Foraging in the Global Workspace: The Central Executive Reconsidered

David L Barack, Postdoctoral Research Fellow, Salzman Lab, Columbia University

What is the cent­ral exec­ut­ive? In cog­nit­ive psy­cho­logy, exec­ut­ive func­tion­ing con­cerns the com­pu­ta­tion­al pro­cesses that con­trol cog­ni­tion, includ­ing the dir­ec­tion of atten­tion, action selec­tion, decision mak­ing, task switch­ing, and oth­er such func­tions. In cog­nit­ive sci­ence, the cent­ral pro­cessor is some­times modeled after the CPU of a von Neumann archi­tec­ture, the mod­ule of a com­pu­ta­tion­al sys­tem that makes calls to memory, executes trans­form­a­tions in line with algorithms over the retrieved data, and then writes back to memory the res­ults of these trans­form­a­tions. On my account of the mind, the cent­ral pro­cessor pos­sesses the psy­cho­lo­gic­al func­tions that are part of exec­ut­ive func­tion­ing. I will refer to this com­bined con­struct of a cent­ral pro­cessor that per­forms exec­ut­ive func­tions as the cent­ral executive.

The cent­ral exec­ut­ive has a range of prop­er­ties, but for this post, I will focus on domain gen­er­al­ity, inform­a­tion­al access­ib­il­ity, and infer­en­tial rich­ness. By domain gen­er­al, I mean that the cent­ral exec­ut­ive con­tains inform­a­tion from dif­fer­ent mod­al­it­ies (such as vis­ion, audi­tion, etc.). By inform­a­tion­ally access­ible, I mean both that the cent­ral executive’s algorithms have access to inform­a­tion out­side of the cent­ral exec­ut­ive and that inform­a­tion con­tained in these algorithms is access­ible by oth­er pro­cesses, wheth­er also part of the cent­ral exec­ut­ive or part of input or out­put spe­cif­ic sys­tems. By infer­en­tially rich, I mean that the inform­a­tion in the cent­ral exec­ut­ive is poten­tially com­bined with any oth­er piece of inform­a­tion to res­ult in new beliefs. The func­tions of the cent­ral exec­ut­ive may or may not be conscious.

Three con­cepts at the heart of my mod­el of the cent­ral exec­ut­ive will provide the resources to begin to explain these three prop­er­ties: intern­al search, a glob­al work­space, and foraging.

The first concept is intern­al search. Newell fam­ously said that search is at the heart of cog­ni­tion (Newell 1994), a pos­i­tion with which much mod­ern cog­nit­ive neur­os­cience agrees (Behrens, Muller et al. 2018; Bellmund, Gärdenfors et al. 2018). Search is the pro­cess of trav­el­ing through some space (phys­ic­al or abstract, such as concept space or the inter­net) in order to loc­ate a goal, and intern­al search refers to a search that occurs with­in the organ­ism. Executive func­tions, I con­tend, are types of search.

The second concept in my ana­lys­is is the glob­al work­space. Search requires some space through which to occur. In the case of cog­ni­tion, search occurs in the glob­al work­space: a com­pu­ta­tion­al space in which dif­fer­ent data struc­tures are loc­ated and com­pete for com­pu­ta­tion­al resources and oper­a­tions. The glob­al work­space is a notion that ori­gin­ated in cog­nit­ive the­or­ies of con­scious­ness (Baars 1993) but has recently been applied to cog­ni­tion (Schneider 2011). The glob­al work­space can be con­cep­tu­al­ized in dif­fer­ent ways. The glob­al work­space could be some­thing like a hard drive that stores data but to which many dif­fer­ent oth­er parts of the sys­tem (such as the brain) sim­ul­tan­eously have access. Or, it could be some­thing like an arena where dif­fer­ent data struc­tures lit­er­ally roam around and inter­act with com­pu­ta­tion­al oper­a­tions (like a lit­er­al imple­ment­a­tion of a pro­duc­tion archi­tec­ture; see Newell 1994; Simon 1999). The cent­ral exec­ut­ive is partly con­sti­tuted by intern­al search through a glob­al workspace.

The third and final concept in my ana­lys­is is for­aging. Foraging is a spe­cial type of dir­ec­ted search for resources under ignor­ance. Specifically, for­aging is the goal-directed search for resources in non-exclusive, iter­ated, accept-or-reject decision con­texts (Barack and Platt 2017;Barack ms). I con­tend that cent­ral exec­ut­ive pro­cesses involve for­aging (and hence this third concept is a spe­cial case of the first concept, intern­al search). While cent­ral exec­ut­ive pro­cesses may not lit­er­ally make decisions, the ana­logy is apt. The intern­al search through the glob­al work­space is dir­ec­ted: a par­tic­u­lar goal is sought, which in the case of the cent­ral exec­ut­ive is going to be defined by some sort of loss func­tion that the sys­tem is attempt­ing to min­im­ize. This search is non-exclusive, as oper­a­tions on data that are fore­gone can be executed at a later time. The search is iter­ated, as the same oper­a­tion can be per­formed repeatedly. Finally, the oper­a­tions of the cent­ral exec­ut­ive are accept-or-reject in the sense that com­pu­ta­tion­al oper­a­tions per­formed on data struc­tures either occur or they do not in a one-at-a-time, seri­al fashion.

