Competition. Alerts generate BU flow in the alert network, which excites its node’s DM network to yield TD DM flow. The existence of a surprising alert means that there is no adaptive response that is specific to the alerting event. Instead, many relevant adaptive responses are mobilized, and compete to be included in the emerging flexible response. Relevance is determined by the reach of alert flow, a function of the present context (the nodes already excited when the alert arrives) and past responses (existing UQS connec- tions). In particular, nodes representing responses to events that share some feature with the current one are excited.
The mechanism that implements the selection of a focused response from the candi- date pool is calledcompetition. Competition and subsequent sustained quax execution via
synchrony are mediated by brain-widecoordinationnetworks consisting of GABAergic inhibitory interneurons (IINs) (i.e., locally projecting neurons). IINs sometimes also partic- ipate in quaxinitiation. Hence, the role of GABA in R17 is to support effortful execution, contrary to the standard descriptions viewing it as inhibiting execution34.
JOS. Coordination utilizes an IIN-mediatedjoin-or-stop (JOS)mechanism that works as follows. Suppose that two neurons have similar physiological characteristics and are targets of the same oscillating inhibitory IIN. If their energy drives are of comparable magnitudes (e.g., when they represent the same thing), and if the IIN’s firing rate is not faster than their charging rate, the IIN’s inhibitionsynchronizestheir firing (it resets their electric charges, allowing them to start charging at the same time. Due to their comparable charging rate, they then fire at the same time). We refer to this as a join operation. If one neuron is supported by a strong drive while the other is not, the IIN wouldstop(silence) the weakly driven one. If the IIN fires faster than the two charging rates, it would silence both neurons. Thus, the effect of activated IINs is to magnify signal (defined as being in synch with the IIN) and quench noise. We say that two excitatory neurons engage in acompetitionif an IIN needs more than one firing to affect JOS.
Any innervation of an IIN by an excitatory neuron can be viewed as a coordination request. Coordination neurons are driven by excitatory network flow, both locally and over long-distance connections. When flow enters an area and activates its excitatory neurons, these quickly activate its coordination networks, to separate the area’s excitatory neurons into those according with the flow (join, definingwinners) and those not according with it
34Large exogenous amounts of GABA can inhibit execution by taking advantage of its mechanistic action,
(stop, defininglosers). Thus, JOS is an arbitration mechanism between neurons for getting included in the quax. After competition resolution, JOS sustains local synchrony between the winners. Synchrony is propagated because single IINs innervate many content neurons, and because IINs are connected by gap junctions, which facilitate synchronized firing (see below).
We refer to the IINs active in a quax as itsscaffold, and to its content and Rgen neurons as itsskeleton. The skeleton determines the currently executing actions, the object(s) that they act upon, and why they are executed. The scaffold maintains skeleton neurons in local synchrony and protects from interference from other excitatory neurons. The skeleton uses glut in the abrain, and glut and acetylcholine (ACh) in the ibrain. The scaffold uses GABA in the abrain, and glycine and GABA in the ibrain35.
Substantial evidence supports the role of GABA in competition resolution. Particularly convincing are the glut and GABA uncaging experiments described in [Hayama et al., 2013], where GABA induced widespread spine shrinkage (a short term structural plasticity phenomenon reflecting local loss of competition) across the dendrite, except at the spine belonging to the quax skeleton.
Disinhibition. To prevent erroneous responses, responses are suppressed by default through- out the brain. In several key junctions, this is done by tonically active inhibitory neurons. Elsewhere, it is done via a general ambient inhibitory tone. In the DM mode, many different response alternatives should be allowed to compete, and this is achieved by inhibition of the default inhibition. In other words, competition involves the opening of a wide disinhibitory constraint-free path (a ‘hole’) in the default inhibition. After competition resolution, only the winning alternative keeps being allowed (disinhibited), and the default inhibition of the losers is reinstigated36.
The main tonically active disinhibitory junctions are at the basal ganglia (BG)-thalamus, cerebellar cortex-output nuclei, PAG-medulla (innates), and BG-superior colliculus (eye movements)37. The main non-tonically active disinhibition is in the competition coordina-
tion network (CCN), see below.
Automaticity. Repeated task training that exposes the brain to a rich variety of input combinations leads to a UQS that provides good BU flow separation, whereby neurons on the correct response path receive high flow while potential competitors receive weak flow. In this situation, a response can be found in a single BU pass, with minimal TD flow (mainly consisting of precise predictions) and competition. Thus, automaticity involves BU triggering and reduced alerts and TD control38. Conversely, acute events require TD flow to resolve competitions.
If the neurons about to win a competition are already partially excited before sensory input arrives, competition is resolved faster. There are two common cases in which this occurs, repeated immediate execution of the same quax (see repetition suppression below) and anticipations and predictions that match the input. Improved prediction accuracy is another consequence of repeated training on a task, since quax transitions are repeated with training and allow learning processes to refine the UQS.
35The skeleton and scaffold are also supported by additional Rgens, see below.
36Tonically active inhibition is also useful for enabling rapid responses, because excitation of the response
neurons can start before the response should be executed (since it is inhibited). That is, relatively complete responses can be prepared before competition is even seriously started, without risking premature execution.
37The retina also uses disinhibition (BU), at least in some species.
38Note that it is difficult to execute automatic routines in attended TD mode (e.g., try to tie your shoelaces
slowly with attention). With automated tasks, the brain’s connections are optimized for BU flow, so forcing the usage of TD flow disrupts execution.
Initiation via GABA. GABA is the brain’s main inhibitory agent. Nonetheless, it can be excitatory under certain conditions. GABA works through two families of receptors, ionotropic GABAA receptors, which act quickly through ion channels, and metabotropic GABAB receptors that induce slower effects [Benarroch, 2012]. GABAARs are comprised of five subunits of various isoforms. Synaptic GABAARs mainly contain the alpha1 or al- pha2 isoforms, and extrasynaptic GABAARs mostly contain the delta isoform. When they bind to GABA, GABAARs induce the opening Cl- channels. Flow direction depends on the Cl concentration gradient between the cell’s inside and outside. If the intracellular con- centration is higher, GABAAR binding results in Cl- outflow, a situation calledexcitatory (depolarizing) GABA.
Early in development, the transporter NKCC1 continuously moves Cl- into the cell, resulting in excitatory GABA. During development, NKCC1 is downregulated and the potassium-chloride cotransporter KCC2, which moves Cl- out of the cell, is upregulated [Ben-Ari, 2014]. In the common adult state, the Cl- equilibrium potential is more nega- tive than the resting membrane potential. In this case, GABAAR binding lets Cl- into the cell and GABA is hyperpolarizing. However, depolarizing GABA is not limited to early development, and occurs in the adult state after initial strong stimulation, e.g., in the retina [Lindstrom et al., 2010], hypothalamus [Bains, 2014], spinal cord, amygdala, cerebellum, and hippocampus [Marty and Llano, 2005]. Acute situations yield excitatory GABA by downregulating KCC2 via BDNF (a structural plasticity agent released during acute re- sponses) [Rivera et al., 2004] and/or by alpha1 NERs indicating alert [Bains, 2014].
Thus, at least in some cases, GABAergic IINs induce quaxinitiation, providing neu- rons with an initial excitatory push after surprising or strong sensory input. GABA-mediated coordination includes quax initiation, quax formation by disinhibition and competition, and sustained quax execution via synchrony.