Decision Making Flashcards
Decisions in everyday life
happens unconsciously most of the time
Small decisions during your day:
Decisions can be seen as a link between memory and future actions
A lot of decisions are based on experience from past memory
These decisions will influence future actions
Memory is updated/influenced after experiencing consequences of your actions; “was the experience better/ as you expected?”
Decision-making is not an isolated process.
How does it relate to LTM?
You can think of decisions as a link between memory for past experiences that help to guide future actions.
Need to activate your past experiences in memory to inform your decision. Usually, you make a decision in order to do something. And in most cases, your decisions will be informed by what you have experienced in the past.
When making the decision, you might think of pleasant memories when you experienced a really good mood.
These memories then drive your decision to go to the Peak District.
Experiences = stored/ accessed in LTM
You will use the information from long-term memory to form predictions about the outcome of your decision. You will predict that your experience this time will be very similar to last time or very different, depending on what your last experience was.
What is the Prediction-Choice-Outcome Loop?
It explains the relationship between predictions, choices, and outcomes
Form of a loop: because the outcome of past choices influences our decisions in the future
We make:
decisions based on our goals
predictions of what we are expecting, after weighing all the pros and cons of different options
form a decision and take appropriate actions that should get us closer to our goal
Internally compare experiences
This outcome will be subjected to internal monitoring processes if our decision and the corresponding actions have achieved our goal (or at least brought us closer to it).
If we haven’t reached our goal?
Brain generates a prediction error:
a signal indicating how large the discrepancy is between what we had predicted originally and what the actual outcome was.
- Use these prediction errors to adjust future expectations
- Update our memory to more precise decisions the next time
when we face the same or a similar choice.
For example, your goal could be to have a tasty meal at a restaurant where you haven’t been before. When looking at the menu, you make predictions on how much you would like each option based on your experience in other restaurants. You then decide to order something (-> this is the action) that seems to be very similar to a dish that you enjoyed a lot elsewhere. When your meal arrives, you realize that you don’t like it as much as the version you had in mind. This difference between your anticipated meal (= that it would be delicious) and the meal that was served (= i.e. not as good) reflects the prediction error. You can use this prediction error to update your memory (you do not order this meal in this specific restaurant again).
This will help you to make better decisions next time.
There is a circular relationship between predictions (derived from memory content), decisions and outcomes which will be used to guide future predictions.
Biases in decision-making might have been developed to cut down the time it takes to make a decision (which can be a cost in itself).
Name some examples of biases in decision-making:
Various biases in DM: from economic games
1- Stick with same option you have chosen before
[stick with default]
eg. same food choice on a long menu, reducing decision time
2- Choosing certain gains over gambling situations
3- Choosing gambles over certain losses
4- Temporal discounting
choosing immediate rewards!
over future rewards (unless benefits are made explicit)
Which term s used to describe a signal indicating how large the discrepancy is between what we had predicted originally and what the actual outcome was?
Prediction errors
Prediction-Choice-Outcome Loop
What are the general features/ aims of decision making?
(We rely on memory content to make predictions.)
Also need to make predictions of possible decision outcomes to optimize our decision-making process.
What are the factors to consider before making a decision?
-Difficulty of the action (effort)
Need to execute in order get some form of reward or avoid a punishment (long exercise/ high cognitive effort to solve prob)
-Probability of success and failure
Is it likely that I will succeed? Or is there a high risk of failure?
(Risk evaluation)
-Value of reward/ choice
- Value of reward might change depending on the context/ current goals a given = reward may be more/ less appealing
(eg. a chance to win a pizza as price for a quiz might be of really high value to you if you are hungry. But not if you just ate.)
-Missed opportunities
- We decide for one option, we often decide against other options.
Often cannot go for several options
(travel by train= miss out on travelling by plane excluding the other option)
Name the 2 different levels of decision-making:
Simple perceptual decisions
-perceptual decisions
you have to decide which colour a stimulus has had or what sound
you have heard
More complex decision
- Taking several factors into account to make a decision
a wider range of factors that will influence your decision
compared to the simple decisions
(consider costs, colours, time consumption)
Levels of decision-making:
Which level consists of several factors being taken into account?
