Photo: Bokeh / Tomedia. Two Military Jets in Flight.
Welcome to AGI · Part 5 of 6
This is part five of Welcome to AGI, a six-part Brainwaves series. The last article stayed with the bot war on websites and the pressure to build better defences. Now I want to follow the same problem into the physical world, where better tools can help somebody get a job done and give somebody else more power over the people around them.
Real Steel, apparently
I will admit I would gladly go and watch robots getting absolutely destroyed by each other, Real Steel style. There’s something very entertaining about a machine trying to do something impressive and falling over instead, or two robots attempting to kick each other’s heads off while people watch. I can see the appeal of turning that into a competition between companies. Everybody gets to show what they’ve built, the robots get put through something a little less predictable than a carefully arranged demonstration, and I get to watch robots fight. I would probably go and see it in person.
China’s robot games are already giving us that. Reuters filmed humanoids kickboxing, running and performing tai chi at the August event, with people using remote controllers for the kickboxing. So the fighting you’re watching involves people operating the machines, with all the physical work of keeping a robot moving and upright still happening underneath. The footage description[1] is clear about that. Even at that stage, you can understand why there is so much interest. We are getting better at making machines that can move around and interact with spaces built for people, while separately developing models that can work through increasingly complicated tasks. I find both interesting, and then I start thinking about what someone might want the combined system to do.
The Lego brick
The Generalist demonstration that stayed with me was a Lego brick getting stuck to the robot’s fingertips. It used its other hand to pull it off and could carry on with what it was doing. That was GEN-1.5, announced on 19 August. It looks so ordinary when a person does it that you wouldn’t normally give it much thought. Something gets stuck to your hand, you use the other one to remove it, and you get back to the job. But watching a robot do it makes you think about how much of dealing with the world involves those little adjustments. You can give a machine the main task, and then some awkward physical thing happens that gets in the way. Somebody has to deal with that before the main task can continue, and in this demonstration the robot dealt with it itself.[2]
Generalist also showed a robot removing paper that covered a bowl so it could finish putting a block inside. The company describes these as examples of improvisation beyond the demonstrations used to teach the task. They’re still working with simple, short tasks, but you can see why recovering from interruptions would be useful. If a person has to intervene whenever something catches or an object ends up somewhere unexpected, they have to keep paying attention to all those little failures. A machine that can deal with some of them needs less help. I can see the value of that for people trying to make robots useful, and I can also see why the video feels different from watching a machine repeat a movement. The robot is dealing with something going wrong during the work, which is a fairly familiar part of doing anything with your hands.
AI and robotics are already mixed together, and I would like to see what else people can do with it. Unfortunately, while I’m watching a robot work out what to do with a Lego brick, other people are already investigating what these systems could do for a military.
Then there is the military
Reuters reported on 7 September that China’s military was pushing humanoid development towards military applications. After the robot games, PLA Daily called for faster movement from laboratories to military training grounds, and Reuters examined research and procurement documents concerning those uses. The investigation found no evidence of an armed humanoid deployed with an operational PLA unit. That is where the reporting stands, while the military interest and development work are already there. Reuters’ investigation, republished by Internazionale,[3] describes people working towards capabilities they would like to have. It is perfectly possible to find that direction horrifying before every technical problem has been solved and someone has put the finished machine on a front line.
And my concern is any government doing this. China is particularly relevant when you’re watching from Australia, but the United States is horrifying too. The US Army’s xTechHumanoid programme[4] seeks humanoid systems and components for military roles, including reconnaissance, security and hazardous operations. It is conducting development and experimentation of its own. I don’t get much reassurance from seeing a country I’m supposed to regard as friendly pursuing the same sort of capability. The people in charge still have their own interests, and they still have the ability to use power against other people.
