Decision making and free will Flashcards
In simplified terms: What two factors drive human behavior when it comes to decision-making?
1.) An estimation of what is perceived
2.) Preferences
What statistical framework heavily influenced the neuroscience of decision-making?
The signal detection theory (SDT)
Briefly explain the main components of the SDT. How is information accumulated and plotted?
In signal detection theory we make a row of sequential, uncorrelated observations at different time points (e.g. the intensity of a single pixel on a radar operator screen, every second). Based on each value a decision is made, whether a particular stimulus is present (e.g. an airplane or no airplane). Both stimuli must have equal probabilities. Doing so will allow us to build two histograms that are of gaussian distribution.
Given these histograms, explain what each distribution represent, and what the Y and X axis are.
One of the two distributions represents a baseline signal (B), and the other represents a deviant signal, made up of the baseline signal and an increment (B+ΔS). Both distributions plot a signal intensity (e.g. the activation of a neuron or the intensity of a radar signal) against the probability that a certain stimulus is present.
Given the basic structure of the SDT, name the main two central parameters of SDT, and what they represent.
The two main parameters are the criterion and d’prime. The criterion represents the cut-off value of the x-axis that determines whether a certain signal intensity is rated as a stimulus is present or not. D’prime represents the degree of overlap between both gaussian distributions and hence also represents the discriminability of stimulus.
What four response categories arise from placing a criterion?
By placing a criterion in an absent stimulus distribution (B) you can either achieve a correct rejection or a false alarm. In a present Stimulus (B+ΔS) you will have hits and false alarms.
How does changing the criterion parameter in SDT influence the presence of hits and false alarms and misses and correct rejections?
By placing the criterion higher you increase the number of misses and decrease the number of hits in (B+ΔS) whilst increasing the number of correct rejections and decreasing the number of false alarms in (B). You would so to say minimize the number of false alarms at the cost of more misses.
What is D’prime and what does it take into consideration in its computation?
D’prime represents the discriminability and amount of overlap between the given gaussian distributions. It expresses the distance of means in units of (pooled) standard deviations of the signals.
Is there an alternative account for how d’prime can be computed?
Yes. If there is a homogeneity of variance and both given distributions are of gaussian nature, you can calculate late d’prime by subtracting the hit rate (HR) from the false alarm (FAR) rate.
Additional information:
This is possible due to the fact that the hits and misses as well as correct rejections and false alarms both amount to 100% of their respective distributions. Consequently, HR and FAR are both expressed by H/1 and FA/1 and can be calculated as (H-FA) /1
What is a ROC? What does it express and how is it modulated through d’prime and why is it sensible to report this in papers?
ROC is the receive operating curve. It plots Hits as a function of false alarms (bang for the buck principle). The larger d’prime is, the stronger the curve climbs before plateauing, and hence the Area under the curve (AUC) is maximized. This shows how well distributions can be discriminated. It does not require a lot of prerequisites.
How and why is SDT applicable in the neuroscience of decision-making?
If our decision is dependent on identifying a certain stimulus, SDT is important to select a certain response .
In decision-making neuroscience, SDT can also be used to research the threshold for the detection of stimuli. What could be observed and how is it connected to SDT?
Over a large number of trials, participants were presented with stimuli varying in intensity. Participants should then report whether they could perceive the stimulus. The closer we get to the lower threshold of perception the stronger the overlap of the stimulus-present and the stimulus-not-present distributions overlap.
We need to define a certain criterion to discern if a stimulus is present or not (e.g. 50%). There might not be a definite threshold for stimulus perception, and if so it would not be particularly statistically stable, as the chance level is more important.
Why is there a gradual and not definite threshold in perception experiments with the SDT?
Stimulus fluctuations (e.g. photon noise) can influence or perception
Sensory processing is not precisely repeatable, as the brain operates with probabilities and has inherent variability (even if the same process is repeated, there might be different ions)
Central processing may result in differences in attention wich might make things less repeatable
What is a Random dot kinematogram (RDK) and what is the central parameter of this paradigm?
RDK is a task in which participants are presented with a cloud of randomly moving dots of homogenous velocity. The amount of dots moving in the same direction is manipulated (i.e. coherence). In this paradigm, information is accumulated across time and an (arbitrary threshold) is set based on the level of coherence at wich direction can be correctly identified given a proportion of trials.
Explain how the RDK is implemented in the Shaiden & Kiani (2013) study on decision-making in macaques.
An RDK was placed in the monkey’s receptive field and two target stimuli were presented left and right of it. the monkey had to observe the direction of the point’s movements and had to perform a saccade after a short delay.
The directional-movement sensitive cell of MT, as well as the LIP cell that indicates the location a saccade will be made (i.e. preparation to move to such cell) to are observed.
How does the MT cell respond to null and preferred movements of RDK movements in relation to their coherence?
With low coherence, there is little discriminability between preferred and null movements. The higher the coherence gets the cell responds with more spikes per trial for the preferred direction it is tuned to, than to the null direction.
Consequently, we can deduce, that MT cells have information on the direction of movement.
How can we more precisely determine the amount of information an MT cell has on movement direction in the RDK experiment with macaques?
By treating the distribution of responses of the MT cell to preferred and null movements we can apply the SDT and discern Correct and incorrect judgments. We can then assess the proportion of correct/all judgments and plot this against the coherence level.
What is a neurometric function? How does it compare to a psychometric function? Explain this with the 2013 Shaiden & Kiani (2013) study on decision-making in macaque monkeys.
A neurometric function is a function, that measures how well one can access stimulus information from the brain
By assessing the proportion of correct answers and plotting it against the coherence level we receive a neurometric function that describes how well an MT cell can discriminate between movement directions in relation to coherence.
In this case, the neurometric needs less coherence than the psychometric function to reach a certain predefined threshold.
This means that the macaque monkeys neurons have more information for the decision than they show in their behavior.
Explain how the accumulation model of MT and LIP for decision-making works. How are MT and LIP modulated through stimulus intensity and how do they interact?
We have a coding of sensory evidence in the sensory area (e.g. MT) which is constant throughout the trial and this sensory evidence is accumulated/integrated across time this is what we have in the parietal area, which again reflects the output of the decision-making process
MT cells activation is constant and proportional to stimuli intensity, LIP cells activation are a function of information accumulation of MT cells, which makes their incline in activation over time proportional to MT activation
Is information accumulation over time truly necessary for decision-making?
Some studies suggest, that “at glance” information may be sufficient for early choice signals in high-level cortical regions.
Explain why the ramping model of evidence accumulation in the LIP of macaques might be challenged.
Using bayesian modeling a research group from Princeton university could show, that discrete steps from no decision to decision may account for information more fitting than the ramping diffusion-to bound model suggested earlier.
