Robot dietary requirements and Jedi
I am definitely not the first person to have blogged about the wince-inducing news that the American military have developed robots that can power themselves by eating organic material. However, I’ve not read much that speculates on what interesting things could happen if you marry this robot technology with several other recent developments in the field of computing – like “adaptive behaviour” for instance. (I use “interesting” in that very British way: gross understatement, superficial calm and underlying blind panic all at once.)
The Energetically Autonomous Tactical Robot (EATR – geddit?!) was sponsored as a business project by the Defense Advanced Research Projects Agency (DARPA). This is the U.S. Government’s geek arm, where techno-nerds work on things like removing debris from space and developing an autonomous walking quadruped platform. (Basically they are all Star Wars fans and want to build a real life Imperial AT-AT – George Lucas should sue for breach of patent. And I bet they all put “Jedi” in the religion box on census forms.)
DARPA has been responsible for lots of sci-fi military technology.
They sponsored the development of the MQ1-Predator. This is the un-manned drone missile platform that costs $4.5 million a pop and that enterprising Iraqis hacked using a €30 piece of software called SkyGrabber – PayPal accepted. (If you are not quite sure yet what you’d do with your very own MQ1-Predator there is a trial version here).
Rather more “out there” DARPA projects include research into military-trained bees, remote-control rats and thought-controlled weapon systems (see ST081-022) . (If you are feeling really brave, there are 522 pages of declassified projects for 2010-2011 on the DARPA website. )
I first became aware of EATR in a report by Stephen Sackur on Radio 4, excerpted from a report he prepared for the BBC World Service. (If you’ve time, have a listen. It is chilling.)
Really?! I am glad they are straight about that!
Considered in isolation, I can fully accept that the problem we are dealing with here is one to do with the ethics of the individuals charged with programming these robots. There is only a problem with the Geneva Conventions if someone set the parameters for their diet too widely. (I am not entirely convinced on this point. I am diabetic – and I know something about the vagaries of diet control.)
Chaos Theory and “adaptive behaviour”
All well and good.
However, a few weeks ago, I was among those who marvelled at Professor Jim Al-Khalili’s programme “The Secret Life of Chaos”, a scientific tour de force which sought to outline the role played by chaos theory in producing the order we see about us every day. Whether or not you accept his assertion that the simple yet unpredictable rules of chaos which underpin evolutionary theory are the sole reason for our existence (rules best depicted by the infinitely variable intricacies of the Mandelbrot set) his analysis of process and the role of mathematics in revolutionising our understanding of biological processes was simply brilliant. (As it happens, I disagree with his assertion that this understanding takes science beyond philosophy and religion – for even if you argue this understanding to be true, there is still no scientific explanation for the existence of this behaviour-dictating rule-set in the first place. Btw you should really watch the Mandelbrot set zoom sequence above. It is incredible.).
One fascinating part of that programme was a demonstration of evolutionary behaviour in successive generations of computer avatars. Games software has often been at the cutting edge of computer technology. The increasingly complex coding necessary to create more life-like games enjoys a symbiotic relationship with computer hardware manufacturers producing faster and more powerful computer chips.
The software development company Natural Motion grew out of work at Oxford University, commercialising research into human and animal movement. Co-founder Torsten Reil, described as an “animating neural biologist”, worked on creating simulations of nervous systems based on genetic algorithms. He and his team set out to teach stick figures to walk using virtual neural networks analogous to that bit of the nervous system in the spine, something described in computer terms as “adaptive behaviour”. (It’s important to distinguish between this “learning” process (complex) and simply programming a computer avatar to walk (simple). This was the former, effectively the computer-generated avatar teaching itself how to walk).
Reil’s team started with lots of neural networks. By their very nature, those in this first generation were going to behave in a random manner. However, a genetic algorithm selected those examples that showed some promise, for example the avatar managing a small step rather than falling over, and then included that behaviour in the next generation of avatars. In twenty generations, the avatars had taught themselves to walk in a straight line. If you are interested, there is the most phenomenal video available at Technology, Entertainment, Design: Ideas Worth Spreading, in which Reil gives an inspirational and fascinating presentation to a live studio audience:
Natural Motion’s white paper on “Dynamic Motion Synthesis”, published in March 2005, sums up the power of adaptive behaviour technology:
“[If] animation assets are synthesised by a sufficiently fast CPU, they need not be static but can be dynamic and adaptive. This means that animations can be fully interactive and adapt to user input and a changing or unpredictable environment.”
This technology is now an integral part of computer game design in some of the world’s largest studios. Natural Motion’s endorphin technology is used by firms like Sony and Electronic Arts, whilst its morpheme software is used by Codemasters and Eidos, amongst others.
The moment I heard Sackur’s piece on Radio 4, I found myself asking the “What if?” question.
What if some bunch of loons, with a multi-billon dollar budget, thought that there might be some merit in at least experimenting with a synthesis of EATR and adaptive behaviour technology? Could those be the sort of loons that spend millions on bot-bees and robo-rats? The very same techno-brilliant weapons nerds you might find digging into the deep pockets of DARPA?
Throw the robo-rat technology into the mix and suddenly the prospect of a corpse-eating robot, that adapts itself to its fighting environment and, from time to time, turns captured soldiers into DARPA’s version of the “Borg” (or even EATR’s version of the “Borg” – the ones it doesn’t eat, anyway), and suddenly it all looks even more alarming.
Is it really too fanciful to conceive of robots that are imprinted with an evolutionary artificial intelligence and that then calculate it is in their best survival interests to over-ride the protocols relating to the Geneva Conventions, eat the corpses of dead soldiers to remain fuelled and re-deploy captured enemy combatants by remote control to bolster their offensive capabilities? After Abu Ghraib, Guantanamo Bay, Project MK-ULTRA and Binyam Mohamed, reliance on the programming ethics of human beings seems a rather flimsy defence. It seems even flimsier in a context of unconventional warfare, billion-dollar research budgets and ever-diminishing physical resources.
Tech-heads and Terminators
James Cameron’s Terminator franchise depicts a world in which an artificially intelligent computer, Skynet, takes over the world, with computer-controlled robots deployed to destroy humanity. The long trailer for Terminator 2: Judgement Day gives you a pretty good idea of what it is all about. Not even Cameron had Arnie eating the corpses of fallen fighters for fuel. (Incidentally, for all you conspiracists out there, Skynet does indeed exist as a family of military satellites providing strategic communications to the UK Armed Forces and its NATO allies on coalition operations).
We may laugh about these things, but as we do, teams of DARPA boffins are beavering away in the classified bowels of the U.S. Department of Defense. If EATR, robo-rats and mind-control target acquisition are declassified, we can only wonder at what is going on behind DARPA’s firmly closed doors – and suddenly Cameron’s Skynet is looking distinctly ZX-81 compared to what could be coming down the military technology track…
And now for something (not) completely different…
In the meantime, rest assured that it’s not just the DARPA boffins that risk losing the techno-weaponry plot. Deep in the badlands of Texas, the gun nuts of Mil-Spec Monkey™ have discovered a need for flash-lights that double as submachine guns.
I kid you not.
Watch… And snigger. Or groan. And be glad you don’t live in Texas.