acquire that education. Here then, educational efforts become even more important, and we have to be realistic about how quickly we can ramp those educational changes up. For whom is upskilling or retraining available? How flexible is the labour market? If we make changes faster than people can adapt to them, or if people do not have the resources to adapt, it will be a difficult transition for entire groups within society, independent of whether they are in developing or developed countries.
The second bucket of questions is around the quality of jobs. For example, quality in data-labelling jobs is an issue as companies rely more and more on contingent, temporary workforces. It’s beneficial for them to bring in a workforce only when they need them, but at the same time, our societal structures are not set up to support workers involved in those kinds of work. Where worker benefits, healthcare, and pensions are tied to an employer, it is very important to be a full-time employee, so the question of portable benefits and other support structures for crowd platform workers is very important and needs to be addressed.
Read the full interview here 📖
What We’re Reading
Facebook’s AI can generate MRI images in minutes instead of an hour
(Engadget)
Teams from Facebook AI and NYU Langone Health have developed a neural network that can cut the amount of time people have to spend in an MRI machine from more than an hour to just a few minutes.
The network, dubbed fastMRI, shortens the scanning time because it only requires a quarter as much data to resolve the image.
Talking points:
Gathering only a portion of data that traditional MRIs gather, the technology highlighted in this article then fills in the missing data using AI, meaning that MRIs can be performed in a fraction of the time. This example very clearly frames when AI is a beneficial technology: whenever it is easier to predict additional data than it is to gather additional data. The technology hasn’t convinced everyone yet — the article notes that one out of the six radiologists that examined AI and non-AI MRIs were able to note differences. - Khiran O’Neill | email
Prediction, Machine Learning, and Individual Lives: an Interview with Matthew Salganik
(Harvard Data Science Review)
Machine learning techniques are increasingly used throughout society to predict individual’s life outcomes. However, research published in the Proceedings of the National Academy of Sciences raises questions about the accuracy of these predictions. Led by researchers at Princeton University, this mass collaboration involved 160 teams of data and social scientists building statistical and machine learning models to predict six life outcomes for children, parents, and families. They found that none of the teams could make very accurate predictions, despite using advanced techniques and having access to a rich dataset.
Talking Points:
“Black box” methods (where the inputs and outputs of an AI model cannot be understood or explained by a human) are compared with explainable ML and traditional four-variable regression in this interview. The findings? There was hardly any difference between these methods’ effectiveness for predicting individual life outcomes. While black box AI has been shown to work well for technologies like computer vision, here it underperforms significantly in predicting individual human behaviour and metrics for social success such as high-school graduation or GPA. This challenge marks an important contribution to the ongoing debate about the ability of AI, explainable or no, to make meaningful and ethical decisions about individual livelihoods and futures. - Faun Rice | email
Research Visualized
As part of our report Searching for Hidden Talent: Experience and Expertise in New Brunswick’s Cybersecurity Community, we looked at cybersecurity job posting data by province. This graphic compares the percentage share of jobs posted in cybersecurity versus the percentage share of Canada’s population, by province. The biggest takeaway, of course, is the size of Ontario’s cybersecurity job market. It shouldn’t be missed, however, that the per-capita numbers in Atlantic Canada are also ahead of those in the rest of the country.
Our Research
Artificial Intelligence and the Forestry Sector
Technological and research breakthroughs in the field of Artificial Intelligence (AI) have put it on the brink of revolutionizing almost every industrial sector. It is important to not only consider its applicability in certain domains but also understand when it is not required, in the presence of existing, well-defined computational techniques.
Developing AI systems requires access to high-volume, high-quality datasets, and a key complement to this disruption will be the network infrastructure of the Internet-of-Things (IoT), notably 5G deployment. The forestry sector is an example of an industry that can leverage this combination, where data collection and analytics will be crucial to effective operations and supply chain management.
Twitter Highlights
Talk to Us 💬
Send your comments, questions, and tech policy insights to:
- Tyler Farmer 🚀 t.farmer@ictc-ctic.ca
- Faun Rice 🦌 f.rice@ictc-ctic.ca
