Dealing with Unknown Effects

If the effects of certain actions are described relatively (e.g., temperature increases relative to the current temperature), then ALFA grounds such effects based on the current conditions using rules (6.32) and (6.33).

• If there is an intention to get or reGet a set of results $\beta$ and there is some $\beta'$ in $\beta$ without an $\effect$ assertion that relates $\beta'$ directly to a set of actions that can cause $\beta'$, then use the current knowledge base to create an $\effect$ assertion. If $\beta'$ involves having the property $Prop$ of an object $Obj$ at value $Val$ and if there are $\effect$ axioms that associate set of actions to changing the value of $Prop$, then use those $\effect$ axioms and the current $\observed$ value of $Prop$ to determine the new effect assertion. This is given by rules (6.32) and (6.33).

  $$\begin{split}& \intention(Id, Type, \beta, T_1, T_2)) \wedge ... ...owright \effect(Id, \alpha, \beta', \tau_1, \tau_2) \end{split}$$ (62)

  
  

  $$\begin{split}& \intention(Id, Type, \beta, T_1, T_2) \wedge (... ...owright \effect(Id, \alpha, \beta', \tau_1, \tau_2) \end{split}$$ (63)

  
  
  Whether (6.32) or (6.33) gets triggered depends on the currently (or lastly) observed value of Prop. The presence of a new last-observed-value predicate $\observed$ causes previous last observed values to be distrusted; all the formulas originally derived from the newly distrusted formula, also get distrusted. As a result, an effect axiom associated with an old observation gets distrusted and a new effect axiom based on the new observation gets added. This provides a mechanism for acting on intentions to maintain certain results for lengthy durations. For instance, to maintain the temperature of an object at a specific value, an agent can perform one set of actions if the temperature goes above the specific value, and perform another set of actions in case it goes below, in order to bring it back to the required level.