Language selection


Artificial Intelligence for Insights into Regulations (DDN2-V15)


This video features AI experts and senior federal officials, who demonstrate artificial intelligence methods for the review and analysis of federal regulations.

Duration: 00:07:28
Published: October 2018
Type: Video

Event: Artificial Intelligence for Insights into Regulations

Now playing

Artificial Intelligence for Insights into Regulations

Transcript | Watch on YouTube


Transcript: Artificial Intelligence for Insights into Regulations

Budget 2018 announced that the Government of Canada would be doing a review of its regulatory stock and this was the first time to our knowledge that we've ever done a review of our regulatory stock. So 2700 regulations, some of them are this big, some of them are that big, that's going to be a lot of paper. That's going to be a lot of review. Why don't I flip this around. Why don't I say here's our data, tell us what you can do for us. Show us what you can do with it. The proposal was essentially, we weren't going to pick one winner. Here's our problem. Come with your ideas. We’ll shortlist a bunch of you and then we'll keep working with you. We will give you access to our regulators. So this is a really, really exciting moment, because while artificial intelligence is not new, it’s one of the very first times that artificial intelligence is being used in the policy world.

The opportunity for applying AI to regulation is broad, because at this state of where we are, AI is a very powerful tool for taking data complexity, large data, complex environments and providing insights. And the insights, in fact, are particularly useful when it comes to identifying trends, patterns, and also predictions.

We're also talking about AI in this context, as informing human judgment rather than replacing it. So just so we're clear, this is about insights for regulatory policy analysts, as they think about their regulations. So when they're doing their stock review of regulations, as the Cabinet directive asks them to do, these are about tools to help regulatory policy analysts do their jobs better.

We actually use an unsupervised learning approach. We're able to bring all this data together and organize it and essentially fingerprint the data and have it self-sort. Now we can actually, you can do a query and in a couple of seconds, it'll get back all the regulations that could overlap. An administrative baseline was not available, but we provided one. So the interesting thing of course is now there's a metric, or something that can be consistently applied, because those administrative baseline burdens were done by human beings. Now you can measure it, doing it consistently, with the data processing pipeline.

We grouped all the regulations we were given into clusters, using machine learning. We rapidly filter regulations by names, clusters, degree of, and trends in prescriptivity, and outdated technology, and even, we’ll try to interact with a chatbot.

We took our U.S.-based Regulatory Explorer tool, and we Canadianized it. We applied the same types of artificial intelligence and analytics to that set of regulations, to create a set of insights from, from a Canadian perspective.

So we decided let's give a tool, using artificial intelligence, that allows you to get information that you can after digest and use, in order to do things like modernizing of them, and automate the process of the update of the regulations. You can see the complexity of these networks.

So we've developed a supervised classifier, with thousands and thousands of training documents going into it, that let us assess a probability that a given regulation is targeting a given industry or sector of the economy. In Canada we're seeing restrictions at the provincial level, in some cases rival or even surpass how many we're seeing at the federal level.

We analyzed the regulatory impact assessment statements that accompany any regulatory change in Canada. The presence of constraint in regulatory activity, over time, is extremely constant, while it varies enormously from one topic to the next.

We actually read regulations and acts and say, is there something in the text that'll say if this will be challenged or not. We actually were able to correctly predict, just solely based on the text of the act, using a very simple procedure, 77% of challenged acts were correctly predicted. And we can actually show why, which is more interesting.

You don't have to go it alone. As you're building into this technology, please tap into the resources that are available. There's over 1100 PhDs that are studying AI at universities. They want to solve these challenges. They want real-world expertise. They want to stay in Canada. So by giving them the access to the really exciting challenges that exist in government departments as well as in industry, I think will have a really great impact in the community.

The public sector wants to become a digital sector and I think that's really where we're starting to see a tremendous amount of leadership, and we're seeing a lot of excitement coming from the private sector to help us out. And I think the technology leaders that are here today, and those that are trying to really get their great ideas and their technology to help us make a difference, we're now receiving that, and we're asking for that partnership. And that's pretty game-changing. How do we make sure it still follows public interest. How do we make sure we're still protecting citizen data. And how do we make sure, as we work with our partners in the technology sector, that they're also thinking about that problem.

We do need to start in small bite-size pieces, right, so pick a few data sets, let's create some automation around them, let's share them, let's access them, let's leverage the technology available to us, let's do some, you know, really good sort of algorithmic insights and see what happens.

How do we actually surface the biases that are already implicit in the system, so that we can identify them and address them. And I think AI and machine learning are tools in doing that. These algorithms are going to be hugely powerful in refining our ability to do the work, but we need to monitor them. We need to have humans in the loop. We need to be making sure that they're doing what they ought to be doing, but it's still a much easier lift for humans, and those machines can do things that we just can't do as humans. It’s going to be the same thing with artificial intelligence.

So what a great morning for the federal participants in the room, especially the regulators, and also I think some of the policy makers, and I've seen a few policy makers in the room. You have seen the future, I think, in different venues today. The capacity with which you can glean insights, in terms of the current stock, is breathtaking. But that's just step one. Once you have those insights, then it's what do you do with them, and what do you do with them in terms of revising current regulations, and what do you do with them in terms of how you write the next generations of regulations.

Related links

Date modified: