Model-Based Algorithms
Latent learning separates the occurrence of learning from its immediate expression in behavior.
Learning Before Visible Action
An agent can learn about an environment before its behavior clearly shows that learning. This separation between learning and its immediate behavioral expression is called latent learning. The knowledge is acquired during one period, but it may become visible only later, when a goal makes that knowledge useful.
A Map of Relationships
Tolman connected latent learning with the idea of a cognitive map. A cognitive map is a learned model of an environment or task space. It represents relationships within that space, including routes that can later support reaching a goal. The important idea is that the model can be learned even when no reward or penalty immediately makes the learning visible.
A cognitive map is more than a record of one reinforced sequence. It is a learned model of relationships in the environment or task space. That distinction matters because a model can support selecting a different route when circumstances change.
Exploration Without Immediate Reward
From Exploration to Goal-Directed Planning
An animal explores an environment without receiving a reward or penalty for doing so. Later, reaching a goal becomes important. How can Tolman's interpretation explain a later route to that goal?
1. Exploration: The animal explores the environment without receiving a reward or penalty. Its behavior may not provide an obvious sign that learning has taken place.
2. Learned model: According to Tolman's interpretation, the animal can learn an environment model during exploration. This model represents relationships in the environment.
3. New motivation: Later, reaching a goal becomes important. The goal makes the previously learned model relevant to planning.
4. Route selection: The animal may use what it learned to plan a route. If it uses a different route rather than simply repeating a practiced sequence, the behavior suggests flexible planning based on the model.
Learning occurred before it was visible in goal-directed behavior, and the learned environment model later supported flexible route planning.
From Cognitive Maps to Algorithms
The connection to model-based algorithms is the use of an internal model for planning. In Tolman's account, exploration produces a learned model of the environment or task space. When a goal becomes important, the agent can use relationships in that model to consider routes and select behavior. The behavior is therefore based on more than repeating a response that was previously reinforced.
Why Route Flexibility Matters
A different route to the same goal is important evidence in this explanation. Simple repetition would reproduce a practiced sequence. By contrast, using another route suggests that the agent is using relationships represented in an environment model. The route change therefore illustrates flexible planning rather than simple repetition.
Common Interpretive Errors
Assuming that no immediate behavioral change means no learning occurred.
Latent learning separates the occurrence of learning from its immediate expression in behavior.
Fix:
Ask whether the learned model could become visible later when a goal makes it relevant.Treating the later behavior as mere repetition of a reinforced sequence.
A different route suggests flexible planning based on relationships represented in an environment model.
Fix:
Distinguish repeating a practiced response from using a learned model to plan.Assuming that a reward or penalty must have caused the original learning.
Rewards or penalties can make an already learned model relevant to planning without being required for the original learning.
Fix:
Separate the conditions under which the model was learned from the later conditions that made it useful.
Apply the Model-Based View
An agent explores a task space without receiving a reward or penalty. Later, a goal becomes important, and the agent reaches it using a route different from a previously used route. Explain why this pattern supports an interpretation based on latent learning and an environment model.
Hints
- First identify when learning could have occurred.
- Then explain why the later goal changes the relevance of the learned information.
- Finally explain why a different route suggests flexible planning rather than simple repetition.
Practice Response
Explain the pattern using the concepts from this article.
Latent learning: The agent may have learned during exploration even though its behavior did not immediately show that learning.
Cognitive map: The exploration may have produced a learned model of relationships in the task space.
Motivation: The later importance of the goal made the existing model relevant to planning.
Flexible route: Using a different route suggests that the agent used the model to plan rather than merely repeat a practiced sequence.
The pattern is consistent with latent learning followed by model-based, flexible planning when the goal became motivating.
Key Takeaways
- Latent learning is learning that is not immediately reflected in behavior.
- Tolman's cognitive map is a learned model of an environment or task space.
- Exploration can produce a model before a reward or penalty makes that learning behaviorally obvious.
- When a goal becomes important, the model can support planning and route selection.
- Using a different route suggests flexible, model-based planning rather than simple repetition.
Key Takeaways
- Latent learning can occur before it becomes visible in goal-directed behavior.
- Tolman's cognitive map describes a learned model of relationships in an environment or task space.
- A newly important goal can make an already learned model relevant to planning.
- Flexible route selection connects cognitive-map-based planning with model-based algorithms.