Author's note:
This research was conducted by the author, as part of their dissertation project (Popa, 2013). It is supporting material for the thesis outlined under "The Shape of Will":
Popa, A. (2019). Psychology 2.0: The Emergence of Individuality. https://doi.org/10.31234/osf.io/m87an
It provides an example of how bio-sociocultural variables work together to produce ADHD symptoms in living organisms. Parts of this work were discussed in: Popa (2014 - 2017); Popa & McDowell (2011); Popa, Calvin, & McDowell (2012, 2013, 2014); in Popa & McDowell (2016); in McDowell (2019; p.141-142); in Calvin (2019; p. 10-11, etc.), ...
Disrupted Neuronal Dynamics as the
Formal Cause for ADHD symptomatology
Andrei Popa
Department of Psychology, Emory University, Atlanta, USA
The ADHD literature abounds in knowledge about contributing factors, both biological and environmental (Mash & Wolfe, 2013), but their mechanisms of action are still to be identified (Mash & Wolfe, 2013; Mick et. al., 2010). Research on an evolutionary theory of behavior dynamics (ETBD; McDowell, 2004, 2013b) showed that certain experimental contexts produced behavioral symptoms that shared descriptive and functional similarities with core symptoms of ADHD (Popa, 2013). The purpose of this section is to explain how ADHD-like symptoms emerged in ETBD and to explore the implications for the emergence and developmental trajectory of ADHD symptoms in humans.
Learning as evolution. The evolutionary account of behavior dynamics conceptualizes behavior as a tool that operates on the environment, with more or less success. Behavioral strategies that prove successful tend to be repeated; those that do not, tend to be abandoned (Skinner, 1981, 1984).
ETBD. The computational model discussed here (McDowell, 2004) instantiates the Skinnerian analogy between positive reinforcement and natural selection (Skinner, 1981). If learning - long-term change in behavior and knowledge due to experience - is an emergent property of evolutionary-like principles, then synthetic behavior produced by Darwinian processes should be indistinguishable from the behavior of living agents. For details, see ETBD.
Corroborating evidence. Extensive experimentation showed qualitative and quantitative agreement between ETBD and real-world behavior (for a review see McDowell, 2013b).
Predictions and translational relevance. In addition to recreating known phenomena, ETBD was shown to predict human behavior (Popa, 2013, Phase 2); to inform about mathematical descriptors of behaviors (Popa & McDowell, 2016); and to shed light on the nature and developmental trajectory of ADHD (Popa, 2013 - 2017; Popa & McDowell, 2011, 2016; Popa, Calvin, & McDowell, 2012, 2013, 2014).
ETBD and ADHD
Popa (2013) showed that under certain experimental conditions (described later), the preference patterns produced by ETBD were strikingly similar to textbook descriptions of ADHD-diagnosed children. When compared to "controls", "impulsive agents" exhibited:
“… diminished sensitivity to reinforcement, high frequency of switching between alternatives, low frequency of target responses and acquired reinforcers (Kollins et. al., 1997; Taylor et., al., 2010), short amounts of time spent on task, decreased productivity, rapid task abandonment, difficulties re-engaging in task behavior once the activity was abandoned (Mash & Barkley, 2003), high levels of motor activity, easily distracted by interfering stimuli/activities (Abikoff & Gittelman, 1985; Mash & Barkley, 2003; Waslick & Greenhill, 2004)."
The figure shows two sets of behavioral characteristics. The only difference between the experimental contexts that produced them was the rate of mutation. This is a property of the digital agent (non-environmental) that takes values between 0% and 100%. ADHD-like symptoms emerged when mutation rate was higher than 30-40% (Popa, 2013, page 42).
Good environments improved the behavior of impulsive agents, whereas bad environments worsen their symptoms; in extreme cases, they disrupted the behavior of controls to the extent that it was impossible to distinguish between controls and impulsive agents (Popa, 2013; pp. 43-45).
Polar charts can be used to display and compare multiple behavioral measures at once (Popa & McDowell, 2016). All graphs depict z-scores, unless specified otherwise (for details, see DATA 4000). The ten y-axes go from -3 to +3, with -3 at the center; the white, dotted line represents zero (0), the mean of the z-distribution. Each surface provides a comprehensive visual overview of the effects of an experimental condition on behavior as a whole. Each surface corresponds to an average z-score (top-right corner). Note that for seven variables, low scores indicate high variability, but for changeovers, inter-bout intervals, and topographic variability (delta_phenotype), low scores indicate low behavioral variability. For consistency, their corresponding z-scores were reversed (R) so that small surfaces and low z-scores correspond to disorganized behavior.
"control"
"impulsive" agent
Good environment
control
impulsive
Bad environment
control
impulsive
Great
Bad bad
Equifinality and Multifinality
Similar to the behavioral symptoms observed in ADHD, disorganized behavioral patterns emerged from many combinations of independent variables; the six panels describe outcomes observed under different experimental conditions. Note the quantitative and qualitative (visual) resemblance between panels 2 and 5, and 3 and 4, respectively. If such similarities would be observed in human participants, it would be difficult not to think of them as subtypes.