More complex decisions
Levels of decision-making:
Which level consists of perceptual decisions?
Simple decisions
Researchers can also experimentally control noise levels in these
tasks. This is important because in everyday life, we also do not always have very clear information available. On a perceptual level, our view on an object might be partly obscured, or more abstract information that we require for a decision might be uncertain.
Evidence accumulation in
Simple perceptual decisions:
Perceptual decision task
Explain the noisy sensory signal
A noisy signal, that needs to be evaluated over a certain period of time, which is converted into a discrete motor action
(move left or right) when the monkey makes this decision.
The decision process is quite noisy so u need time to evaluate and come up with a decision
- time needed to accumulate the evidence, then monkey makes the decision which is transferred into a discrete motor act
This is not just the case in a random-dot lab task, but in real life the sensory input is also often not clear immediately, e.g. when our view on objects is partly obstructed or symbols or keys, that we need to process for our decision, look very similar.
Accumulating evidence in perceptual decisions:
they start to fire when they detect more evidence
Neuronal recordings in monkeys suggest that this is what is happening in the brain during such a perceptual DM task:
When the stimulus, consisting of random-moving dots (and some coherently moving dots), is presented, neurons that are tuned to detect a specific motion direction will start to fire. As you might know from other lectures, neurons in sensory brain areas are tuned to a preferred feature. For instance, some neurons respond most strongly to leftward moving stimuli while other neurons respond most strongly to upward moving stimuli, etc. The more dots are moving in a given direction coherently, the stronger these corresponding motion detector neurons will fire. The firing rate will increase as more evidence for a given motion direction is being accumulated. You can see this in the illustration on the right side. Evidence accumulation for leftward motion is shown at the top, evidence accumulation for rightward motion is shown at the bottom. The evidence for a left motion direction increases faster than the evidence accumulation for a right motion direction. The curves indicating the firing rates/evidence accumulation are not straight, but go up and down a bit around a general upward trend. This reflects the noise in the stimulus. The more dots move coherently in one direction, the stronger the evidence for this direction. This would be reflected in the evidence accumulation curve being steeper and reaching the threshold faster.
Studies suggest that evidence accumulation always increases up to a certain threshold. When this threshold is reached a decision will be made in line with the evidence leading to a corresponding action, i.e. if the evidence accumulation curve for the left motion direction reaches this threshold first, the monkey will decide to make a left-sided response.
If the evidence accumulation curve for a right motion direction would reach this threshold first, the monkey would decide to make a right-sided response.
If the stimulus is quite noisy, that is, there is more distracting information present and the relevant information is not as clear, the evidence accumulation curves would reach the decision threshold closer in time and sometimes would lead to incorrect decisions, if the accumulation process for the less dominant direction meets the decision threshold first.
Simple perceptual decisions:
Which brain areas does evidence accumulation take place in?
Brain areas responsible for encoding the relevant feature,
e.g. area MT/V5 if motion is relevant for decision
You need to specify with the motor system is responsible for the decision making:
Ie. if the task is colour detection then colour coding areas will be involved in perceptual decision-making tasks
But recent evidence: sensorimotor areas (Parietal and dorsal prefrontal cortex) are representing possible actions, accumulating evidence as well
What are the 3 stages of perceptual DM?
Detection of sensory evidence; What are the alternatives that can be detected (left/right, red/blue, etc.)?
Integration of evidence over time
-> because evidence is noisy
Checking if threshold has been reached
-> if so, elicit appropriate action
-> if not, accumulate more evidence
There are 3 stages in perceptual DM:
The first is: Detection of sensory evidence; Here individuals identify what kind of sensory evidence can be detected (e.g. all possible colours or motion directions) and start to accumulate the evidence
Then, the evidence is integrated over time to detect the relevant signals among the noise.
As a third stage, a mechanism is assumed that regularly checks if a threshold has been reached. When this is the case, an appropriate action will be elicited. If the threshold has not been reached yet, more evidence will be accumulated.
