How to train your Warjack: Reinforced Steel.

Cortex Banner
Cortexes are really nothing like computers… They are simulated brains, not calculating machines executing lines of code. They are more trained than programmed.” – Doug Seacat

The moment I read those words, I started thinking about how warjacks might learn. Given that they are magical simulations of brains, it seems likely to me that warjacks learn in similar ways to humans and animals. The computer metaphor is very prevalent in mainstream cognitive psychology, and in the popular understanding of psychology, but I (and a lot of researchers) take strong issue with it. We contend that human brains are nothing like computers either – they don’t execute lines of code any more than warjacks do.

Since I think an operant conditioning account can successfully explain complex human behaviour (see http://www.contextualpsychology.org and Blackledge (2003), if you’re interested) I’m obviously going to look in that direction for an account that can guide warjack development in your IKRPG games. I’m going to etch out the basics of that account, and then talk about where warjacks are likely to diverge from humans in how their behaviour is shaped.

The first principle of human learning is that behaviour is shaped by consequences. A behaviour that is followed by a reinforcing consequence is more likely to be emitted again. A behaviour that is followed by a punishing consequence is less likely to be emitted again. Two things are important to note in these definitions:

1) Behaviour is probabilistic. Learning changes the probability of a particular context evoking a particular behaviour, but does not determine it.

2) Whether something is defined as a reinforcer or a punisher is determined by its effect on behaviour – nothing is universally reinforcing or punishing, and a particular thing can have its function change via learning, or via context.

Armed with a knowledge of what is reinforcing/punishing for an organism, you can begin to shape behaviour. Behaviour is shaped by controlling the consequences of the organisms actions – you reinforce desired behaviour, and punished unwanted behaviour. Complex behaviours can be shaped by reinforcing behaviours that are like the behaviour you want, and then slowly raising the bar for specifically what you’ll reinforce.

The first question I asked myself for this article was “what is innately reinforcing for warjacks?” Humans have some innate reinforcers (food when hungry, water when thirsty, sex when safe) and some innate punishers (mostly pain, possibly social exclusion) so I would expect warjacks to have the same. However, human innate reinforcers are evolved, selected for by evolutionary processes. All of our innate reinforcers are about survival, as that is the selective consequence for genes. Warjacks are made, not evolved, and that Some Other Thing must shape what the innate reinforcers of a warjack are. (This, I think, is an interesting little mystery of warjack cortexes).

Most of the ideas I talk about in How to Train your Warjack will be being used in the IKRPG game I’ll be running in a few weeks. Of course, none of this is canon, but I like it! Hopefully it’ll inspire some things in some of your games – please do share in the comments!

Doug Seacat notes that attempts to put cortexes in control of boats and trains didn’t really work, and hinted at some aggression problems. There’ll be more about Embodied Warjack Cognition later in this series, but, for now, I’m going to suggest that steamjacks are reinforced by physical work. Humans seem to be somewhat reinforced by effectively practicing their body’s competencies – listening to music to practice the ear, sports to practice different types of physicality, etc. Steamjacks might be reinforced similarly by doing physical things – not just anything physical, but human-like physical.

Warjacks in particular, I expect, will be reinforced by physical damage to an enemy/obstacle. They’re creatures of war, and their instincts tend towards aggression. Angry humans are reinforced by this too, but I would suspect that warjacks are reinforced by this in general, not simply when provoked to anger. (Or you could think of it as them being more generally angry all the time. Certainly, that helps suggest a shape for warjack personalities that differentiates them from labourjacks narratively.)

Being incongruent with their controller.  Humans find incongruent thoughts punishing (see cognitive dissonance). Steamjacks in my campaign similarly find incongruence with their controller similarly punishing. I like this as an idea because it creates all sorts of interesting interactions in a narrative. In my game, ‘jacks will find it highly aversive to have their actions and desires (however simple) be out of sync with what their controllers want. This can be used as a metaphysical whip to train a ‘jack whose behaviour is out of line, or as an unintended effect which causes warjacks to slowly change their personalities to match what their casters want from them.

