Week 30 — Can you teach a machine a new trick?

Scott McNaughton
6 min readOct 4, 2019


What happens when you want a machine to do something you don’t know how to do?

There are upsides and downsides to being on the cutting edge of government innovation. On one hand, you get to do really cool things and get to push the boundaries and art of the possible in what is typically seen as a by the books kind of place. On the other hand, the cutting edge meets the way things are and those two realities begin to clash. The realities of an institution built for a different era which has incrementally changed over the years (often for good reason) clashes against a world that wants to evolve constantly. It’s an interesting dynamic to watch from my vantage point leading a project that seeks to disrupt. What is the key takeaway? Whether we see it or not, things are changing but maybe not as fast as we would like. As someone looking to push things forward faster than what would naturally happen without said push the best piece of advice I can offer is be patient, pick your battles and create a situation where the incentives favour what you want to do (e.g. what’s the worst that could happen/if this works what is the upside). Easier said than done but being the change driver in an organization is like going to the gym. You do it long enough and you see the gains but if you stop you can lose your gains quickly!

Picture unrelated

AI Demonstrator Projects (Incorporation by Reference and Regulatory Evaluation Platform)

This week has been a good week of moving files ahead. I’ll try to focus on a handful of items of interest but some I want to save for future posts to do deep dives.

Regulatory Evaluation Platform: The inspiration for this week’s title came from meetings we had on the Regulatory Evaluation Platform project. The challenge is that we are attempting to build an AI system to do something we are not currently doing. When you are trying to build something to replace an existing function that means you have documented processes or experts who can outline what they do to accomplish the task. In the case of this project, we have not undertaken a number of the tasks we want AI to do. This is creating an interesting situation and thought exercise about how we can use AI to do something we don’t do. This becomes problematic as you realize that the lack of doing a task means there is no data about what a good and bad outcome is. In a supervised learning model, how would you know whether the model is working? If you don’t know the result because you don’t know the process and you don’t know what is right and wrong it creates a number of challenges.

Despite this shortcoming, we have begun to unpack these tasks by challenging our subject matter experts to think in terms of “what would I do if I was forced to do this”. It remains to be seen if you can train a machine a new trick. In the meantime, it’s been an interesting exercise to stretch people’s thinking beyond “I don’t know because I don’t do it” to “let’s work through this and think how I would do it if I had to do it”. From a facilitation point of view, the biggest lesson learned is around how we frame questions and help people work through problems. Re-framing the ask to move away from what do you do today (we don’t do it) to what would you do if you had to do it (here’s how I would do it) has been powerful and fruitful.

We are also starting to have discussions around how to do a “peer review” of the AI system. As per the Directive on Automated Decision Making and the Algorithmic Impact Assessment, we are pretending (for sake of being a pathfinder project) that we are scoring as a level 2 project (the truth is we are level 1). One of the requirements of a level 2 project is the need to conduct a peer review. Being the first out of the gate is daunting but valuable. We have started to define what a peer review is, consulted experts to help and are close to locking down our next steps. Being the first also means learning the hard way how to do one which I hope will be insightful for others who come next.

Incorporation by Reference: We had a good discussion with the Canada Standards Authority today about accessing critical data for this project. As I reflect on this conversation, it is amazing to realize how open and collaborative outside organizations can be especially as they realize the ambition of what you are trying to do and how that will benefit many people. With movement on this item, the project will be that much better.

I’m saving for next week a discussion about procurement for AI so stay tuned. Debated doing it this week but I haven’t had enough time to put thought into words. I did present on the topic so the topic is fresh in my mind so I will take the next week to put together some interesting thoughts and reflections.

Rebuilding the Public Service From The Ground Up: Week 17

Week 17 already? You would think after 17 weeks I would be running out of ideas to talk about but you are wrong. For week 17, let’s talk about the future of where we work. Yes, inspired by a visit to the GC Co-Working Space, here is a thought about re-framing how we think about the space we spend a lot of time of our adult life: our office.

Idea 17: User Designed Modular and Flexible Workspaces

The topic of working space (cubes, offices, home, coffee shop whatever) has been something on my mind for awhile. The general acceptance of co-working, working from home, teleworking whatever you want to call it has seen a dramatic increase in the GC. This is reflected in the shift to Activity Based Workplaces and the opening of GC Co-Working Spaces. I’m sure many other changes are notable at the more local level as well.

So here’s the problem, open workspace offices do not work. The research is clear on this and countless workers can attest to this (at least privately). The old joke is that government is always behind on the trends and that appears to be the case here. However, I would argue that the more traditional private, high walled cubicle built using visual decoration from the 1960s has to go. It’s drab, uninspiring and leaves a bad taste in everyone’s mouth. So what’s the alterative? Here’s an idea: build workspaces around the needs of teams. Invest in modular and flexible office equipment that is movable and can change depending on needs.

How does this look in practice? Assign each team a certain amount of space (100 square feet for example). Then ask them to define how they work. Do they do heads down focused work? Do they yell at each other over cubicles? Do they need to work with their colleagues on a regular basis? Are they working with sensitive files? Are they on the phone a lot? Based on these needs, show them what kind of movable and flexible furniture and equipment would suit their needs. Are they heads down all the time? Then give them a lot of private quiet space. Do they collaborate a lot? Then give them shared furniture where they can huddle on an ad-hoc basis.

Cost wise, it makes sense to try to build a “one size fits all” let’s squeeze everyone into the same configuration style work space. In reality, teams need to lead how their workspace is set-up so they are set-up for success.

Thanks for reading and I’ll see you next week!



Scott McNaughton

Working on public sector innovation one problem at a time. Found biking and hiking on weekends. Father of young baby… what is sleep?