Arousing decisions: Environmental Cues to Danger Rationally Alter Evidence Integration and Impulsivity

Windy Torgerud, University of Minnesota Dominic Mussack, University of Minnesota  Paul Schrater, University of Minnesota

Abstract


Arousal in animals is a fundamental behavioral state with a functional role of preparing the animal to respond to possible threats in the environment.  However, in psychology and neuroscience, arousal is widely viewed as a physiological state, largely due to its ease of measurement through its impact on the autonomous nervous system.    Here we show how arousal can be manipulated and its functional consequences on decision making are measured and isolated.   We do X and we show Y.   

Introduction

Decision making occurs under a variety of different contexts each of which place competing demands on the decision maker.  In natural environments, the importance of one particular decision will be regularly modulated by signals emanating from the environment. Signals indicating potential nearby threats or social cues such as peer competition can increase or decrease the salience of a particular decision relative to everything else currently occurring in the action-ably nearby environment.

Structure of the argument:
1) Animals need to modulate decisions to deal with threat.  This modulation is graded, use the Ecology of fear Mobbs paper to frame the issue Mobbs 2015.
2) Depending on the context, there are different responses that are most appropriate for the animal. To monitor the threats you may have to split attentional resources, either across space or time, or completely disengage from the current task to appropraitly respond to an immediate threat. Sarter 2006 (Lang 2010)
3) Arousal is a key manifestation of threat preparedness  Lo 2008 (survival optimization system (Mobbs 2015), however, other types of ecological responses will engage physiological arousal, but for different purposes (eg. mating. reward. positive valence, circumplex model of emotion). 
4) In this view, physiological arousal will not be a good predictor of decision modulation, so we use a data-driven methodology to find brain/body signatures associated with decision modulation
5) We predict that decisions will be impacted by a process that frees up attentional resources across decisions to monitor the background.  Resources could be split equally at all times, but visual information seeking requires temporal allocation (cite Rothkopf, PNAS), which incentivizes freeing up time by making certain decisions faster.
6) This is a time allocation problem, trading-off on-task performance vs. background monitoring.
7) These predictions can be mapped into two hypotheses about the impact of threat stimuli.  a) threat should be TRACKED, integrating cues across time, using a time window.  b) increases in threat should alter decision making by decreasing threshold, and increasing gain, to free up time to look while minimizing performance impact.  

Results

3 claims we have evidence for (drawing the connections):
1) discovered a latent state influencing decisions, which we show using PLS & diffusion model, that we argue represents a CNS signal corresponding to arousal that gears up brain for threat via resource reallocation.  This is complementary to Arousal as PNS (autonomic) gearing up the body by energetic resource availability and increase in sensory and motor.  Overall, there is a conceptualization of threat signal processing as being integrated in the brain, and this signal broadcast out to other brain areas to effect a coordinated change in both cognitive and metabolic resource availability and allocation.  To better show the distinction between PLS defined "brain/body" arousal, we need to include more standard arousal measures as predictors, and let them try to explain decisions in regression.  We expect this will fail to add anything to our predictions of decisions, moreover, the arousal signals we get this way may be better tied to stimuli.   This analysis seems to need a comparison predictability of stimuli vs. predictability of decision RT.  
Predicting RT is different than predicting stimuli.  We show that stimuli are poor predictors of RT - that means they are essentially independent of each other.   Now it's interesting which measures predict stimuli well, and which predict RT well. 

2) this arousal state tracks the environmental stimuli; shown via the time-course of latent parameters and linear systems
3) there is a tradeoff in time; indicated by the alternating drift and threshold parameters 


Replace this text with your caption
Figure 1. A fox out hunting a hare. While stalking the hare, the fox has to make a quick decision to dive for it, expecting the hare to run behind the trees to the top or bottom. Then the fox hears a noise in the bushes behind it. The fox has many predators itself, including wolves and coyotes, so should the fox run for safety, keep stalking the hare, or try for both?
Story setup:
A fox is walking through the woods when she spots a glimpse of a rabbit jumping behind a tree ahead of her. She crouches down so as to avoid being seen by her prey. Her ears are forward and her eyes are sharp. In her stalking approach she hears a sudden rustle in the bushes behind her. Fearful of becoming a meal herself she freezes and turns her ears back to listen to the sound. Shifting her head slightly in that direction she sees a gray paw peeking out of the bushes to her rear. Having been spotted the wolf lunges forward to attack the fox. As the fox had detected the threat she is able to escape just in time, though her stomach still rumbles, she'll have to find another rabbit or mouse to eat later, but at least for now she is still alive. 
Intro 1.  Animals need to modulate decisions to deal with threat

Animals making decisions engage in an economic cost-benefit analysis; how do I maximize my benefits and minimize my costs of different options (Stephens and Krebs 1986)? However, when performing any activity, animals must account for the costs of predation, i.e. the chance of death (Stephens 2007Lima 1990). While an animal can engage in activities to reduce this, such as hiding or being vigilant, constant vigilance produces its own costs, such as starvation or never reproducing. This forces a tradeoff in the allocation of time, attention, and other resources between the foreground task the animal is engaged in and the background threat of predation. Resource allocation is a graded response based on the priority of either outcome - a graded modulation of fearful responses - including different ways of splitting attentional resources across either time or space (Trimmer et al 2013, Mobbs 2015).

