A week or so ago in the superstition thread, I said something along the lines of “it’ll be hard enough to make the AI work well without making it work badly.”

Well, that was before I was thinking so much about solving problems in a fake economy. Past me is so passe. I bet he still listens to that band who were cool last week but are so over now.

Anyway, partially inspired by this Chris Dillow blogpost, a potential list of single variable “irrational” weightings that could be used to create diversity of behaviour in NPCs.

Each of the items below can be considered to be a single trait or variable. An NPC might be at 0 or 1 on the scale, not at all to very, but most would be in the middle following a normal distribution. Their position on the scale would make them over or underweight certain factors when creating plans for future action. These would have differing effects on the micro and macro economy.

• Optimism/Pessimism, a.k.a. “Risk Aversion”. Optimistic NPCs overweight positive potential outcomes and are more prepared to take on riskier strategies. Pessimistic NPCs overweight negative outcomes and are extremely risk-averse, prefering low-reward but safe options. Optimistic NPCs are the “entrepreneurs” who get rich quick and then fly their spaceship into the sun. Pessimistic NPCs are miners and bank managers who keep the universe working while the optimists are off trying to invent a new kind of spam.
• Novelty Preference. I’ve mentioned this one before. Paying a higher price for the new and unique even if it doesn’t deliver much benefit. Paying a lower price for something old and common even if it’s perfectly decent. NPCs with high novelty preference are trend setters, early adopters and fashion victims. If you bring back something from a far-flung sector, they’ll want it bad, no matter what kind of worthless tourist tat it is. If you keep manufacturing the same old stuff, they’ll get bored and seek out better suppliers.
• Hindsight bias. An NPC with this trait, having taken a risky path that succeeds, is more likely to try it again. This simulates a common cognitive fallacy where most people presume they were talented rather than lucky if things went right.
• Complacency - If nothing has gone wrong yet, assuming it won’t go wrong in the future. "Nobody’s attacked my trade route so I don’t have to guard it as heavily." NPCs with high complacency underweight familiar risks that have so far not manifested as actual threats.
• Conservatism/Skittishness - Conservative NPCs overweight their current course of action and resist change, even when things look bad. Skittish NPCs will jump ship as soon as something goes bad, even if there’s a chance it’s just a temporary setback. Similar to optimism/pessimism, but path dependent.
• Hoarding/sentimentality. An attachment to items which were used in important/meaningful events for the NPC, or which they've simply had for a long time, leading to reluctance to sell or risk them. Keeping hold of the first ship you made a million credits in, for example, even if it’s hangared up and you never use it any more.

These seem to be the best human biases for easy adaptation into a computational system of weights and numbers, but there may be others. Irrationality, and especially irrationality that varies over a population and over time, can lead to interesting swings and shifts in macroeconomic situations, and that's always fun.

I agree. Would love to see this put into the game. It would probably give it an unpredictable trend.

McDuff wrote:A week or so ago in the superstition thread, I said something along the lines of “it’ll be hard enough to make the AI work well without making it work badly.”

Well, that was before I was thinking so much about solving problems in a fake economy. Past me is so passe. I bet he still listens to that band who were cool last week but are so over now.

Anyway, partially inspired by this Chris Dillow blogpost, a potential list of single variable “irrational” weightings that could be used to create diversity of behaviour in NPCs.

Each of the items below can be considered to be a single trait or variable. An NPC might be at 0 or 1 on the scale, not at all to very, but most would be in the middle following a normal distribution. Their position on the scale would make them over or underweight certain factors when creating plans for future action. These would have differing effects on the micro and macro economy.

• Optimism/Pessimism, a.k.a. “Risk Aversion”. Optimistic NPCs overweight positive potential outcomes and are more prepared to take on riskier strategies. Pessimistic NPCs overweight negative outcomes and are extremely risk-averse, prefering low-reward but safe options. Optimistic NPCs are the “entrepreneurs” who get rich quick and then fly their spaceship into the sun. Pessimistic NPCs are miners and bank managers who keep the universe working while the optimists are off trying to invent a new kind of spam.
• Novelty Preference. I’ve mentioned this one before. Paying a higher price for the new and unique even if it doesn’t deliver much benefit. Paying a lower price for something old and common even if it’s perfectly decent. NPCs with high novelty preference are trend setters, early adopters and fashion victims. If you bring back something from a far-flung sector, they’ll want it bad, no matter what kind of worthless tourist tat it is. If you keep manufacturing the same old stuff, they’ll get bored and seek out better suppliers.
• Hindsight bias. An NPC with this trait, having taken a risky path that succeeds, is more likely to try it again. This simulates a common cognitive fallacy where most people presume they were talented rather than lucky if things went right.
• Complacency - If nothing has gone wrong yet, assuming it won’t go wrong in the future. "Nobody’s attacked my trade route so I don’t have to guard it as heavily." NPCs with high complacency underweight familiar risks that have so far not manifested as actual threats.
• Conservatism/Skittishness - Conservative NPCs overweight their current course of action and resist change, even when things look bad. Skittish NPCs will jump ship as soon as something goes bad, even if there’s a chance it’s just a temporary setback. Similar to optimism/pessimism, but path dependent.
• Hoarding/sentimentality. An attachment to items which were used in important/meaningful events for the NPC, or which they've simply had for a long time, leading to reluctance to sell or risk them. Keeping hold of the first ship you made a million credits in, for example, even if it’s hangared up and you never use it any more.

