44th Parl. 1st Sess.
November 20, 2023
- 12:19:45 p.m.
- Watch
- Read Aloud
Good morning, and thank you very much.
My name is Théo Lepage‑Richer, and I'm a post-doctoral researcher at the University of Toronto.
First, I want to thank you for the opportunity to share a few thoughts with you today. These are the product of my research on artificial intelligence governance, a topic I address by combining historical research with public policy analysis.
In previous meetings, several members of the committee raised the following question: how can we develop governance frameworks adapted to technologies that are evolving as quickly as artificial intelligence? This is indeed a legitimate issue that is regularly raised by the providers of this technology to encourage some restraint by public policy-makers. However, I'd like to qualify this question by pointing out the broader trends that the history of artificial intelligence in Canada highlight.
The first federal AI programs provide a useful historical precedent to examine the impact of this technology on the organization of work.
Starting in the 1960s, the Pearson government identified artificial intelligence as a promising technology to reduce the costs associated with hiring qualified public servants.
In 1965, the National Research Council of Canada was mandated to develop a first artificial intelligence program to address the translation of official documents from English to French. As a strategy, program managers opted for the development of software tools that would allow the translation process to be broken down into simple sub-tasks. One of those tools, for example, was designed to produce literal translations of common names and verbs in a text, with the idea that operators would then take care of them by fine tuning them, adding the necessary determiners and revising everything. The purpose of these tools was to standardize specialized tasks such as translation, to the point where they could be assigned to workers without prior training, and, above all, at a lower level on the pay scale.
Although inconclusive, this program launched a series of reforms aimed at reducing the federal government's dependence on skilled workers and, above all, restoring a certain level of control over the federal machinery.
Under Pierre Elliott Trudeau, initiatives such as the CANUNET network and the Télidon system were put in place to create the necessary infrastructure to produce new data on the work of federal employees. In a recent article published by Fenwick McKelvey and myself, we suggest that the objective of these programs is to quantify the work of public servants so that it can be framed more narrowly using new data analysis tools developed in government and elsewhere.
Fifty years later, the applications of artificial intelligence in the Government of Canada and elsewhere have changed. However, there are early warning signs of broader trends that can be identified in these early programs. Rather than completely replacing positions, artificial intelligence tends to be deployed in such a way as to restructure tasks so that they are assigned to workers with more precarious status, limit the opportunities that workers have to exercise their judgment, reduce the dependence of organizations on certain forms of expertise and replace investments in training and workforce development.
These trends go beyond artificial intelligence, of course. However, as Paola Tubaro and her colleagues point out, these trends nevertheless tend to characterize the platforms, management practices, reforms and business models that depend on the deployment of this technology.
As such, it is therefore urgent that the impact of artificial intelligence on the workforce become a key perspective for developing tailored policy responses. This position is shared by a number of people, including Emanuel Moss and Valerio De Stefano, who point out the inability of the risk-based approach that characterizes the current regulatory instruments to account for issues related to worker protection. To reflect the impact of artificial intelligence on the workforce, these instruments would have to take into account the impact of this technology on the distribution of wealth, the quality of jobs and the loss of salaried jobs to precarious or subcontracting positions.
Until now, artificial intelligence has been perceived in Canada as an industrial policy issue, and not without success. However, it is crucial that investments in the AI industry complement, rather than replace, similar investments in human capital.
While future applications of AI are difficult to predict, the structural effects of AI on the organization of work remain stable and can therefore inform policy responses to the test of technological change.
Artificial intelligence is a challenge both in labour law and in industrial policy. I therefore encourage the members of this committee to consider the trends in which artificial intelligence has been embedded over the past 60 years to put in place the necessary safeguards to ensure that workers also benefit from the deployment of this technology.
Thank you very much.
803 words
- Hear!
- Rabble!
- add
- star_border
- share
- 12:42:14 p.m.
- Watch
- Read Aloud
That's a very good question. I can't think of an international example off the top of my head.
In fact, we have to think about the hidden costs that are often associated with artificial intelligence. When you interact with a platform like ChatGPT, the human work behind it tends to be somewhat erased. However, behind a system like ChatGPT and all the other artificial intelligence systems that are trained using large amounts of data, humans have to work to label that data, format it and organize it, among other things, which is not a well-paid job.
There are a lot of countries, especially emerging countries, that are going to train a whole workforce to do these tasks at a very low cost. The example that comes to mind is not necessarily an example that Canada wants to emulate because, when we talk about low-paying jobs associated with artificial intelligence, we can think about data labelling. This work is essential to all the artificial intelligence systems we use and develop, but it depends on thousands of workers who manually crunch data for pennies. The international examples that come to mind are not necessarily examples that we want to emulate, but that we want to keep in mind to remember the social and human costs associated with the development of these technologies.
224 words
- Hear!
- Rabble!
- add
- star_border
- share
- 12:45:07 p.m.
- Watch
- Read Aloud
The approach currently used in Canada to assess and anticipate risks and the impact of this technology is mainly based on self-assessment. The proposed artificial intelligence and data act promotes the idea that we must create a model so that businesses can govern themselves by taking certain parameters into account, while making sure that the effects on their work are as limited as possible.
One of the problems I see with this approach is that AI is deployed in a very wide variety of sectors. Therefore, at some point, these tools need to be tailored to each sector and industry in which AI is deployed. This will allow us to properly represent the reality of workers and users whose quality of life, work and well-being are directly influenced by this technology.
One of the first ideas that comes to mind is that risk assessment tools should be set aside for different industries. In fact, at all levels of government, there are specific frameworks to assess the environmental, financial, social or human impact. However, we do not see the same degree of precision in the evaluation of this technology when it is deployed.
Off the top of my head, I would say that we need a more specific development of analytical tools.
213 words
- Hear!
- Rabble!
- add
- star_border
- share
- 12:47:23 p.m.
- Watch
- Read Aloud
Right off the bat, I welcome your comment with enthusiasm.
That's more or less the strategy that Europe has adopted. The European model relies a lot on independent or semi-independent committees to assess the impact of the deployment of this technology.
However, I wonder to what extent this approach would be realistic in Canada. I'm thinking of the size of the European government and public compared to that of the Canadian government and public. Realistically, although I'm excited about your comments, I'm wondering to what extent the Canadian government could implement a such an evaluation model. That's why developing analytical tools that are better adapted to the various industries and sectors seems to me to be a realistic compromise in the Canadian context. I'm not hiding my preference for possible solutions.
138 words
- Hear!
- Rabble!
- add
- star_border
- share
- menumenu
- notificationsnotifications
- home
- mailmail
- searchsearch
