Editor’s Note
This is a guest blog by author Andrew Skinner. Andrew grew up in South Africa’s coal-mining heartland, amidst orange dust and giant machinery. He now works as an archaeologist and anthropologist, interested in folklore, rain-making arts, and resistance; but the machines aren’t done with him yet. Steel Frame is his first novel. Andrew Skinner can be found on Twitter at @apocrobot.
Guest Blog
A few years ago, in the early stages of writing what would become Steel Frame, I read an article that fundamentally challenged the way I saw the future of robotics. The article’s been delisted now*, but I’ll rehash the part that really caught me off guard (bear with me).
In the early 2000s, a pair of researchers at the University of Sussex set up an experiment; they developed a piece of software that could mimic a kind of natural selection. It had a ‘genome’ that represented different configurations of transistors, and instructions to put that genome through mutations that changed the connections between those transistors, until it evolved into something that could function as an oscillator (a simple circuit that puts out a controlled, repetitive signal). Every generation that got closer to putting out the right kind of signal ‘survived’, and became the basis for the next generation.
Fast-forward to the end of the experiment, and their creation has succeeded. From the barest building-blocks, it has evolved something that does what oscillators do, and puts out exactly the signal they’re looking for. But! It hasn’t built an oscillator.In fact, the researchers can’t find any part of it that should be putting out a signal. Eventually, they trace the source – it isn’t coming from the circuit itself, it’s coming from all around them.
The source revealed…
The experiment was set up in a computer lab, and their creation has worked out how to hear the other machines in the room. Using components that didn’t include an antenna, it sensed that there were other things in its environment; other things that put out oscillating signals.
Instead of building an oscillator, it built a radio. It evolved a sense of hearing, and then built something that sang along to the sound of the machines around it.
Interpreting instructions — computers are like fairies
There’s some old hat about computers being very good at doing what you tell them to, but not very good at understanding what you mean. That isn’t the issue here – at least, not exactly. This program followed the original instruction, and did what the researchers directed it to; what makes it all so weird is how differently it understood the problem, and the world, to its creators.
I think we envision the endpoint of robotics and AI to be something like ourselves. I know I did. When I wrote my machines, my objects-as-characters, they tended to be pretty recognisable. A kind of ‘Us 2.0’, more intelligent and capable but driven by intentions we could understand. It makes sense if you think about the way we portray the creatorsof sentient machines; they’re the Doctor Calvins, the greying engineers and career scientists; they’re startup kids and Elon Musk lookalikes; visionary and fallible all at once, but most of all, recognisable.
The future looks alien
At the moment, it’s looking less and less like that will be the case. Both for abstract AI and practical robotics; the world is complex problem atop complex problem. How do you move across this surface or that? How do you survive changing conditions? How do you interact with the strange things that inhabit the spaces beyond program and signal? It’s the kind of problem-solving that people tend to do badly—we rarely have the perspective—but that evolutionary development does really well. After all, it’s a system that’s been solving this kind of problem for millions of years.
In practice, this means treating nascent machine intelligence as less of a mathematical problem, and more as something to be grown or nurtured. A child given senses and raised**, where all of the weirdness of experience, the things we would never think to teach something, can mould an entity with a more complete understanding of the world. What’s more, there’s every chance we won’t be the ones doing the raising or the nurturing. In fact, there’s every chance that this evolution will be set in motion by another machine. A previous generation, that learned by chance and experimentation in a world that looks and feels very different to our own.
Imagine…
Writing science fiction is an opportunity to bend rules, and to toy with the parameters of many different times and spaces. It’s also a chance to imagine what may come of this exact moment; right here, right now, and maybe just a little to the left.
The more I think of my machines, the artificial things that inhabit the stories I want to tell, the more I want them to have this particular DNA. I want our mechanical descendants to be familiar aliens – maybe shaped like us, and maybe with a few, minor things in common; but ultimately, absolutely weird.
Because I think our children will be strange.
Footnotes
*You can find a more detailed version in Superintelligence by Nick Bostrom.
**If you want to try this idea on for size, I recommend The Life Cycle of Software Objects by Ted Chiang.
Steel Frame by Andrew Skinner
Rook is a jockey, a soldier trained and modified to fly ‘shells,’ huge robots that fight for the outer regions of settled space. When her shell is destroyed and her squad killed, Rook is imprisoned, left stranded, scarred and broken. Hollow and helpless without her steel frame, she’s ready to call it quits.
When her cohort of prisoners are sold into indenture to NorCol, a vast frontier corporation, Rook’s given another shell – a near-decrepit Juno, as broken as she is and decades older – and sent to a rusting bucket of a ship on the end of known space to patrol something called “the Eye,” a strange, unnerving permanent storm in space.
But they’re not alone.
Book details
Steel Frame can be found at Rebellion Publishing and all good book outlets.
UK: 9781781087053 • 22 August 2019 • £8.99
US: 9781781087046 • 20 August 2019 • $15.99
