Meaning 'standard' in Greek, 'criterion' was advanced already by Plato, in a very capitalistic connotation: 'Your success, I admit, is fine evidence of the wisdom of the present generation as compared with their predecessors, and it is a popular sentiment that the wise man must above all be wise for himself; of such wisdom the criterion is in the end the ability to make the most mone/ (Hipp. Maj. 283b). Although to the naive mind the mere notion of'criterion' may look straightforward, to philosophers it poses a real problem. Consider the following statements: in order to reach the conclusion that one knows a thing, one must posses criteria for the instantiation of that thing; but in order to know the criteria, one must already know what the thing is. This Catch-22 situation is termed 'the problem of the criterion'. It has haunted philosophers since ancient times, and even brought the Greek sceptics to conclude that we know nothing for real because the aforementioned statements are both correct (Amico 1993). Not surprisingly, 'the problem of the criterion' was dubbed as 'one of the most important and difficult problems in philosophy (Chisholm 1982). Modern epistemology and philosophy of language are occupied with various facets of the meaning and use of criteria (e.g. see discussion in Glock 1996). For our purpose, before taking a much more limited yet pragmatic attitude, suffice it to note that a criterion provides indication that something is the case (definition 2), but unless specifically qualified as such, its satisfaction does not entail the occurrence of what it indicates.
Now to the pragmatics. The fact that 'criterion' stirs debates among philosophers should not discourage neuroscientists from using qualified criteria as powerful tools in experiments1 and "models. Suppose we were to contemplate an elementary and pressing problem in memory research, namely, the relevance of neurobio-logical observations to learning and memory (Dudai 1994b). Attempts to address this issue can benefit from application of a number of criteria, which refer to the following.
1. Correlation. Some biological models of learning and memory are proposed, at least at the time of their inception, on the basis of correlations in space or time of certain biological observations with the behavioural phenomena of learning and memory. In practice, the correlation may be of the first or higher order. It is of the first order if the molecular, cellular, or "system phenomenon is directly correlated with learning and memory. It is of a higher order if the phenomenon is correlated with another intermediate phenomenon, which itself was earlier correlated with learning and memory. For example, correlation of synaptic facilitation in "Aplysia with behavioural "sensitization (Castellucci and Kandel 1976) was of the first order; but correlation of the molecular changes that take place in cultured "Aplysia neurons with the sensitization is of a higher order, as the events in the culture are correlated with the synaptic facilitation rather than directly with the behaviour (Sun and Schacher 1998). Correlation is a very popular criterion, and at the same time a rather weak one. Phenomenological correlation does not necessarily imply mechanistic correlation. A common correlative fallacy is the post hoc argumentation (post hoc ergo propter hoc, Latin for 'after this hence because of this'), which argues that because one event was correlated later in time with another, the second happened because of the first. Correlation nevertheless is a useful guideline for the design offurther experiments intended to establish more stringent criteria (see below).
2. Similarity. Sometimes phenomenological similarity is used to conclude that a process or mechanism are related to learning and memory. For example, some authors promote the claim that "long-term potentiation plays a part in memory because it displays properties expected of memory, such as "persistence following a brief stimulus, "associativity, order-dependency in associativity, and localization in brain regions assumed to subserve memory (see in Dudai 1995). Such similarity is appealing, but could be misleading. Furthermore, there is no reason to assume that parts of a whole should display the properties of the whole and vice versa, hence that the molecular and cellular devices should display the properties of the behaving organism (e.g. Bechtel 1982; "reduction).
3. Usefulness. Are the mechanisms or processes useful for memory? There are two versions to this criterion. The first is 'pragmatical usefulness', i.e. can the process or mechanism be used in an experimental protocol to induce learning-related alterations in a given preparation. For example, serotonin is useful in inducing presy-naptic facilitation in "Aplysia (Sun and Schacher 1998), and "acetylcholine in inducing lasting plasticity changes in "hippocampal neurons (Markram and Segal 1990), supporting the notion that these "neurotransmitters play a role in memory. Experiments of this kind occasionally merge with an attempt to demonstrate that the manipulated agent is sufficient to induce the change (see below). A second version of the usefulness criteria is 'conceptual usefulness', i.e. can the implicated mechanism be productively incorporated into models of learning and memory, leading to a more coherent model and new testable hypotheses. An example is provided by the introduction of the unique kinetic properties of the cyclic adenosine monophosphate-dependent "protein kinase into models that attempt to explain the persistence of use-dependent "plasticity in neurons that encode memory (Buxbaum and Dudai 1989). Usefulness is suggestive, but does not prove that the process or mechanism involved do play a part in learning and memory in vivo.