Consider this scenario:

A warcaster and his ‘jack have been together for a long time, when some tragedy occurs which sends Caster into a spiral of depression. ‘Jack and Caster have always been active, upbeat, happy go lucky types, but now Caster’s mental state is creating a punishing incongruity for ‘Jack. ‘Jack wants this punishing state to end, so tries to do things which remove the incongruity – trying to make Caster happy with little gestures and dances, or trying to hurt people who he thinks are making Caster sad. In time, the constant punishment of the incongruence will drive Jack into non-behaviour. Constant punishment doesn’t just suppress one behaviour, it suppresses them all, creating learned helplessness.

This follows the usual processes of behaviour change that follows a lot of punishing consequences:

1) If the punisher predictably follows a certain behaviour, stop that behaviour
2) If the punishers are unpredictable but still constant, attempt to escape (Make Caster Happy)
3) If escape does not work, attack (attack possible sources of Caster’s sadness)
4) If both of the above fail, stop behaving. (Learned Helplessness)
(For warjacks, aggression is probably likely to happen earlier than escape.)

(Aside: This, by the way, is why using punishment to shape behaviour in humans and animals is generally a Bad Idea and Should Not Be Done. Shape behaviour with reinforcement.)

I’m not entirely sure what else might innately reinforce a warjack – I’m casting about for the warjack equivalent of “throwing them a snack”. As I think about it, I think that Punishing Incongruence is probably one of the main ways in which a steamjack learns. And, as outlined above, extensive use of punishment has some negative consequences, not least of which is increased aggression. Which, as noted earlier, is something warjacks have a bit of an issue with. Kinda makes sense, really… especially since I’m quite sure “damage to an obstacle/enemy” is reinforcing. If both of those things are innate to warjacks, their aggressive instincts would naturally emerge and multiply.

There’s a couple of other reinforcers to think about, briefly – Novelty and Control. Both of these have been consistently found to be reinforcing to even very simple organisms. Organisms will work hard to gain opportunities for novel experiences, even choosing them over more reliable food/water. The same is true of being in control of the consequences of your actions – in a behavioural sense, organisms are reinforced by making reinforcers more predictable. I think these would be interesting to include in a list of steamjack reinforcers, because it gives a narrative direction for smarter, older, more capable warjack – they will select behaviours that give  them a little more autonomy, a little more novelty.

This article is getting pretty long at this point, so I’m going to call it a day… More later this week!

Te Nosce,

Anthony

Advertisements

One thought on “How to train your Warjack: Reinforced Steel.

  1. Woosh, that’s a heady mix of information, there. I do like the idea of the cognitive dissonance being a primary shaper of behavior – a ‘jack will want to behave more like it’s controller in order to be more in sync with the controller, which plays nicely into the idea of bonds and imprints. Would definitely enforce the idea that warcasters need to be of disciplined mind, can’t afford for a moody warcaster causing multi-ton killer robots to get “antsy.”

    Worth noting is that as far as I’ve been able to tell in the fluff, there is no difference in the fluff between a cortex used in a warjack and a cortex used in a laborjack. The difference seems to be in the cortex grades, with the higher grades having a better learning curve, higher ability to learn complex behaviors, better coordination (higher MAT/RAT, 2-weapon fighting, ambidexterity), and the ability to better connect with their controller. Assuming that, it seems likely that the aggression of warjacks is more likely a learned behavior than an innate characteristic (I suspect Doug’s comment about boats and trains attempting to throw and slam each other was a joke rather than a behavioral observation, I always took it to mean cortexes have “instincts” for humanoid shapes), which would make sense if doing new things and doing things in general were both reinforcers, since combat would tend to be a constantly shifting experience which they would engage in frequently.

    I’m also curious what your take is on the training of ‘jacks used by Jack Marshalls. It seems like the difference between training a dog that you have to command with hand gestures and voice commands (admittedly, it only obeys YOUR hand gestures and voice commands) would be rather different from training a dog that reacts based on your mental impulses.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s