The Australian Robotics Paper

The folks that care about Australian robotics are devising a national plan.

The discussion paper is here.

I believe the definition in the paper restricts the vision, so it limits the role Australia can play.

Here’s their definition: “Robots are machines with a degree of autonomy that can move within their physical environment and manipulate objects. Robots have 4 essential characteristics: sensing, movement, energy and intelligence.”

Here’s my definition: “Robots are individual and federated devices that have three characteristics:

? the ability to be influenced by their environment which presumably includes human designers;

? the ability to perform some reasoning; and

? the ability to effect change in the world.”

Systems that are termed “AI” lack that third component.

This definition allows for three different notions of linked affordance, with each presenting technical challenges and advantage.?

? Affordance one is the ability to affect, direct or inform the robot;?

? Affordance number two is the ability to internally affect, direct or inform the reasoning system;?

? Affordance three is the ability to affect, direct or inform the world.

The first is often considered as a combination of sensors and directives, implicit and reactive. The second is termed ‘learning’ in the popular press. The third is usually considered to be movement of "objects". All three of these in the proposed document use obsolete definitions and limit the vision and opportunity.?

If we want to have a progressive strategy, we need to escape the norm.

The strategy in the discussion paper and presumed plan unwisely mixes the benefits of the use of (current and near future generations) robots with the benefits of developing the next generation of devices that we presume may be radically different and possibly unimaginable in current terms.

If you want to worry about both, you need two different and unrelated strategies. Goals of the former address end users, technologists and engineers, each of which are sophisticated and motivated in different ways. Goals of the latter require advances in science and underlying foundations.?

Once the document’s roadmap gets into recommendations, it abandons the research and startup potential of these disruptions. For example, a theme throughout is how to support, even subsidise companies to sell and use the current technologies. But promoting the benefits of the existing generation — beyond current market forces — could divert attention from the fact that we are likely at maturity of one on a scale of ten and there are several disruptions to be found in the sector in the mid term.

I would instead like to see some recommendations on how to overcome the massive deficiency we have in the root of our innovation chain. Elsewhere, I’ve noted our research deficiencies. It is less worrying that we are not competitive internationally, than that no responsible actors admit it.

If we were to have a robots-focused research policy, our suggested definition may help guide it, as the focus on ‘affordance’ emphasises the scientific challenges. For example, I’ve read a lot recently about how to trust military robots. All the results I’ve seen focus on the trust part of the equation, reslicing ethics and regulations developed for other contexts.?

Think instead of abstractions that capture society’s ideal constraints and how they translate into the reasoning and learning environment that computers use. About eighty percent of that reasoning environment works by imposing constraints already. But that’s only half of the problem, because the effects of the original constraints need to convey to constrained effects of robots as they act.?

An analogy is how we can shape sun, water, and a few chemicals into an intermediate ecology of food and have the affordance chain carry through to healthy bodies. There are good formal frameworks for understanding how these abstractions can convey through the affordance layers to guide breakout research. I’ve used these in my US labs, the DARPA vacations, and now in our recent ambitious work.?

A good research manager will have their own framework of course — and mine may not map as well as I suppose. But the point is that you need to have one and that should be the job of a national framework. If you don’t dig deep into the problem space, your work will not be world-changing. For example, the products of Open AI that you’ve read about — breathlessly reported — are not a breakthrough in any fundamental way. We are just seeing the application of scale to principles we mastered a decade ago.?

While many folks are getting excited about Open AI and its new siblings, some small groups are working on the next set of underlying foundations that the money folks will leverage in five years. We need to have a few of those groups here in Australia. If we don’t, we’ll be basically adopting tech devised elsewhere, and sending our money away.

Does it matter that we had a CSIRO team that mastered robot tech to do well in a DARPA-managed navigation challenge? Of course it does; we need good engineers in the robotic space.

But that is not where you build futures. Those are operators, not creators.