In the worst case, threat is immanent and requires direct engagement; either fight or flight. In these cases the priority of the threat takes precedence over whatever other action the animal was engaged in, and available resources are allocated to dealing with the threat. But this response depends critically on the degree of immanence; threats that are less immanent should be monitored rather than directly engaged. This gradient of threat responses means that animals allocate more or less resources to monitoring the threat based on perceived threat level (Mobbs 2015, etc). Animals can take active measures to increase their chance of detecting a predator, and will increase their vigilance based on perceived threat.

Vigilance in ecology is often operationally defined based on head movement of foragers (e.g. Treves 2000, Blanchard et al 2011, Bednekoff and Lima 1998, Shettleworth 2010). For example most birds have to alternate between scanning their environment looking for danger with their head up, and foraging for food with their heads down. In general an increase in threat will decrease the time allocated to vigilance behavior for these animals (Wahungu et al 2001, Gustafsson et al 1999, Stephens 2007).

Of course vigilance is operationally defined differently depending on the animal, as the allocation of attention in this way depends crucially on both the threat level and sensory system used. Attentional processing can sometimes be allocated without a change in overt behavior, as seen in split-attention paradigms (e.g. Broadbent 1958, Dukes and Kamil 2000). If attention can be allocated without behavior changes, as might be possible for auditory processing, then attention might not need to be allocated across time as a distinct vigilance task. However the same is not true for visual processing in many animals including humans, since high resolution needed for predator identification is limited to the fovea. For visual tasks this produces a time tradeoff, where visual attention cannot be split at one time and must be split across time (Nobre 2001, Hoppe and Rothkopf 2016). 

Intro 3. Arousal is a key manifestation of threat preparedness.
This threat monitoring and preparedness is often revealed in emotional states of fear and arousal. Arousal and fear can take on different levels, depending on the level of the threat. However, arousal itself does not represent an animal's threat level. Instead arousal can be a manifestation of multiple different factors, both positive and negative in rewarding value or valence. Different models of affect separate out arousal from valence, indicating that while high arousal can indicate fear or a high threat, it can also correspond with positive excitement or approach behaviors such as mating.
Intro 4. Data-driven methodology to find brain/body signatures associated with decision modulation
Because arousal can correspond to multiple different ecological responses, physiological arousal by itself is not a good predictor of an animal’s decision. Generally arousal is measured directly via heart-rate, however in the literature ‘arousal’ is often operationally defined as or measured by heart rate alone. Arousal, in terms of how it impacts decisions under threat, is a combination of a host of biological and neural factors that result in the decision. An agent tracking the threat level of an environment should integrate
Replace this text with your caption
Intro 5 or 6.
Past research has investigated the temporal aspects of decision making, of when to make a decision (Ratcliff 1978, Shadlen and Newsome 2001, Smith and Ratcliff 2004, Wagenmakers 2009). This problem can be generally viewed as an optimal stopping problem, which has resulted in a set of theories that attempt to explain the response time of decisions (Wald 1948?, Luce 1986, Bogacz et al 2006). However most of these these theories only consider a single decision, and do not address the problem of how to allocate time across multiple decisions. In natural settings humans and animals face sequences of decisions, and time spent on one activity is time not spent on another. Importantly this applies to scenarios in which one of these decisions involves physical harm to an agent, something which can be a constant threat and so can influence seemingly unrelated decisions.
Resource Allocation
Kalman Filter threat integration model. placeholder image. actual image will be a qualitative look of kalman filter model data next to real data.

Results

Results claim 1:
Replace this text with your caption
Results claim 2:
this arousal state tracks the environmental stimuli; shown via the time-course of latent parameters and linear systems 
Maybe show this image side by side with a model generated image... to show what we'd be expecting to see if such and such were true, then show real data by it 
Results claim 3:
there is a tradeoff in time; indicated by the alternating drift and threshold parameters 
See if we can recreate the donut effect with a model generated data. if we do model generated data does donut show up or more gausian cloud? what would we have to change to get donut

References

  1. Dean Mobbs, Cindy C. Hagan, Tim Dalgleish, Brian Silston, Charlotte Pr??vost. The ecology of human fear: Survival optimization and the nervous system. Frontiers in Neuroscience 9, 1–22 (2015). Link

  2. Martin Sarter, William J. Gehring, Rouba Kozak. More attention must be paid: The neurobiology of attentional effort. Brain Research Reviews 51, 145–160 (2006). Link

  3. Peter J Lang, Margaret M Bradley. Emotion and the motivational brain.. Biological psychology 84, 437–50 Elsevier B.V., 2010. Link

[Someone else is editing this]

You are editing this file