There are different amounts of autonomy involved in remotely operating a robot, letting it work out how to move through a space, and letting a system decide when to use force. Those decisions have consequences beyond whether the robot manages to stay upright. A remotely operated machine still has a person directing it, which leaves me asking what that person has been told to do and who authorised it. If more of the work is handed to the system, we also have to consider how well the people supervising it understand what it will do. A system can fail to follow the limits people intended, and people can deliberately use a system to do something horrible. That is where my brain goes to Terminator.
Attrition, funding and silicon
I was already getting into the money side of this in Ghosts in the War Machine. I wondered what war starts to look like if more of the soldiers can be replaced with machines, and then what happens when both sides are trying to afford enough machines to keep fighting. Money already pays for equipment, replacements and keeping a force supplied. We haven’t discovered a new relationship between wealth and war. What worries me is how much further that advantage could go when the side with the money can treat more of its own losses as damaged hardware, while the people it is attacking still have to deal with the consequences in their own bodies and homes. An attacker can replace a machine that has been destroyed. The people it reached before that happened may have lost things they cannot replace.
Imagine being able to send machines into a war zone while keeping more of your own people away from the danger. You lose some of the machines, order replacements, repair what can be repaired and keep going. That is the possible future I’m concerned about. It would still require production, transport, maintenance and everything else involved in keeping equipment working, so having the money would have to translate into being able to supply the force. But if you can afford to keep doing that and the people you’re attacking cannot afford to keep stopping it, their problem gets worse with every replacement you can send. From their side of it, what are you meant to feel about a force that can keep replacing the things you manage to destroy? Surviving one attack doesn’t mean you have the means to survive the next.
I don’t find the thought of robot-versus-robot warfare much more reassuring, either. In the earlier article I wondered whether it becomes a budget problem, with people stuck in the middle paying the price. If you need your own machines to keep the other machines away, you still have to afford them. You have to keep them working and replace them when they fail or get destroyed. The people who can’t afford those defences don’t get to remove themselves from wherever the war is happening. They are still there while somebody else works out whether it can fund another round. Calling the equipment expendable describes what its owner is prepared to lose. It doesn’t tell us much about what happens to the people near it, or whether they ever had a choice about taking that risk.
I wrote about resistance rising in Ghosts in the War Machine, and I’m finding less reassurance in that now. People can resist and still face something they cannot afford to keep fighting. I don’t have a satisfying answer for what gives them protection or bargaining power against someone who can keep replacing its losses. And isn’t that already part of war anyway? The number of people you can send, the resources you can keep supplying, the amount of damage you are willing to inflict and absorb. I think we’re moving the line further into a type of war fought through attrition, funding and silicon, with a growing difference between what the attacker has to risk and what the people underneath it stand to lose. I don’t see why those people should feel reassured because fewer of the things coming towards them have a human inside.
The useful jobs still exist
Obviously there are benefits to all of this. I can see the value in rescue, repairs and getting people out of harm’s way. If a robot can do something dangerous without putting a person there, I understand why somebody would want it, and why the people building it would be proud of making it work. Being able to deal with an obstruction or recover from a mistake could help with useful work as well as the military applications. I use AI in my own work and want it to help protect my clients, so I understand wanting to keep developing something useful even while being concerned about the other ways it could be used. The useful work still needs doing, and there are people trying to build something that helps.
But people will still be corrupt. They will still use systems in ways they weren’t originally intended, and some will build them for purposes they already know will harm other people. Giving a system better instructions or making it more reliable doesn’t make the person operating it trustworthy. Absolute power corrupts absolutely, and the people holding that power may not completely control the machinery themselves either. So I end up watching these developments with much the same feeling I have when I’m building with AI: I can understand the appeal and the useful work, while also seeing why somebody with bad intentions would want it. The people with the money are already exploring what these machines could do for them. I don’t know what protection the people who can’t afford to join them are going to have.
In the final article, So what’s the solution? I don’t really have one, I want to come back to accountability. I think there should be consequences when the people operating these systems lose control or choose to use them against others. Then I get stuck on who has enough power, and enough trust, to make those consequences happen. I’m still trying to build useful things in the middle of that, so I don’t get to leave my own responsibility out of it.