Subjective values of options:
Areas associated with subjective value of decision options
(Fellows, 2018):
Reward value is reflected in activity strength in relevant brain areas: ↑ activity = ↑ expected reward value
Damage to the striatum may disrupt some aspects of reward learning.
The more activity in these areas, the higher the expected reward value of a given option.
Suggest that goals are represented in the medial OFC.
Suggest that the vmPFC and ventral striatum track the expected value in line with the current goals
(i.e. the value representation will change for a given option depending on the current context)
Name a brain area associated with decision making:
Lateral prefrontal cortex (PFC):
But active in many decision paradigms
(NOT involved in value-based choices)
What is Evidence accumulation?
How the brain combines the info stored in memory with new incoming info
Memory must both be durable and flexible:
2 competing models of how this can be implemented in the brain
Name the 2 models of how evidence accumulation can be implemented in the brain:
A) Homogeneous model
B) Heterogeneous model
Evidence accumulation:
Memory must both be durable and flexible:
2 competing models of how this can be implemented in the brain
Homogeneous model
Homogeneous model:
All relevant neurons (to encode evidence) become active at the same time. The more evidence presented, the greater the activity shown in cells.
Sensory events make a wave of activity
Evidence accumulation:
Memory must both be durable and flexible:
2 competing models of how this can be implemented in the brain
Heterogeneous model
Heterogeneous model:
Neurons don’t become active at the same time,
Early-responding neurons become active quickly and pass on activity to other (slower-responding) neurons.
Creates a wave of activity in the network.
Evidence collectively grouped to respond to create memories that are flexible to adapt new info.
The accumulated evidence in the network then reflects when the decision threshold will be reached. By having both neurons that react fast to incoming new information and those neurons that react in a more sustained way over longer periods of time, the memory for accumulated evidence can both be flexible and durable.
=predicted accumulative evidence
Mental maps in decision-making
Hypothesis:
Decision making processes rely on internal models of the current task so experiences need to be organized in internal models or mental maps
Explain this:
We have an internal representation stored in our memory that provides us with some guidance about the different options that are available in a choice situation; this fits with the loop idea that I’ve presented earlier. The internal model helps us to predict the different outcomes of the available options based on our experiences.
For instance, if you decide which route to take for your commute home, you will have an internal model representing the different options, including their advantages and disadvantages. You will also have some experience regarding which route is better on certain days of the week or during certain times of the day based on your experience of traffic jams in the past, for example.
In order for the internal models to be as accurate as possible, new experiences that contribute to the accuracy of our internal models, need to be added to these models or mental maps.
For instance, you might experience that there is a new construction site on one of the routes, so your internal model needs to be updated with this information.
Spatial tasks such as long ques on Fridays are easier to think about and explain
To illustrate where these hypotheses (mental maps) in decision-making came from:
Tolman had rats experiencing a spatial maze
The shortest path to the goal was blocked so the rat had to go all the way round.
‘A’ and ‘B’ here just indicate spatial locations in the maze for illustration purposes, but are not actually indicated in any way to the rat.
Findings showed?
Rats created a mental map representation of the maze
Points a and b were closer together taking the shortcut
-encoding the relation between points a and b
If then the path was unblocked later, the rats quickly realized that they could take a shortcut and go directly from A to B to reach the goal faster.
Thus, the rats seem to have positions A and B encoded as being close together in space, although they had never experienced going straight from A to B.
This means that the rats had encoded transitive relations of the different locations in the maze and had used this information to build a mental map of the maze.
This mental map informed their decision at the first intersection to go straight ahead instead of taking the path they had been trained on.
Mental maps can be transferred to non-spatial tasks:
Give an example
e.g. how to code an experiment in Psychopy
This experience will get embedded in a cognitive map on how things work and relate to each other. This cognitive map will guide your decisions on how to set up an experiment the next time you’ll be confronted with a similar task. Thus, you will choose the shortcut to a solution immediately and make your decisions accordingly.
So this idea of mental maps is not restricted to spatial maps, but can be applied to paths for problem solving as well.