4. Necessity. Is the mechanism necessary for learning and memory? The "methodology here is to intervene in the physiological process and infer normal function from dysfunction. This is an extremely popular approach. Many types of intervention are possible, involving anatomical lesions (Glassman 1978), mutations ("neurogenetics), or pharmacological agents (e.g. Morris etal. 1986; Berman etal. 1998). Many examples for each of the above are mentioned throughout this book. The types of caveats that typically arise involve post hoc argumentation (see 'correlation' above), doubts about the specificity of the intervention, and the possibility that the effect of the lesion is masked by compensatory mechanisms in vivo.
5. Sufficiency. Does the mechanism suffice for memory formation? In practice, this criterion is more difficult to satisfy than the previous ones. It requires mapping candidate loci of the "engram. The methodology involves mimicry experiments, resembling those mentioned in 3 above. It is common practice to infer that if event A is both necessary and sufficient for event B to take place, B is caused by A.2 The following examples illustrate the use of this criterion: induction of conditioned phototaxis in the mollusc Hermissenda by altering membrane properties of photoreceptors in vivo (Farley et al. 1983); microinfusion of the neurotransmitter octopamine into the brain to show that octopamine encodes the unconditioned stimulus in "classical conditioning of the "honeybee (Hammer and Menzel 1998); a similar experiment with serotonin and long-term facilitation in Aplysia (Sun and Schacher 1998); switching on "gluatamtergic N-methyl-d-aspartate "receptors in the "mouse in an attempt to prove that this receptor is crucial in "consolidation (Shimizu et al. 2000); induction of long-term memory in Aplysia, "Drosophila, or the honeybee by activation of the cyclic adenosine monophosphate "intracellular signal transduction cascade, "CREB, and modulation of gene expression (Yin and Tully 1996; Muller 2000); reversible inactivation of the "cerebellum and brainstem pathways to show that they are necessary and sufficient for conditioning the nictitating membrane reflex in the rabbit (Thompson et al. 2000); and microstimulation of visual cortex in order to alter the behavioural response of the behaving "monkey (Groh et al. 1997). Considering the number of variables involved in any brain function, it is rather unlikely that a single molecular or cellular event will suffice to induce memory faithfully. Nevertheless, activation or inhibition of key 'molecular switches', such as CREB, could have substantial behavioural effects.
6. Exclusiveness. The most demanding criterion is exclusiveness. Such a claim cannot be currently made for any candidate mechanism in learning. Some discussions of CREB do dare to hint that the manipulated cellular process plays an exclusive part in memory consolidation (e.g. Yin and Tully 1996). Again, as in the case of the criterion of sufficiency but even more so, the question is whether exclusiveness can at all be expected, as multiplicity of mechanisms and parallel pathways appear to be the rule in the brain systems that subserve learning.
The above or other criteria (e.g. Rose 1981) are applicable and useful in multiple domains of memory research. Here is another central question that calls for the formulation of criteria: How should one delineate and classify memory systems? (Shettleworth 1993; "taxonomy.) Occasionally the discussion of such issues leads us back to the good old philosophical 'problem of the criterion', even if this problem is not explicitly stated (see Sherry and Schacter 1987).
Selected associations: Method, Reduction, System
'An example of the use in learning experiments of a generic type of criterion that fits definition 1, is 'learning to criterion' in 'acquisition.
2This inference is not a given. Some philosophers will raise the possibility that B is merely 'supervenient' upon A. Briefly speaking, a 'supervenient', or 'consequential' attribute comes along in addition to other attributes but is not necessarily entailed by them (Kim 1978; Davidson 1980; Hare 1984; *reduction).This is not an argument commonly employed by brain scientists. It does come up in discussions that concern the mind-body problem.
Was this article helpful?