Evidence of meeting #107 for Industry, Science and Technology in the 44th Parliament, 1st Session. (The original version is on Parliament’s site, as are the minutes.) The winning word was chair.
A recording is available from Parliament.
Prof. Andrew Clement
As the witness mentioned in his testimony, there's insufficient focus in this bill on curation, and curation of the training data is incredibly important. If, as he just said, you can't trace back where something has come from, that is a big problem in itself. The New York Times case is going to be a big one.
Conservative
Michelle Rempel Conservative Calgary Nose Hill, AB
Do you believe there's something this committee could do, or that the government should be doing, again to have a legislative imperative as opposed to just leaving this to interpretations by the courts, which are using laws that did not anticipate this type of technology?
Prof. Andrew Clement
Copyright is not an area of my expertise, but I would say that updating the Copyright Act and then linking that into this bill would be a way to go.
Conservative
Michelle Rempel Conservative Calgary Nose Hill, AB
[Inaudible—Editor] this bill to address the challenge that Canadians are facing with regard to AI regulation?
Conservative
Michelle Rempel Conservative Calgary Nose Hill, AB
Would you characterize that particular aspect as an example of how this legislation fails to adequately address or provide a proper and comprehensive regulatory framework to address some of the challenges that Canada is facing with regard to artificial intelligence governance?
Prof. Andrew Clement
I believe it is an area that is missing here or needs to be developed.
Conservative
Michelle Rempel Conservative Calgary Nose Hill, AB
Dr. Papernot and Dr. Clement, simply put, do you believe this legislation is up to the task, is adequate, salvageable?
Dr. Clement, you had strong feelings on this.
Prof. Andrew Clement
Yes, I made clear that I don't think this is suitable AI legislation. Whether it's salvageable depends on what your criteria are. There are clearly many areas where it could be improved.
Conservative
Michelle Rempel Conservative Calgary Nose Hill, AB
Do you also believe that Canada is at risk—because this is so inadequate—of promulgating regulatory systems that are put in place in either the EU or the United States through trade agreements like CUSMA and CETA? We haven't really thought about our own domestic frameworks before looking at how these provisions in these trade agreements could force us to promulgate the data ownership or AI regulations of other major jurisdictions.
Prof. Nicolas Papernot
In terms of the act, I don't think it's adequate in it's current form.
Conservative
Conservative
Michelle Rempel Conservative Calgary Nose Hill, AB
On effective accelerationism, would you say yes or no?
This is the mantra that's being espoused by Marc Andreessen and Yann LeCun that we should all leave AI to be self-regulated by industry and leave humanity to the vices of our benevolent AI overlords.
Prof. Nicolas Papernot
I think we both made it very clear that there is need for a consultation with more stakeholders in this process.
Conservative
Liberal
Liberal
Iqwinder Gaheer Liberal Mississauga—Malton, ON
Thank you, Chair. Thank you to all the witnesses for making time for this committee.
My first question is for Ms. Bednar.
In your opening testimony, you mentioned the term “algorithmic collusion”.
Could you elaborate on what that is, if you think it's already happening and how we are already seeing it?
Executive Director, Master of Public Policy in Digital Society Program, McMaster University, As an Individual
I will do my best.
This is an area where we have complementary tools and levers that would go alongside this legislation right through our Competition Act. We also have provisions in the Competition Act under false and misleading advertising, which I think could be a helpful tool for some of the deception that people see or experience when they are communicated with or receive the result of an algorithmically generated decision or result.
On algorithmic collusion, the question there is, when these systems are speaking to each other to either determine a price or in bidding—we see that Walmart and other companies are using increasingly algorithmic systems to negotiate with other algorithmic systems—is this a form of digital collusion that we would say is unacceptable or is this just an advancement in terms of our efficiency or our ability to negotiate rapidly to set prices or other outcomes?
It's not my main area. It's an area of interest and curiosity. It's something I've been researching. I'm happy to defer to other witnesses, too.
Liberal
Prof. Nicolas Papernot
The example I mentioned in my opening statement of the photocopy illustrates that when you have multiple AI systems interacting, it can eventually lead to very poor performance because they each respectively lose their ability to model the underlying phenomena.
We are currently studying this negative interaction in our lab. It's not going to help people who are already at risk of being harmed by AI.
Liberal
Iqwinder Gaheer Liberal Mississauga—Malton, ON
Is this already happening?
I'm a lawyer. I'm not a competition lawyer. A simple example is when two companies sort of collude to set the price of bread, for example. When AI does this, is there any evidence? Is there a trail of evidence that this collusion has taken place or not?
