“The affective gradient hypothesis: an affect-centered account of motivated behavior“: everyone agrees that feelings and actions are intertwined, but cannot agree how.
According to dominant models, actions are directed by estimates of value and these values shape or are shaped by affect. The article proposes instead that affect is the only form of value that drives actions. Our mind constantly represents potential future states and how they would make us feel. These states collectively form a gradient reflecting feelings we could experience depending on actions we take. Motivated behavior reflects the process of traversing this affective gradient, towards desirable states and away from undesirable ones. This affective gradient hypothesis solves the puzzle of where values and goals come from, and offers a parsimonious account of apparent conflicts between emotion and cognition.
Existing models of decision-making propose several potential roles for feelings (e.g., affect, emotion, mood):
(A) as a read-out of ongoing evaluations;
(B) as learning signals that are abstracted into an estimate of value or utility;
(C) as relevant (integral) or irrelevant (incidental) inputs to estimates of decision value; and
(D) as drivers of a separate (‘hot’ or Pavlovian) system for action control. These roles are illustrated individually, but contemporary accounts often combine several of these, in some cases differentiated by the types of feelings evoked (e.g., shorter- vs. longer-lasting).
(E) Independent of their specific role(s) for feelings, all of these accounts maintain that a separate system exists to evaluate potential outcomes in a way that abstracts from or sidesteps feelings.
The current account proposes that feelings alone (in the form of affect; cf. experienced utility) may be sufficient to account for all motivated behaviors, without the need for a separate system for ‘colder’/more goal-directed evaluations.
- Models of motivated behavior suggest that feelings (e.g., affect, emotions) interact with other types of evaluations to drive action. This article proposes instead that affect is, on its own, sufficient to drive all motivated behavior.
- People represent potential multiple potential affective consequences at any given time. These describe the ways in which expected affect can be improved or worsened by one’s actions (the affective gradient), which in turn drives dynamic adjustments in behavior.
- This affective gradient hypothesis centers on affect as the sole driver of both reflexive and deliberate behavior, replacing past accounts that propose distinct systems (e.g., hot vs. cold) or selves (e.g., present vs. future) underpinning these behaviors.
- This account also helps resolve a longstanding puzzle of how goals are prioritized and maintained. It proposes that goals are an emergent property of the affective associations under consideration at a given time
It is now well-established that humans and other animals maintain an internal model of their environment, consisting of rich and structured associative maps relating potential locations/contexts, objects, episodes, and concepts. These maps enable us to revisit past states of our environment and project into potential future states, a process that can transpire voluntarily (e.g., through directed search) or involuntarily (e.g., through an automatic spreading of associations; cf. priming).
These state-space representations form a critical interface between perception, memory, and action. I will argue that, in so doing, they enable action to be dynamically influenced by affective content embedded in these state representations. To ground this argument, I begin by drawing on existing work to extrapolate a set of basic principles that characterize this embedded structure (Figure):
- There are affective qualities or features to every state that a person can represent. The state of eating a meal, being given negative feedback, or paying one’s bills all have affective features that carry a specific identity (e.g., the taste of a particular food) and scalar intensities along a limited set of dimensions (e.g., valence and arousal) (Figure A). In other words, affect is both multivariate and – similar to perceptual features like color and depth – evoked by any stimulus or context that is brought to mind.
- Affective features can be evoked by experienced, recalled, or imagined states. Affective features are evoked while a person is in a relevant state (e.g., eating a meal) and when those states are brought to mind through recollection (e.g., recalling a recent meal) or prospection (e.g., imagining having that meal at a restaurant tonight) (Figure B).
- Affective features can be evoked in a ‘bottom-up’ or ‘top-down’ fashion. A particular state can be brought to mind via an external prompt (e.g., a poster for a particular restaurant) and/or subsequent spreading of associations (e.g., being reminded of meals you have eaten at this restaurant and other similar restaurants), thus evoking the affect tied to that situation in a ‘bottom-up’ fashion. These same states can also be brought to mind through forms of directed search (e.g., considering potential dinner options), constituting a ‘top-down’ route to accessing the same affective experiences (Figure C).
