Michael Milford: Positioning the future

And was it an ah-ha moment back when you were undergrad, or pre-study, that you decided this was going to be your career?

I’ve been interested in aspects of robotics and AI for a long time, ever since high school. I liked science fiction, I saw the movies … the Terminator films were some of my favourite films, and still are.

But I’ve been inspired in a few different areas. I think toward the end of my undergraduate degree, I’d been doing a lot of programming, a lot of AI-related stuff. I actually had to write a Rhodes Scholarship application about what my big vision was, and what I see sort of the biggest challenges to overcome and solve would be, and that’s when I sort of really fell into the robotics and AI areas. That then led me on to a PhD, and from there I’ve never looked back.

When was that Michael, what’s the timeframe you’re talking about?

That was oh, showing how old I am … my last year of undergraduate was 2002, and I started my PhD in 2003.

I’m partly comforted that Terminator was your inspiration, and that you know full well the robotic and AI future we don’t want!

Yes, yes. I have pictures of Terminators in my office, but they’re intentionally not carrying guns, so they could theoretically be peaceful, or protective good Terminators.

Excellent, so more like the latter films in the series, when Arnie was less gun-ho!

Yeah, the friendly Arnie who only shot you in the kneecaps, something like that.

I see that you’ve studied overseas too. Other than QUT, where have you been, and what were you doing?

In terms of my core career, I spent 11 years first as a student, then as a PhD, and then as a researcher at the University of Queensland.

During my time at QUT, I’ve done a couple of short stints overseas. I spent six months in Boston working collaboratively with Harvard and Boston University. And I recently spent a shorter sabbatical in the UK, in Scotland and then in London, working with Edinburgh University and Imperial College London.

All right, now moving back away from you for a little while, and into the world of hypotheticals … you know of a problem that you’d like to fix. Someone’s given you a big bucket of money to do it, and a reasonable timeframe. What is it you’d like to attack?

I’m very interested in autonomous vehicles of all descriptions, and there’s really two reasons for that. One is if we do a really good job and we solve all the technological problems and we implement them properly and it all goes really well, obviously we can radically change how cities are designed, we can save a lot of lives, we can hopefully improve quality of life for a lot of people, especially people who are marginalised by current transport technology. That’s one reason.

The more scientific part of the problem is I think autonomous vehicles, and the research being done in terms of the AI that enables them to make decisions. I think we have the opportunity to make the biggest advances in artificial intelligence research in the domain of autonomous vehicles. It’s a beautiful area to explore these problems of intelligence, because it combines … it’s not trying to solve general intelligence. This is a somewhat constrained problem of how to drive robustly and safely on the roads, but it’s rich and complex enough, and we haven’t solved it of course yet, that you would be able to make a lot of advances I think in AI.

Of that switch to connected and automated vehicles, if we wanted to bring it on more quickly than perhaps it’s moving now, what do you think would need the biggest injection of resources?

I think what we’ve seen not only in this general field of autonomous vehicles, but also just AI and robotics in general, is that conventional methods have been … the most recent sort of trench of conventional methods have been optimised and refined, and they’re really, really good, but they may not be able to be further refined to get them to a 99.999% reliability level.

So I think people are looking at stepping back a little bit, and trying to sort of rethink how we do some of this stuff. One particular example, this is in AI, are the machine learning algorithms that control the smarts in autonomous vehicles and robots. People in some areas are increasingly looking towards biology for inspiration about how they can improve their systems, so looking towards how the brain does it and trying to mirror some of that in software.

What do you think will be the hardest part of the switch to autonomous vehicles? Will it be dealing with the changeover from mixing legacy vehicles with part automated vehicles on the road, or running full autonomous environment? Which do you think might be a more troublesome time along that path?

I think the most difficult part of all of this is solving the core technological problems. If you’re thinking in a longer timescale context, things like the industrial revolution were incredibly disruptive over the course of a generation. But in retrospect, we’ve perhaps lost a sense of just how disruptive technological change in that period in history was. But we got through it, and the modern world owes a lot of what we have to that revolution.

Yes, there will be significant, very disruptive, very confronting changes that we’ll have to manage well. But I think really the core problem still is can we make the technology capable enough that that’s actually a choice we get to make?

History in the rear-view mirror can somewhat diminish the sense of disruption.

Yes. It’s important not to use it to sort of gloss over the incredible challenges that still remain, but it does give us some sort of precedent and some inspiration.