1
2
3
4
5
6
Implications for ADHD diagnostic
Examining the behavioral constellations above, is difficult to infer the specific configuration of variables that produced it. This is relevant for clinical research because assignment to control and experimental groups is done based on symptoms similarity, which may obscure etiological variability.
This suggests that the studies that failed to find significant connections between dopamine-related genes and ADHD (Li, Sham, Owen, & He, 2007) may have overlooked some environmental variables that could have prevented (or lessened) their expression (e.g., high reinforcement density).
Similarly, in studies that failed to find clinically-relevant improvements in ADHD symptoms from behavioral interventions, participant’s biological predispositions may have overcame the effects of environmental variables. This would explain, for example, why the performance of ADHD-diagnosed children varies with context (Corkum & Siegel, 1993) and why, although many ADHD-diagnosed clients benefit from behavioral interventions, some benefit a lot more than others, as suggested by a recent examination of ~170 studies (Fabiano et. al., 2009).
Finally, they provide an explanation for the differences observed in medication effectiveness: is not unreasonable to assume that environmentally-induced variability, a natural reaction to resource scarcity, (Neuringer, 2004; 2009) may respond less to medication than biologically-induced variability.
In ETBD, these effects were produced by the interaction between high mutation rate, reinforcement rate, magnitude, and changeover delay; in ADHD, these apparent inconsistencies have been attributed to Gene x Environment interactions (G x E; Laucht et. al., 2007).
Clinical research and practice would benefit from complementing assessment of symptoms with a functional analysis of the bio-social context that produced said symptoms. Such analyses, however, cannot be conducted without knowing how and where genes and environments meet and interact and how ADHD symptoms emerge from these interactions.
ADHD as an emergent manifestation of
disrupted neuronal dynamics
ETBD has no goals and no memory. All behavioral constellations emerge, unguided, from the reiteration of Darwinian processes, which transform each generation of integers into a new one.
Since emissions are chosen at random (from the existing population) it follows that the only way in which computational parameters can influence long-term outcomes is indirectly, by altering the composition of each generation, hence the likelihood of emitting one behavior or another.
By examining the low-level effects of these variables, I noticed that the experimental contexts that produced disorganized behavioral patterns had one thing in common: they were altering the population structure in ways that decreased the probability to exhibit consecutive target responses, i.e. "to stay on task" (Popa, 2013, page 46-47).
High mutation rate. Mutation diminishes the "concentrating" effects of selection, events, introducing spontaneous variation in the new population. A high rate of mutation, it follows, can cause even very fit populations to dissipate extremely fast, thus counteracting the effects of even very strong selection events.
Low reinforcement rate/magnitude. Selection events concentrate the "next" population of behaviors around the reinforced emission. Non-reinforced emissions, on the other hand, causes the parents for the next generation to be chosen at random (and not on their fitness). The effects of selection, therefore, dissipate with each non-reinforced emission, at a rate inverse with the strength of selection. The effects of seldom or weak selection events dissipate faster than the effects of frequent/strong selection events.
High conduciveness. The Hamming changeover delay - the average number of bits necessary to "shift" behavior from one class to the other - does not affect the rate of dissipation or the level of concentration, so they don't affect behavior quantity, but its allocation among the target classes, with low values increasing the likelihood to change-over to the "other" class.
From bits to neurons
The moment-to-moment activity of biological organisms is produced by groups of neurons (Edelman, 1978), hypothesized to constitute a material counterpart for ETBD’s integers (McDowell, 2010). Neurons are interconnected entities, located in a 3-dimensional space, a 3-dimensional graph in which the structure remains fixed, but the configuration of activation states changes. This means that the succession of generations in ETBD corresponds to the succession of configurations of active neurons that underlie behavior. So whatever changes we see in behavior, must have a dynamic counterpart in the alternation of activation states of the neuronal groups that underlie it.
In ETBD, disorganized behavioral patterns emerged in contexts that prevented the formation of bouts, contexts described by various combinations of reinforcement rates, reinforcer magnitudes, mutation rates, and Hamming arrangements. Their real-world counterparts may act in a similar manner, by altering the succession of neuronal configurations in ways that disrupt sustained behavior and sustained attention.
High activity in the brain's Default Mode Network (DMN)
Popa and McDowell (2016) hypothesized that mutation rate may be functionally equivalent to the level of activation of the brain’s default mode network (DMN), a collection of interconnected regions that shows strong spontaneous activation at rest (Buckner, Andrews-Hanna, & Schacter, 2008; Raichle et al., 2001). Failure to suppress this network during tasks that require sustained attention or response inhibition was associated with increased levels of behavioral variability (Weissman, Roberts, Visscher, & Woldorff, 2006; Feige, Biscaldi, Saville, Kluckert, & Bender, 2013), presumably by interfering with the activity of task-positive regions, involved in goal-directed activity (Kelly, Uddin, Biswal, Castellanos, & Milham, 2008).
