Towards Transparency by Design for Artificial Intelligence

Science and Engineering Ethics - In this article, we develop the concept of Transparency by Design that serves as practical guidance in helping promote the beneficial functions of transparency...     

Mechanisms of source confusion and discounting in short-term priming 2: effects of prime similarity and target duration

D. E. Huber, R. M. Shiffrin, K. B. Lyle, and K. I. Ruys (2001) tested two-alternative, forced-choice (2-AFC) perceptual identification in a short-term priming task. For repetition priming, passive viewing of primes resulted in a preference to choose repeated words, but actively responding to primes resulted in a preference against choosing repeated words. These results were explained with a computational model, responding optimally with unknown sources of evidence (ROUSE), using the offsetting mechanisms of source confusion and discounting. An analysis of ROUSE revealed conditions under which discounting efficacy should diminish, causing a preference for primed words even with active prime processing. Two new studies confirm 2 such conditions: very short target flash durations and very low similarity between primes and primed choice words. These a priori predictions contrast with the a posteriori data fits of a multinomial model developed by R. Ratcliff and G. McKoon (2001).     

Syllable structure and sonority in language inventory and aphasic neologisms

Phonological theories have raised the notion of a universally preferred syllable type which is defined in terms of its sonority structure (e.g., ). Empirical evidence for this notion has been provided by distributional analyses of natural languages and of language acquisition data, and by aphasic speech error analyses. The present study investigates frequency distributions of syllable types in German, which allows for a rather complex syllable structure, and in neologistic utterances of a German speaking jargon aphasic. The findings suggest that the sonority structure of the patient's neologisms is generally in accordance with the notion of theoretically preferred syllables. Moreover, comparative analyses suggest that the predominance of the preferred syllable type is especially pronounced in the aphasic data. On the basis of these findings, the influence of sonority in impaired phonological lexical processing is discussed.     

An fMRI study on conceptual, grammatical, and morpho-phonological processing

The aim of the present study was to determine whether processing of syntactic word information (lemma) is subserved by the same neural substrate as processing of conceptual or word form information (lexeme). We measured BOLD responses in 14 native speakers of German in three different decision tasks, each focussing specifically on one level of lexical processing (conceptual, syntactic, and morpho-phonological). The test parameters were natural gender, grammatical gender, and word form derivation, respectively. Discrimination between words played backwards and complex sounds served as control task. Complex contrasts revealed a functional fractionation of the left inferior frontal gyrus for each level of lexical processing.     

Object permanence in common marmosets (Callithrix jacchus)

A series of 9 search tasks corresponding to the Piagetian Stages 3-6 of object permanence were administered to 11 common marmosets (Callithrix jacchus). Success rates varied strongly among tasks and marmosets, but the performances of most subjects were above chance level on the majority of tasks of visible and invisible displacements. Although up to 24 trials were administered in the tests, subjects did not improve their performance across trials. Errors were due to preferences for specific locations or boxes, simple search strategies, and attentional deficits. The performances of at least 2 subjects that achieved very high scores up to the successive invisible displacement task suggest that this species is able to represent the existence and the movements of unperceived objects.     

Pigeons use item-specific and category-level information in the identification and categorization of human faces

Three experiments examined the pigeon's ability to adapt to the specific tasks of identification and categorization of complex visual stimuli by attending to information at different levels of abstraction. In Experiment 1, pigeons were successively trained to discriminate first between pairs of human faces from the same sex and then between male and female faces. A final test revealed that the associations formed in the first training phase did not survive categorization training, indicating that the pigeons used item-specific information to discriminate faces of the same sex, but shifted their attention to category-level information for discriminating between the male and female categories. Two further experiments disproved alternative explanations. The results support the feature-learning account of open-ended categorization in animals.     

Information search in the laboratory and on the Web: With or without an experimenter

The focus of this study is the effect of the location (laboratory vs. Web) of experiments on active information searchin decision-making tasks. In two experiments, participants were confronted with two different search method versions (list vs. keyword) for acquiring information about a task from a database. The amount and type of information gathered and the time required for task completion were measured. In Experiment 1, significantly more information was searched for in the laboratory than on the Web when the list version was employed, whereas there was no difference between locations in the keyword version. In Experiment 2, the participants were assigned randomly to the Web or the laboratory condition. The results of Experiment 1 were replicated. Whereas location (and the presence or absence of an experimenter) had an effect on the absolute amount of information gathered in both experiments, the relative distribution and type of information items did not differ.     

The road not taken: creative solutions require avoidance of high-frequency responses

To investigate individual differences in creativity as measured with a complex problem-solving task, we developed a computational model of the remote associates test (RAT). For 50 years, the RAT has been used to measure creativity. Each RAT question presents three cue words that are linked by a fourth word, which is the correct answer. We hypothesized that individuals perform poorly on the RAT when they are biased to consider high-frequency candidate answers. To assess this hypothesis, we tested individuals with 48 RAT questions and required speeded responding to encourage guessing. Results supported our hypothesis. We generated a norm-based model of the RAT using a high-dimensional semantic space, and this model accurately identified correct answers. A frequency-biased model that included different levels of bias for high-frequency candidate answers explained variance for both correct and incorrect responses. Providing new insight into the nature of creativity, the model explains why some RAT questions are more difficult than others, and why some people perform better than others on the RAT.     

Cognitive function, physical activity, and aging: possible biological links and implications for multimodal interventions

Increasing evidence shows the positive effect of physical activity (PA) on maintaining cognitive function. Both processes seem intrinsically linked to each other. The most likely mechanism is a reciprocal stimulation of neuroplasticity. Based on extensive experimental work on animals and humans, the concept of an enriched environment, including PA and challenging cognitive tasks, has provided the basis to systematically assess possible interventions for successful aging. I will use recent findings on brain mechanisms associated with PA and its effects on higher cognitive function at a systems and molecular level to demonstrate the need to design effective interventions. Such interventions should be designed to take advantage of the presumed compensatory mechanisms of elderly individuals, thereby limiting functional decline in higher cognitive performance in aging people. My review of the most recent relevant publications concerning this topic supports the notion that it is a promising approach to provide cognitive training and PA in conjunction, since the combination may generate synergistic beneficial changes than either one individually. Multimodal training programs should therefore be tested. However, at present there is insufficient evidence to conclude that multimodal interventions are superior to isolated cognitive or physical exercise interventions, since major studies addressing this topic are lacking. These studies are needed to conclusively prove that both strategies will positively interact when used in combined interventions. Finally the use of modern technology for these interventions will be discussed.