Artificial Intelligence 2

Trying to Keep Prices Reasonable

The difficulty in designing a rational AI player in a free market economy (no crime, all contracts honoured, everyone is rational) is that it may not give the results we would like, versus a script based system with a controlled economy.  Prices can move too high or too low.

The situation is discussed in a previous post, each individual runs on an artificial intelligence to attempt to maximize wealth in a rational manner.  However, the ultimate result gave rise to the following situation:

Every individual begins as a farmer.  They would stochastically (randomly) choose to try out another profession.  Those who produced common wares who had early success could build a store of wealth.  Any subsequent competitor had to price themselves against someone who had a store of wealth who could engage in a price war, take a loss over a period of time, driving competitors out of business and then engage in monopoly pricing again to rebuild the store of wealth.  This process would repeat to the point where one person would sell a single product over a long period of time worth many many turns of food production and never worry about competition.  There would also be a less successful group of "rich" individuals.  Everyone else remained very poor.

Here are some ideas (not necessarily mutually exclusive) that could be used to bring prices down so that there can be a daily market for goods (which would make street markets more interesting in game) and reduce the level of wealth disparity.


Saving Rate
Individuals have a stochastically chosen saving rate, where a given amount of disposable income is put away into savings thereby lowering the available amount of disposable income that can be used to spend.

Maximum Spending Limit
Additionally, individuals may have a budget limit which grows slowly from day to day.  The amount of wealth used to purchase goods grows slowly from day to day.  Rather than add the entire disposable income from each day to the pool, only a portion of it is added and the rest is used as savings.

The starting budget is based on disposable income gained in a day.  Subsequent budgets have disposable income "refill" the budget, and if there is still more additional disposable income, the budget only grows slightly at the margin.

Price Expectation
A person can make a snap judgement on whether they believe the price will rise or fall the next day.  Based on this they will purchase or not purchase a good, with a fuzzy decision making mechanism (ie. they have a percentage chance of purchasing or not purchasing based on the strength of their belief).

A person can also expect the price per unit of aggregate bonus (health and happiness quantified as a number and added together) to be within a certain range as an absolute number or as a percentage of their income.  So if the price per unit of aggregate bonus is 5% of disposable income, but they expect it to be 4%, then there is a low chance for the person to purchase the good.

There can also be, which is most likely to arise in a barter system, a price expectation based on equivalent opportunity cost labour input in terms of possible food production.  That was a verbose statement but simply put: if instead of producing that good you spent the whole time farming, how much food would you have gotten?  If you made a pot, then let us picture instead of making the pot, the person had farmed and the person who mined the clay had farmed, how much food would you have had instead?  Then, convert the food into the equivalent units of wealth (perhaps based on the current average barter price).  This allows you to set a price based on "real cost of input".  However, it's somewhat problematic in that, how much of a profit margin is acceptable to a person (because a price at cost or below makes the endeavour pointless for the seller).


Assumed Bonus from Wealth
Another method to create a certain price range is for people to value holding wealth in a way where they can measure it against a purchase.  The benefit of purchasing a good is the health/happiness bonus it provides.  A shelf, for instance, could provide +1 health.  A bath gives +2 health.  A bottle of perfume gives +1 happiness.  So if holding 100 units of wealth is considered equivalent to +0.1 happiness, then spending anything over 1000 wealth for a good that gives +1 aggregate bonus would be "not worth it".  This would set a hard upper limit on prices and may not be desirable (especially since units of wealth can be defined however you wish in a developing economy, such as having units in terms of seashells or in horses or in gold coins).



Price Per Unit of Aggregate Bonus
An easy way to differentiate different goods is to think of them in terms of "price per unit of aggregate bonus".  If a good costs 120 wealth and gives 6 health+happiness,  then we can think of the price as "20/bonus".  Another good may cost 90 and gives 2 bonus, thus it has a price of "45/bonus".  We would purchase the 120 wealth good.

This helps to differentiate between goods and their bonus values, which helps to deal with unreasonable prices in a market with many different goods.

Item Goal List
There are issues with attempting to figure out reasonable pricing especially when some items are legitimately much more expensive than others.  One such example is housing versus household goods.  Another may be raising a horse versus buying daily food.

There should be a priority to the list of goods that an individual wishes to purchase.  Then money is set aside for each category, perhaps according to a budget.  Then, we attempt to find the lowest price for a good in that category.  Then, finally, we apply a rule to guess at whether that price should be accepted or not (with respect to price expectation and reasonable maximum price).

This way, an individual can intelligently "save up" for a big-ticket item, rather than always balk at the price of a house because milk is so much cheaper.


Trying to Make Luxuries Cost More Than Common Variants

Another issue that arose from the free market that was not discussed was the heavily depressed prices of luxuries which occurred for two main reasons.  Luxuries lasted forever but common wares had to be constantly repurchased (therefore driving up demand for common wares and driving down demand for luxuries).  Second, the lack of disposable income due to high common ware prices hampered the luxury market to the point of oblivion.

Product Stratification
For luxury goods, unlike common goods, you could possibly have more than one of the same type and gain bonuses from all of them. 

As an example, imagine a category of goods as "dining tables".  The health/happiness bonus from "dining tables" is based on the best table a person owns (best is based on aggregate bonus).  If one owns twenty "common" dining tables, then they can only receive a single bonus from the best common dining table.  But, if a person owns a luxury dining table, after a set amount of time it becomes an antique dining table.  At this point, it always gives a bonus whether it is the table in use (then it gives its full bonus) or not used (then it gives a highly reduced bonus).  This makes it useful to buy another luxury good of the same category once an old one as achieved "antique" status.  To limit the bonus, only one luxury item can be trying to achieve antique status at a time.

Shopping Algorithm
The current algorithm attempts to fulfil a need through a common item before ever considering a luxury item.  They should be considered side by side, with a pricing analysis that takes into account the bigger bonus possible from a luxury item (for instance, the price per unit of aggregate bonus discussed above).