- Multiple affective features can be evoked in parallel. A person can (near-)simultaneously bring to mind states relevant, for instance, to: (i) an immediate decision (e.g., different approaches to writing a section of a grant); (ii) their overall task (e.g., whether to try to meet the upcoming grant deadline); (iii) other potential activities (e.g., checking their e-mail); and (iv) basic survival instincts (e.g., pain from a recent injury, hunger) (Figure D).
- The salience of an affective feature scales with the salience of its associated state. Anything that make a given state more accessible also increases the salience of associated affective features. Conversely, variables known to decrease accessibility (or render a state more ‘psychologically distant’) should similarly weaken the relevant affective experience – this includes factors that make a given outcome seem improbable, spatially or temporally distant, or otherwise unlikely to impact oneself. This should be true independent of the affective content, in that more vivid outcomes should produce more salient affective experiences whether intensely positive, intensely negative, or neutral. For instance, the grant writer might be motivated by the possibility of this grant being funded and/or the possibility of failing to secure any grant funding, but the affective features of these states may be less salient than those of states they perceive as more immediate and likely, such as the possibilities of successfully submitting the grant or missing the deadline (Figure E)
According to the current account, every internally represented state (e.g., episode) carries affective features.
(A) Each of these states (e.g., current hunger level, eating a muffin, submitting a grant late, having a grant awarded) are represented along a limited set of affective dimensions (e.g., valence and arousal).
(B) States can be brought to mind through experience, recall, or prospection (e.g., current, past, or future meal) and in each case evoke the associated affective features (e.g., positive feelings towards eating the food in question).
(C) States and associated features can be brought to mind through bottom-up cueing (e.g., an image on the screen) or directed search (e.g., choosing a lunch venue).
(D) A person can have in mind affective features associated with multiple states at the same time (or in rapid alternation) (e.g., current feelings of hunger and mental fatigue, future outcomes related to success or failure at the current task).
(E) States that are most accessible and/or vivid (e.g., ones that are perceived to be more immediate or likely) will have more salient affective features.
The landscape of affective features being held in mind at any given moment can also serve as a guide for how to adjust behavior in that moment.
States that are expected to improve one’s affect serve as attractive landmarks (those to be reached), whereas states that are expected to worsen one’s affect serve as repulsive landmarks (to be avoided).
The actions afforded by one’s current environment can each be described in terms of their relationship with these states: to what extent does taking that action increase the likelihood of reaching more positive states and/or avoiding more negative states?
(A) While working on a given task (e.g., grant-writing), a variety of states may become accessible, some with positive affective features (e.g. grant being awarded, uncovering the contents of an incoming e-mail) and others with negative affective features (e.g., hunger, mental fatigue, missing a deadline), each varying in its salience. Each of these may bring to mind actions (e.g., walking to store, attending to grant, attending to e-mail) that can make a given state more likely (unbroken arrows) or less likely (broken arrows) to occur.
(B) The relationship between a given action (e.g., attending to grant) and a given state (e.g., grant being awarded) can be described by a gradient, with increased attention making it more likely that the person will reach that state and achieve the associated affective experience (e.g., elation). Other actions (e.g., attending to e-mail) produce different gradients with respect to this and other potential consequences (e.g., feelings associated with reading the e-mail). The affective gradient hypothesis (AGH) proposes that actions are optimized so as to maximize expected positive affect and minimize expected negative affect, for instance, in this case by maintaining focus on the task at hand.
(C) Applying this model to a value-based decision task (e.g., whether to eat sushi or tacos), both AGH (blue) and traditional models (yellow) predict that actions should be guided by the affect expected from obtaining a given option (e.g., based on past experiences with each). AGH diverges from these models in how it accounts for one’s engagement in the task (e.g., level of attentional focus, threshold for making a decision). Traditional models assume that the task serves as its own goal and participants adjust control when performance worsens (e.g., by monitoring for errors or conflict). AGH instead proposes that task engagement is directly determined by the salience of potential performance outcomes (e.g., likelihood that one would be perceived to be performing the task poorly) and how these outcomes would feel. Thus, actions (e.g., left vs. right) and control states (e.g., levels of attention and threshold) are both determined by expected affect. Rather than maintaining a task goal explicitly, a person need only represent contingencies between action and performance (e.g., task rules) and the consequences of performing well or poorly. This account generalizes across task rules, including those that focus only on visual features of stimuli (e.g., size or form) rather than their affective features.
