The Behavior Analyst 2002, 25, 161-173 No. 2 (Fall) Cognitive Neuroscience from a Behavioral Perspective: A Critique of Chasing Ghosts with Geiger Counters Steven F Faux Drake University Cognitive neuroscience is a growing new discipline concerned with relating complex behavior to neuroanatomy. Relatively new advances in the imaging of brain function, such as positron emission tomography (PET), have generated hundreds of studies that have demonstrated a number of inter- esting but also potentially problematic brain-behavior relations. For example, cognitive neurosci- entists largely favor interpretations of their data that rely on unobserved hypothetical mechanisms. Their reports often contain phraseology such as central executive, willed action, and mental imagery. As B. F. Skinner argued for decades, cognitive constructs of neurological data may yield nothing more than a conceptual nervous system. Key words: cognitive neuroscience, PET, fMRI, ERP, brain imaging During the past 50 years, cognitive 2001). Further, many in cognitive neu- scientists have transformed their field roscience attempt to give a brain lo- by embracing a variety of different dis- cation to those unobserved processes ciplines and subdisciplines (Gardner, using gross measures of physiology 1985; Solso, 2001). For example, they (e.g., Roland, 1993). have made use of various versions of This paper will update an argument linguistics (e.g., Chomsky, 1959), phi- originally made by Skinner (e.g., 1938/ losophy (e.g., Fodor, 1975), symbolic 1991, 1950, 1953, 1974) that super- logic (e.g., Newell & Simon, 1976), imposing unobserved mechanisms and connectionist architectures (e.g., upon the brain results in little more Grossberg, 1988). This paper is a cri- than a "conceptual nervous system," tique of certain practices that charac- with a great potential to misguide (see terize a popular new version of cogni- also Uttal, 2001). Much of cognitive tive science called cognitive neurosci- neuroscience still relies on mentalistic ence. Its growing popularity is due, in forms of explanation that either explic- part, to its use of sophisticated brain- itly or implicitly appeal to an inner imaging technology involving positron agent, "the ghost in the machine," in emissions, magnetic resonance, and the words of Ryle (1949). Although the brain-electric fields. However, cogni- technology of cognitive neuroscience tive neuroscience has not improved is impressive (e.g., tracking gamma ra- upon the assumptions found in older diation in the brain), it is the opinion cognitive sciences. It often assumes of this author that those measurements that reasonable inferences about unob- amount to little more than chasing served neural mechanisms can be ghosts with Geiger counters. This pa- made from overt behavior (see Uttal, per argues that much of cognitive neu- roscience is mere conceptual neurology I thank W. Scott Wood of Drake University that tends to obscure rather than further for helping me formulate the ideas in this paper scientific progress. It will also address and for comments and reactions to an earlier a number of methodological concerns. draft. A version of this paper was presented at the 27th annual meeting of the Association for Behavior Analysis, May 26, 2001, New Orleans. A BRIEF HISTORY OF Address correspondence to Steven F Faux, COGNITIVE NEUROSCIENCE Department of Psychology, Drake University, Des Moines, Iowa 50311 (e-mail: steven.faux@ Cognitive neuroscience emerged as drake.edu). a new discipline when cognitive sci- 161 162 STEVEN F. FAUX Major Dependent Variables in Journal of Neuroscience 80 Cognitive ~0 en 60 U~~~~~~~~~~~~~~~~~ '- Q) 0 ' °-O = behavioral Irt, error rates, etc.) a 20 0) | = functional (ERP/PET/fMRI) L. 0) 0 0 1988 1991 1993 1995 1997 1999 2001 Year Figure 1. Percentage of empirical articles in each volume of the Journal of Cognitive Neuroscience using either traditional behavioral measures (e.g., reaction time, error rates, etc.) or brain-imaging measures (PET, fMRI, or ERP) in the past 10 years. entists began combining their old cog- at its inception and by its 12th volume nitive-perceptual paradigms with the publishing well over 1,000 pages per new brain-imaging procedures (Servos, year. Citations of cognitive neurosci- 2000). The most important brain-im- ence studies now appear widely in aging technologies that have shaped common textbooks of undergraduate the field are positron emission scan- cognitive psychology (e.g., Solso, ning (PET), functional magnetic reso- 2001; Willingham, 2001) and introduc- nance imaging (fMRI), and event-re- tory psychology (Kosslyn & Rosen- lated potentials (ERPs). These technol- berg, 2001; Myers, 2001). ogies (described in more detail below) In the past, cognitive science has re- give some