Decisions, decisions

Toby Simpson
5 min readSep 3, 2018


Back in the late nineties, when the music was bad and the millennium bug was starting to scare people, I was working on biologically inspired digital organisms. Although they were primarily deployed in games, we put a lot of consideration into other applications. Primarily, this was because humans are really, really smart, and their application to almost any non-trivial problem can be used to significantly improve the outcome. The catch is that many of these applications would drive humans utterly mad and our biological need to pop to the toilet, daydream, eat, drink, gossip and socialise ensure that the unique general purpose learning capabilities delivered by our brain only does the job in utterly unachievable circumstances. If we could get software to behave in an increasingly more human fashion, then we could apply such abilities to an enormous range of tasks — so that is what we tried to do, move our digital organisms from the virtual gaming worlds into the real world.

One of the lighter applications we toyed with back then were traffic signals. Traffic signals are operated by simple electronics that may, or may not, extend to involving a microprocessor and some simple software. They use sensors buried inside the roads and others mounted on the signal poles to establish a simple picture of their little world and then operate accordingly. Sometimes, they plug into other local traffic signals, but mostly they do not. Let’s face it, they’re rubbish. They don’t adapt and they’re dumb. They are “just about passable” at managing traffic flow in that, generally, traffic flows marginally better and with additional safety with them there rather than not there. But in many cases, only just. It takes very little in the way of unexpected situations for them to make things worse. Just one minor accident. An ambulance. A social event. Lots of pedestrians. An incident on another road, kilometres away, that causes a knock-on effect. Rain. No rain. A mother duck and a bunch of ducklings. Honestly, anything can mess it up because it can’t think, it can’t adapt to surprises and it is incapable of viewing the world in the way we do and making sense of it: your induction loop tells you a large metal vehicle is there, but not that it is part of a funeral procession that should absolutely not be split up by a red light at a poorly chosen moment.

Traffic lights bring order to the road, but are far from an ideal solution

So imagine if you stuck a human being in a comfy chair at the top of each signal at a junction. You gave that human some binoculars and the ability to speak easily to the humans at the top of the other lights at the same junction. They are rewarded according to traffic flow: if the important stuff makes it through, the pedestrians are safe and the traffic flows well, they earn a lot more than otherwise. With this incentive structure and human smartness it’s easy to imagine how well things could go. Of course, this is totally impractical for all the bodily functions and mental health reasons mentioned earlier, plus some other whoppers such as the obvious lack of enough humans to man every signal. In a splendidly large nutshell, this is one of those cases where it would work, but only in theory and even then, only for the odd half an hour when everything lined up. But all of that pales into insignificance when you look at the value that’s generated — it’s simply not enough to support a human’s life. Mortgages, healthcare, stability of employment… there isn’t an ice-cube’s chance in lava that manning a traffic signal can support all of that. There’s a reason why only a certain class of hotels uses humans to open doors for other humans, and even then, just the front door. It works well, but it is economically viable in just a few, rare cases.

The problem with replacing humans with digital organisms or agents for all these cases is that those organisms cannot see the world in the way that humans can. Sensors are not enough, because if they were, traffic signals would already be efficient, doors would be magical and you’d never have to touch a light switch ever again for as long as you live. Computer vision coupled with cameras and neural networks improve things (and increasingly so as time goes on), but aren’t some kind of magic bullet here: they build a representation of the world through interpretations of observations without the fundamental understanding and context that humans are so good at grasping.

Ultimately, digital entities (be them just dumb computer programs, splendid computer programs, biologically inspired digital organisms or amazingly advanced AI) are compromised because they are forced to live in our world and see it as our world. Which is a problem because our world is one that 3.7 billion years of natural selection has optimised us to live in, not them. And no matter how smart we made our digital organisms a couple of decades back, this was always the bag of spanners in the works: there was no effective ability to visualise and interact with the real world in any meaningful way. It was like they were living in a sealed box, isolated from any opportunity to flourish and develop by themselves.

For too long, digital entities have been locked out of the real world

What is needed is a version of the world for digital entities — our world, but structured for their eyes. This needs to be a space that arranges what is important, removes what is not, and delivers a unique representation that embodies all the facts and information in a way that can be easily digested by the digital, not the biological. It should be the kind of space that understands the things in the world (rather than just tries to understand from observations) and places them appropriately. Other entities should be able to view that space not just on some form of geographic dimension, but on other useful dimensions, such as those that see decision points or economic factors such as cost, safety and speed. It’s a space for everything to be represented: all the stuff that we know is important, and the rest too because who knows what might turn out to be important? And for the cherry on the top, it would need to restructure itself in real-time using machine learning and artificial intelligence to ensure that each entity sees a world that is perfectly adapted just for it. Wouldn’t it be amazing if such a digital space was available?

The first of many of those spaces is now there, and that’s Fetch.AI. The true “metaverse”, whatever it turns out to eventually be, is the perfect decentralised world that can be inhabited by digital entities and for those entities to be able to be rewarded for the success and economic value that they deliver. They are able to sell each other information in order to operate more efficiently, and they are rewarded for doing so. This is not a zero-sum game. This is a world that defers the decision making to those that are best placed to make those decisions — the things that are right there, where it’s happening.

And for the first time, these entities have a world that is just for them. It’s about time, really.



Toby Simpson

Opinions my own, yours may be different, and that’s cool.