These seem to be the best human biases for easy adaptation into a computational system of weights and numbers, but there may be others. Irrationality, and especially irrationality that varies over a population and over time, can lead to interesting swings and shifts in macroeconomic situations, and that's always fun.

^ This.

McDuff wrote:[...]

Sounds good. Should any of these traits correlate with other traits as proposed here?

I think they're more complementary. Traits in that thread seem to be more about what kinds of goals NPCs want. Irrationality traits are more about how they evaluate the potential paths to those goals.

Take an Aggressive, Greedy, Unsociable NPC. If said NPC were an optimist she would be more likely to act on riskier opportunities to shoot people and take their stuff. If the risks paid off and she were predisposed to hindsight bias, she might take on increasingly risky raids in the future. Etc etc.

McDuff wrote:I think they're more complementary. Traits in that thread seem to be more about what kinds of goals NPCs want. Irrationality traits are more about how they evaluate the potential paths to those goals.

Take an Aggressive, Greedy, Unsociable NPC. If said NPC were an optimist she would be more likely to act on riskier opportunities to shoot people and take their stuff. If the risks paid off and she were predisposed to hindsight bias, she might take on increasingly risky raids in the future. Etc etc.

Right, what I mean is just, would you expect there to be any actual correlation between the "goal determining" traits and the "path determining" traits? Would you expect aggressive NPCs to likely be optimistic, or likely be pessimistic, or neither more or less likely for either? In other words, if you modeled goal determining traits and path determining traits as random variables, would they be independent of each other? If not, what kind of traits from each set do you think would go hand in hand?

I expect that they'd work reasonably well picked independently. I haven't expanded the full gamut and considered what might work well together, and what might fail. I suspect that's best left to a computer test which can model 100,000 NPCs and see what comes out. Especially since we're not just considering what types might work well, but what types might work well in conjunction with or competition with the other types. That's a lot of variables. Probably best leave it to a model, experiment, see what seems to work and what results in excessive quantities of bone-headed behaviour.

McDuff wrote:I expect that they'd work reasonably well picked independently. I haven't expanded the full gamut and considered what might work well together, and what might fail. I suspect that's best left to a computer test which can model 100,000 NPCs and see what comes out.

This would be a perfect problem for genetic algorithms to solve.

You define a hypothesis of the form (F, R, A, C, A, S, RA, NP, HB, COM, CON, HO), where each of the components of the vector correspond to some distribution function that is sampled to determine the goal determining and path determining traits of NPCs. You then create a test simulation with however many NPCs you want. You decide upon what a "desirable" end result to the simulation is, operationalise it and create a fitness function based on that.

You create an initial population of hypotheses, run the simulation with each of them, determine their fitness, breed the best hypotheses and reject the worst, rinse and repeat until you're satisfied.

Yup.

I don't mind if hypotheses breed because they aren't sentient. 3:

I wonder if we could actually measure and aggregate all the economic activity in a region - then you might be able to tie the fitness function to the macroeconomic performance indicators (taken from here):

• Real GDP Growth (short term and long term)
• Jobs (unemployment and employment rates)
• Prices e.g. as measured by the annual change in the consumer price index
• Trade balances and measures of competitiveness
• Productivity of labour and capita inputs
• Average standard of living e.g. measured by per capita GDP (PPP adjusted)
• Quality and accessibility of public services

I wonder if Limit Theory's economy could become so complex and dynamic that it warrants a full-time professional economist, like EVE Online does. That'd be so cool.

There isn't an LT economy, there are potentially infinite numbers of LT economies.

I'm also not convinced some of those things have analogues in LT (or Eve, for that matter).

Unemployment and living standards, for example. Do we even know if those things *are* measurable in a real way for NPCs?

And what's a "public service" in a world where governments are basically private organisations?

McDuff wrote:There isn't an LT economy, there are potentially infinite numbers of LT economies.

Well, the LT macro-macro-economy then. Or the LT economic system.

McDuff wrote:Unemployment and living standards, for example. Do we even know if those things *are* measurable in a real way for NPCs?

Bear in mind that it's not just NPCs but also planetary colonies that will be part of the LT economy. But as for NPCs, if they are modelled with a standard set of needs and wants which need to be met by goods provided by the market, living conditions could be assessed based on how well they're met. However, I'm not fond of this because it implies two things: (1) NPCs and the player are organics, and not computer programs; and (2) things like hunger, thirst, boredom, etc. would need to be modelled - this isn't The Sims.

As for unemployment and public service, I don't know. I just copied and pasted.