The ana­lys­is of the cent­ral exec­ut­ive as foraging-type searches through an intern­al, glob­al work­space may shed light on the three key prop­er­ties men­tioned earli­er: domain gen­er­al­ity, inform­a­tion­al access­ib­il­ity, and infer­en­tial richness.

First, domain gen­er­al­ity is provided for by the glob­al work­space and unres­tric­ted search. This work­space is neut­ral with regard to the sub­ject mat­ter of the data struc­tures it con­tains, and so is domain gen­er­al. The search pro­cesses that oper­ate in that work­space are also unres­tric­ted in their sub­ject matter—those pro­cesses can oper­ate over any data that matches the input con­straints for the pro­duc­tion sys­tem. (While they may be unres­tric­ted in their sub­ject mat­ter, they are restric­ted by the con­straints on the data imposed by the pro­duc­tion system’s trig­ger­ing con­di­tions.) The unres­tric­ted sub­ject mat­ter of the glob­al work­space and the unres­tric­ted nature of the pro­duc­tion pro­cesses both con­trib­ute to the domain gen­er­al nature of the cent­ral exec­ut­ive. This ana­lys­is of domain gen­er­al­ity sug­gests two types of such gen­er­al­ity should be dis­tin­guished. There are con­straints on what type of con­tent (per­cep­tu­al, motor­ic, gen­er­al, etc.) can be con­tained in stored data struc­tures, and there are con­straints on the type of con­tent that can trig­ger a trans­form­a­tion. A domain gen­er­al work­space can con­tain domain spe­cif­ic pro­duc­tions, for example.

Second, inform­a­tion­al access­ib­il­ity reflects the glob­al workspace’s struc­ture. In order to be a glob­al work­space, dif­fer­ent modality- or domain-specific mod­ules must have access to the work­space. But this access means that there must be con­nec­tions to the work­space. Other aspects of inform­a­tion­al access remain to be explained. In par­tic­u­lar, while the glob­al work­space may be widely inter­con­nec­ted, that does not entail that mod­ules have access to inform­a­tion in spe­cif­ic algorithms in the work­space. The pres­ence of a work­space merely insures some of the needed archi­tec­tur­al fea­tures for such access are present.

Third, infer­en­tial rich­ness res­ults from this intern­al for­aging through the work­space. Foraging com­pu­ta­tions are optim­al in that they min­im­ize or max­im­ize some func­tion under uncer­tainty. Such optim­al­ity implies that the executed com­pu­ta­tion reflects the best data at hand, regard­less of its con­tent. Any such data can be util­ized to determ­ine the oper­a­tion that is actu­ally executed at a giv­en time. This explan­a­tion of infer­en­tial rich­ness is not quite the sort described by Quine (Quine 1960)or Fodor (Fodor 1983), who envi­sion infer­en­tial rich­ness as the poten­tial for any piece of inform­a­tion to influ­ence any oth­er. But with enough simple foraging-like com­pu­ta­tions and enough time, this poten­tial wide­spread influ­ence can be approximated.

These com­ments have been spec­u­lat­ive, but I hope I have provided an out­line of a sketch for a new mod­el of the cent­ral exec­ut­ive. Obviously much more con­cep­tu­al and the­or­et­ic­al work needs to be done, and many objections—perhaps most fam­ously those of Fodor, who des­paired of a sci­entif­ic account of such cent­ral processes—remain to be addressed. I intend on flesh­ing out these ideas in a series of essays. Regardless, I think that there is much more prom­ise in a sci­entif­ic explan­a­tion of these cru­cial, cent­ral psy­cho­lo­gic­al pro­cesses than has been pre­vi­ously appreciated.

 

References:

Baars, B. J. (1993). A cog­nit­ive the­ory of con­scious­ness, Cambridge University Press.

Barack, D. L. and M. L. Platt (2017). Engaging and Exploring: Cortical Circuits for Adaptive Foraging Decisions. Impulsivity, Springer: 163–199.

Barack, D. L. (ms). “Information Harvesting: Reasoning as Foraging in the Space of Propositions.”

Behrens, T. E., T. H. Muller, J. C. Whittington, S. Mark, A. B. Baram, K. L. Stachenfeld and Z. Kurth-Nelson (2018). “What is a cog­nit­ive map? Organizing know­ledge for flex­ible beha­vi­or.” Neuron100(2): 490–509.

Bellmund, J. L., P. Gärdenfors, E. I. Moser and C. F. J. S. Doeller (2018). “Navigating cog­ni­tion: Spatial codes for human think­ing.”  362(6415): eaat6766.

Fodor, J. A. (1983). The mod­u­lar­ity of mind: An essay on fac­ulty psy­cho­logy, MIT press.

Newell, A. (1994). Unified Theories of Cognition, Harvard University Press.

Quine, W. V. O. (1960). Word and object, MIT press.

Schneider, S. (2011). The lan­guage of thought, The MIT Press.

Simon, H. (1999). Production sys­tems. The MIT Encyclopedia of the Cognitive Sciences: 676–677.