Indeed. All right, and part two of this hypothetical scenario. This time, you’re to fix a problem, and have an appreciable impact with a very small budget, and quite a strict, short timeframe. What would you like to go at?

You have to be pragmatic in these cases, and look at fields where a lot of the framework, the research infrastructure, and the ability to test and deploy a system is already there. So I’d actually stay in the same area as my big dollar project answer. I’d say autonomous cars and safety systems for cars. But this time perhaps not trying to solve the entire automation problem, but rather trying to solve one tiny little aspect.

Perhaps how these cars detect people in the world around them, and making them rock-solid safe so that they never, ever, ever hit someone. So I think I’d like to carve off a little niche, one particular important sub-problem, and go at that. That would be probably what I would advocate for this small(er) budget hypothetical, and that’s pretty much what we do, and many other people do around the world, who are running research programs at universities with limited budgets.

It’s one way to fix the trolley problem isn’t it, to not have the problem at all?

Yes, and the trolley problem is a nuanced, very complex thing. It’s rightfully brought up in a lot of public debate, but it’s also … even before autonomous cars came along, there were some problems with the trolley problem in terms of it being an absolute hypothetical, and humans really not being very good at thinking about absolute hypotheticals, because they know there’s always some uncertainty involved.

Indeed. Okay, now away from hypothetical land and back to you, of all the work you’ve done so far, what is it that you’re most proud of to date?

During my PhD, with Professor Gordon Wyeth as my supervisor and building on previous student research, we started trying to model aspects of navigation in the brains of rats and make robot navigation systems better. That was a very long and very challenging journey, but over the course of about a decade, from where we started with fundamental theories and neuroscience, we managed to translate that all the way to areas where aspects of that research are now being deployed in mine site technology trials. And doing that sort of full circle, from initial concept all the way to sort of trying to deploy it, that’s been a really challenging but also very fulfilling experience.

And while Gordon is obviously one of the critical people in my career, but there’ve been literally, and I’m not exaggerating, there’ve been literally hundreds of influential collaborators and mentors over the course of my career alone that have led to where I am and what I do.

Okay, so this one might be tricky for you, having done quite a lot of work in quite a lot of areas, but is there something that you haven’t done yet that you interests you and you’d like to get into?

One thing that I would like to do, it’s just a matter of priorities, is commit 110% to a startup for a few years and see where I can take it. That could be in the technology space in self-driving cars, but right now I have a startup in education. It would be interesting to commit to that, forget about everything else and just do that full steam ahead and just see where I ended up after a few years.

What’s the education thing you’re doing?

We combine entertainment with STEM education. We have a series of kids’ picture books and story books, and a young adult novel, in which we embed STEM educational content, but very stealthily, accompanied by more explicitly educational textbooks and worksheets. We sell those around the world. I think we’ve sold them in 33 countries so far.

And so did you say book, or is it an online project too?

It’s books with online support material, such as animations. We’ve also run workshops that act out some scenarios from films, and events at the World Science Festival around maths and physics in the movies. Along with running similar events at pubs to get adults engaged. So all sorts of ages, all different things. But the common theme is drawn from science and mathematics as portrayed in the entertainment world.

In the next three to five years, what is it in all of these areas, be it transport, smart cities, or again, whatever’s associated with those areas, what is it that you’re most excited about?

I am most excited right now about autonomous vehicles. I think if we make a lot of progress in the next few years, we will pretty much for the first time in history have artificially intelligent machines moving around us every day in our everyday lives. Hopefully saving lives, and getting us to where we need to be. All of that would be just such a transformative notion for humanity!

Indeed, and what about things, do you see a problem with connected and autonomous vehicles being too good a thing and will perhaps a la the movie Wall-E become not physical active beings anymore?

You’re giving me one minute to answer that?! OK.

So any new technology can be good and bad. A lot of the new technologies have the extra characteristic of being a multiplier or enhancer on the goodness or the badness of their effect.

I think the reassuring thing is that governments, the public, all the key stakeholders now are far more educated about all of these technologies and the potential downsides, as well as the positives. From everything I see around Australia, indeed around the world, I think we’re going to have a reasonable amount of control over the narrative of how these technologies unfold and how we let them affect us, and I think that can only be a good thing.

Michael is currently working on one of iMOVE’s projects, How automated vehicles will interact with road infrastructure.