actively struggling with the problems of talk would be to imagine something that is, at most, the shell or semblance of human life with language.

language is, a place of capture and release, engagement and criticism,

These are entangled, and with the consequence that you can never factor out, from the pure exercise of the activity itself, all the ways in which the activity challenges, retards, impedes and confounds.

and it is we who are tested and we whose limitations are exposed

text

you can’t define a Turing machine or a Turing-computable function.

integrating them into our lives

proxy

In the absence of disturbance, there is nothing:

human animal is a creature of disturbance

is the decisive move in the conjuring trick.

milquetoast

simplistic picture of human and animal cognitive life

If there is intelligence in the vicinity of pencils, shoes, cigarette lighters, maps or calculators, it is the intelligence of their users and inventors

creation of social and technological landscapes that in turn transform what we can do

AI is beyond our control.

without succumbing to bogus claims about machine minds.

learning the realcategories people learn under the real circumstances in which they learn them),the artificial classification learning task does not seem to reflect a singular,independent cognitive process (such as, for example, making an old/new recog-nition judgment)

Categorization is ubiquitous and fundamental to cognition; it is alsomultifaceted and complex

few recent studies have shown failures of theapproach to successfully predict human performance in the traditional artificialclassification learning paradigm

(1)unlike the traditional view that construal and inference (i.e., going beyond theavailable data) occurs after assigning category membership, DIVA uses theprocess of feature prediction and construal as the basis for making a classifica-tion decision; (2) unlike the traditional use of error-driven learning to adjustitem->class weights or feature->class weights, DIVA uses error-driven learningto adjust the recoding and decoding weights that comprise knowledge of within-category inter-feature relationships, so learning is not driven by classificationerrors but by construal errors along the correct category channel; and (3) thedifficulty of a classification problem is driven not so much by between-categoryconfusability (as follows from stimulus generalization theory) but by within-category coherence which can be operationalized in terms of the ease withwhich each feature of a category member can

imagine a contraption with aset of adjustable dials that produces a graded outcome in response to eachinput. The dials are initially at arbitrary positions, so the contraption producescompletely unsystematic outcomes. However, upon each observation of an itemthat merits a strong response (i.e., a category member), the dials are adjusted tomake it more likely that the contraption produces a stronger response to futureobservations similar to that one. Before long, the system reaches a point in “dialspace” that tends to elicit a strong response to the observed examples of a

clearest path to categorymembership is being highly similar to one or more known members of acategory and sufficiently dissimilar to members of contrasting categories

likelihood of category membership is determined as a functionof the inverse exponential distance in psychological space

the learner could store each student’s attributesand link this information with their category

CCM

fixed cue-based approaches because thefeatures of the stimuli are the cues used to predict the category and thisremains unchanged during the learning process

combined cue-based approach in which thefeatures themselves serve not only as individual cues but they are also groupedinto additional compound cues

any stimulus with values nearer to thoseof the average “A” than the average “B” will be classified as a member ofcategory “A.

weights begin at small random initial values and are updated incremen-tally to optimize task performance.

prototype may have feature values or feature combinationsthat are observed rarely or not at all depending on the characteristics of thedensity distribution

category is not a strict definitionagainst which all members conform but instead an ability to extract the statis-tical central tendency across known members

ower complexity rules with fewer memorizedexceptions are preferred

If no such ruleis found, then the model searches for rules that function successfully in conjunc-tion with memorized exceptions

The logical rules that are possible given a set of attributes and operatorsdefine a hypothesis space that gets reduced each time a posited rule is falsifiedby an observation (e.g., a blue item that is not in Category “B”). This approachhas been most fully realized in the RULEX model

prediction (output)

nature of concept formation and whether that can either explain or beexplained by the psycholog

researchers typic-ally employ a two-choice classification task (“Is it an A or a B?”) in a sharplycircumscribed and caricatured domain

timuli people experience are encoded in termsof semantically laden elements (attributes, features, dimensions) and thatsemantic memory holds a conceptual vocabulary of knowledge of the kinds ofthings people can experience or think about in the world (e.g., chairs, dogs,bicycles, baseballs, planets, pickles, pockets, dragons, etc.

updating the representation of the concept toreflect newly designated members

treatment of the general enterprise of advancingscientific understanding of categorization via computational modeling

a taxonomyto help systematize the range of established approaches

categorization is the process of identifying a targetstimulus as belonging to an established category (i.e., concept, kind, or class).