Next: Learning in Tolman's Mazes Up: Tolman's Cognitive-Maps Task Previous: Theoretical Assumptions Contents
The 'design' of the experimental setups which have been proposed by Tolman, his students and his colleagues is 'inspired' by the leading hypothesis described above.
Thus we have to assume at least the following variables for a 'learning system (LS)':
Several possible settings are possible. In the following I list only those, which are mentioned explicitly by Tolman and which we did use in our experiments.
Not being hungry:
If a learning system is not hungry then it has within the above framework no clear 'goal'. It can dwell round in a maze without any preferences. This is also independent from some preceding learning.
Being hungry/ thirsty:
Here we have to distinguish two subcases:
Hungry/ Thirsty while learning:
Encounter food/water while being hungry/ thirsty induces a change in the behavior in the manner that the LS can 'shorten' the time to find the food/water in the next trial. This works just from the 'beginning' of learning as well as after n-many trials without food/water.
Hungry/Thirsty after learning:
If the LS finds food/water 'as usual' nothing happens. If it does not find food/water at the 'learned place' then this event can 'weaken' the tendency to go to this place in the future again.
The usual timing assumes 9 - 22 days, one trial per day. In the case of the 'delayed rewarding' one introduces the food/water after about 10, 11 trials (days).
From this we extract the following two main cases:
In case one (cf. figure 11.3) the learning system is 'hungry' from the beginning and therefore is directly searching for food. A solution is to find food at some place.
In case two (cf. figure 11.4) the learning system is 'not hungry' from the beginning and therefore there is no immediate 'need' to search food. The learner can 'look around' 'for free'.
In a combination - let us call this case 3- one can start with case two (not being hungry) and after some runs where the learning system could dwell in the maze it will be made hungry before the next runs. Then it has the 'need' to find food.
According to Tolman and others will case three reveal that the learning system has learned in the 'play mode' enough to be able to go directly to the goal when urged to be hungry.
Another case -here called case 4- could be that one after the successful learning of a goal changes the position of the food: removing it from the old place and place it at a new position. Whatever has been learned before has to be 're-learned'!
With regard to all yet mentioned cases 1-4 it can make a difference 'how long' the search path is. Assuming a 'long enough' path, then a movement e.g. to a direction d1 can occur in different 'contexts': sometimes it is followed by a direction d2, sometimes d3, etc. Therefore if the learning system shall be able to distinguish all these cases it must be able to learn 'more complex' concepts which can represent 'larger units' of property sets. We call this case 5 whereby case 5 can be combined with any other case 1-4 before. Therefore we will use the following table of basic properties combined in an experiment:
| EXP.NO | CASE | CASE |
|---|---|---|
| 1 | 1 | -- |
| 2 | 1 | 5 |
| 3 | 3 | -- |
| 4 | 3 | 5 |
| 5 | 4 | -- |
| 6 | 4 | 5 |
A typical experiment with rats presupposes some maze with an entrance, different possible path segments, some of those as 'blind alleys', others as 'leading to a goal', which is a 'food/water box', where the rat can 'eat/drink'. After having find food/water a rat is taking of from the maze.
We have designed a first example of a TOLMAN MAZE No.1 as follows:
TOLMAN1 = !O O O O O O O ! ! ! !O . O O O F O ! ! ! !O . O O O . O ! ! ! !O . . . . . O ! ! ! !O O O . O O O ! ! ! !O O O O O O O !
This is a 6 x 7 maze with 6 rows and 7 columns. The learning system will always start at position (5,4):
-->TOLMAN1(5,4)='*1' TOLMAN1 = !O O O O O O O ! ! ! !O . O O O F O ! ! ! !O . O O O . O ! ! ! !O . . . . . O ! ! ! !O O O *1 O O O ! ! ! !O O O O O O O !
Our second maze according to the paper of Tolman [380] is the following maze called TOLMAN2:
!. . . . . . . . O O O . . . . . . . . . . ! ! ! !. . . . . . . . O . O . . . . . . . . . . ! ! ! !. . . . . . . . O . O . . . . . . . . . . ! ! ! !O O O O O O O O O . O O O O O O O O O O O ! ! ! !O . . . . . . . . . O . . . . . . . . F O ! ! ! !O O O O . O O O O . O O O O . O O O O O O ! ! ! !. . . O . O . . O . O . . O . O . . . . . ! ! ! !. . . O . O . . O . O . . O . O . . . . . ! ! ! !O O O O . O O O O . O O O O . O . . . . . ! ! ! !*1 . . . . O . . . . . . . . . O . . . . . ! ! ! !O O O O . O O O O O O O O O . O . . . . . ! ! ! !. . . O . O . . . . . . . O . O . . . . . ! ! ! !. . . O . O . . . . . . . O . O . . . . . ! ! ! !. . . O O O . . . . . . . O O O . . . . . ! ----------------------------------------------------
Because this maze is really 'large' from the point of larning we have inserted different positions where the food can be placed to allow more 'simpler' learning tasks (cf. figure 11.5).
![\includegraphics[width=4.0in]{Task1_Tolman.eps}](img687.png)
![\includegraphics[width=4.0in]{Task2_Tolman.eps}](img688.png)
![\includegraphics[width=5.0in]{TOLMAN2PartsAll.eps}](img617.png)