Each action can thus be described as occupying a location along a multidimensional gradient, with this location identifying the expected affective consequences of taking that action. For instance, focusing on the task you just started reduces the likelihood that you will miss a deadline or have to stay late; checking your e-mail reduces aversive uncertainty and/or increases the potential for positive surprise arising from a recent notification; and stopping to eat your lunch reduces growing hunger and momentarily increases feelings of satisfaction and enjoyment.
Recasting intrapersonal conflict: one self, many feels
Accounts of self-regulation describe conflicts that arise between one’s current inclinations and those of an idealized actor, where the latter can reflect, for instance, projections to a future self (e.g., one that has health, wealth, and happiness) or other forms of social norms or moral ideals (e.g., being a good person, doing what’s right). Popular (yet controversial) accounts cast these intrapersonal conflicts in terms of competing agents (selves) within one’s mind, each with its own objective function (e.g., maximizing immediate vs. long-term reward, serving personal vs. group/societal interests).
Variability in motivated behavior within and across individuals
People vary considerably in their motivation to engage in various acts, in ways that can be maladaptive. At one extreme are cases where one lacks the motivation necessary to engage in daily activities (e.g., apathy, amotivation). At the other extreme are behaviors that can be described as reflecting an excess of motivation for outcomes that are detrimental to their long-term health and wellbeing (e.g., drug use). From the perspective of AGH, understanding what motivates a person to perform a certain action requires knowing what feelings they expect
(i) while performing that action (e.g., effort);
(ii) as a result of that action (e.g., approval);
(iii) as a result of inaction (e.g., reprimand); and
(iv) as a result of other actions (e.g., foregone opportunities).
For instance, someone can behave impulsively because of their positive feelings about the expected outcome (e.g., drug high), negative feelings about not achieving that outcome (e.g., frustrated craving), and/or lack of negative feelings about longer-term risks (e.g., addiction). Understanding how motivation differs across people, and over the lifespan, requires understanding variability in these same affective expectations.
Notably, these affective expectations each center on particular future outcomes (varying from immediate to longer term). Variability in these expected future states will not always be reflected in common instruments that assess summary estimates of a person’s feelings in the moment or in the recent past. Instead, a more comprehensive approach is necessary to inventory each of the outcomes a person considers when weighing a given activity; how salient (e.g., probable) those outcomes seem; and how they would feel if a given outcome were to occur (i.e., the affective features of this outcome) (cf. measures of outcome ‘expectancies’).
Affective goals can theoretically alter the motivational impact of a given affective configuration (e.g., high-arousal positive affect), such that states with these features are either more or less attractive or repulsive. This would, in turn, promote actions that increase the likelihood of achieving this experience (e.g., skiing vs. hiking).
There is also research to suggest that people vary in the levels of affect they find most desirable. For instance, some people strive for high-arousal positive states (e.g., elation), whereas others strive for low-arousal positive states (e.g., serenity), leading these two groups to pursue different kinds of activities (e.g., skiing vs. hiking). It is possible that some of these affective goals emerge from the combination of states that are most accessible to a person (e.g., people who engaged in more vs. fewer high positive arousal activities growing up) and their associated affective features (e.g., linking arousing activities with more vs. less downside potential). Alternatively, such goals could reflect individual differences in which configurations of affective features (e.g., settings of valence and arousal) are most desirable. This could be conceptualized as a ‘meta-parameter’ that alters the orientation of affective features with respect to potential actions, effectively motivating behavior towards maximizing the desired levels of affect, whatever these may be
Surtil : 