measure of brain activity lied largely upon reaction time as its with varying degrees of temporal and primary dependent variable. Reaction spatial resolution. times were used as an indirect measure In 1988, the Journal of Cognitive of mental chronometry, that is, the Neuroscience began publication, and speed of mental processes (e.g., Posner, shortly thereafter, the Cognitive Neu- 1986). Although reaction time is still roscience Society (with the acronym important, cognitive neuroscience has CNS) was founded in 1994. CNS be- added three brain-imaging technolo- gan with a membership of about 400 gies (PET, fMRI, and ERP) to its ar- and is now about 2,000 and growing mament. Tabulation of dependent var- (Tara Miller, personal communication, iables used in all empirical articles in 2001). The Journal of Cognitive Neu- the Journal of Cognitive Neuroscience roscience is a peer-reviewed quarterly. between 1991 and 2000 (see Figure 1) Its empirical articles are on standard shows an increasing reliance on brain- topics of cognitive science, which in- imaging techniques relative to tradi- clude working memory, verbal perfor- tional behavioral measures in cognitive mance, visual perception, and mental psychology. Cognitive neuroscience imagery. The journal has grown steadi- has become a brain-imaging science. ly, publishing about 300 pages per year It is easy to understand why some COGNITIVE NEUROSCIENCE 163 scientists, editors, and readers would ity may be several millimeters in any be attracted to this imaging technology. direction. PET brain activity is often The imaging instruments themselves measured while the participant contin- are the result of advances in atomic uously engages in some behavioral physics, electromagnetism, and micro- task (e.g., reading a list of concrete computers. Images from functional nouns presented one at a time or lis- brain scans are often color coded (like tening for a target tone that requires a weather maps) and produce spectacular button press). Multiple stimuli (trials) color plates in journals. The data ap- usually are presented over the time it pear to take the reader a step closer to takes to make a single PET scan. Be- the "black box" of brain operations. cause of the temporal constraints, there Hence, a brief description of the three is little or no ability to resolve PET ac- most widely used imaging techniques tivity as a function of trials or stimulus is useful and is given below. presentations within a task. A typical experiment requires six to ten scans THE IMAGING TECHNOLOGY made about 10 min apart (Cabeza & AND DEPENDENT VARIABLES Nyberg, 1997, p. 3). Scans from an in- dividual participant are averaged, and Positron Emission Tomography then data from multiple participants are This technique is essentially a com- grouped to create grand averages. puterized Geiger counter that can iden- tify regions of radioactivity in the Functional Magnetic Resonance brain. Human participants are admin- Imaging istered (by inhaling or by injection) ra- dioactive isotopes, such as oxygen-15, Functional MRI is similar to PET in nitrogen-13, or carbon-Il, that can be that it provides a measure of blood absorbed by the brain. Brain regions flow-in this case, localized blood ox- presumably activated during a cogni- ygen (see David, Blamire, & Breiter, tive task selectively absorb the isotopes 1994, for introductory review). How- and become differentially radioactive ever, it differs in that no radioactive for a short time (see Cabeza & Nyberg, substances are involved. This tech- 1997, 2000; Roland, 1993). nique utilizes magnetic fields (having a Depending upon the protocol, PET magnetic strength of about 2 to 4 Tes- can measure regional variations in la) and the magnetic properties of he- blood flow, blood volume, oxygen con- moglobin to produce functional brain sumption, and glucose utilization, images. Because deoxyhemoglobin is among others (Raichle, 1996, p. 190). more magnetic than oxygenated he- Oxygen-15-labeled water with a half- moglobin and surrounding tissue, the life of 123 s is the substance most often fMRI machine is able to detect differ- used in PET studies (G. I. Shulman et ences in oxygen levels. This technique al., 1997, p. 642). As neurons become is called blood-oxygen-level-depen- active, the labeled water is locally con- dent (BOLD) fMRI. sumed during oxidative metabolism, Although fMRI images can be made creating differential regions of radio- quickly, it takes 3 to 6 s for oxygen to activity in the brain (Roland, 1993). concentrate in brain regions that be- PET scan measures of gamma radi- come neurologically active (Servos, ation are converted to estimates of re- 2000), creating a functional limit to gional cerebral blood flow. The typical scanning rate. An fMRI machine