In ETBD, high mutation rates disrupted sustained behavior (i.e., bout-responding) by constantly “dispersing” the population of integers, therefore counteracting the effects of selection events. In nervous systems, high levels of DMN activity may act in similar ways, by constantly disrupting the patterns of neuronal activity that underlie sustained behavior and sustained attention.
PREDICTION. Considering the interaction between mutation rate and reinforcement rate / magnitude in ETBD, the effects of high DMN activity should be more pronounced in bio-sociocultural contexts that contribute to low activation of the reward system, like reduced number of neurons, dopamine deficits, low frequency of events with reinforcing potential, or in contexts that facilitate rapid switching between alternatives (i.e., low changeover delay; Findley, 1956), such as working on an essay while the instant messaging app is set to "always on top of current window", cell phone within reach, etc. (page number). In contrast, the effects of high DMN activity should be counteracted in contexts that facilitate the activation of the brain’s reward system (e.g., stimulants, interesting activities, rich social interactions, etc.).
Inadequate activation of NS’ reward system
The low reinforcement rates and magnitudes arranged in ETBD could correspond to low activation of the brain’ reward system, which, in turn, could be caused by many variables involved in ADHD: low dopamine levels (Faraone & Mick, 2010; Volkow et. al., 2009), reduced number of neurons (Krain & Castellanos, 2006; Malenka, Nestler, & Hyman, 2009), decreased speed of neuronal communication (D'Agati, Casarelli, Pitzianti, & Pasini, 2010), non-reinforcing environments, psychosocial adversity, family conflict; Biederman, Milberger, Faraone, Kiely, Guite, Mick, Ablon, Warburton, Reed,1995; Nigg, 2009), genetic polymorphisms (e.g., LPHN3, Arcos-Burgos & Muenke, 2010), and so on.
Regardless of what variables interfere with the system's functioning, the outcome is disorganized behavior: few, short bouts of sustained activity, high frequency of changeovers, low sensitivity to reinforcement, etc. In everyday human environments, these discontinuities could be interpreted “easily distractible”, “forgetful”, “careless mistakes”, “driven by a motor”, “leaves their seat”, and so on, depending on the expectations associated with the context in which behavior is recorded.
Low Changeover Delay (COD)
The Hamming_COD in ETBD was shown to correspond to a real-world changeover delay (Popa & McDowell, 2010): a short time interval, contingent on switching between alternatives (i.e., changing over), during which reinforcement is withheld (Findley (1956). A period of 1-2 seconds was often enough to noticeably increase sensitivity to reinforcement and to reduce the frequency of changeovers; increasing the length of the interval beyond what seemed like a critical threshold had no further effects (Shull & Pliskoff, 1967; Baum, 1974, 1979; Davison & McCarthy, 1988; Temple et. al., 1995). Similarly, Taylor et al, (2010) showed that a 2-second COD increased sensitivity to reinforcement in ADHD-diagnosed children, but did not affect controls. These results were in agreement with ETBD (Popa & McDowell, 2010), confirming the equivalence between Hamming_COD and real-world COD.
Popa (2013, Section 2) took this verification one step further: he examined two groups of non-diagnosed college students in a fast matching procedure. One group experienced a 2-second COD; the second group experienced a 0-second COD. The No_COD group exhibited more variable behavior on eight of eight behavioral dimensions.
How does a COD affect the neuronal activity
that underlies choice behavior?
The fast-pace of the procedure and the barely noticeable differences between reinforcement rates led the author to hypothesize that students' behavior was not influenced by explicit cognitive strategies. Rather, the fast pace of these procedures tapped directly in the adaptive potential of the nervous system, bypassing cognition. Is possible that during the COD interval the neuronal activation associated with pre-changeover responses dissipated, thus reducing proactive and/or retroactive interference. This raises the possibility that fast matching procedures can be used to explore memory phenomena
Molecular Dynamics
(follow up)
In ETBD, it seemed that the experimental contexts that produced disorganized behavioral patterns were doing so by altering the population's structure in ways that decreased the probability to exhibit consecutive target responses, i.e. "to stay on task" (Popa, 2013, page 46-47).
To verify this hypothesis, I ran the experimental conditions described here (two examples are shown below) again and recorded the entire population of behaviors, at each generation. This allowed me to compute, for each generation, the probability to emit a response in the same class or the other class, where "same" was defined as the most recently reinforced target class. For each condition, I sampled 10 generations before and after each selection event, and computed the average moment-to-moment probability to respond in the previously reinforced target class (i.e. "same" class).
control, impulsive
bad environment
control, impulsive
good environment
To be continued.