Association for Behavior Analysis International

The Association for Behavior Analysis International® (ABAI) is a nonprofit membership organization with the mission to contribute to the well-being of society by developing, enhancing, and supporting the growth and vitality of the science of behavior analysis through research, education, and practice.

Search

10th International Conference; Stockholm, Sweden; 2019

Event Details


Previous Page

 

Paper Session #7
Topics in Gender
Sunday, September 29, 2019
8:00 AM–9:20 AM
Stockholm Waterfront Congress Centre, Level 2, C4
Chair: Ally Patterson (George Mason University)
 
Behavior Analysis for Machines: Using Single-Case Experiments to Study Gender Bias in Algorithms
Area: EAB
Domain: Applied Research
ADAM ÅBONDE (Stockholm School of Economics), Lise Bergman Nordgren (Karolinska Institute), Camilla Dahlin-Andersson (Newstag), Richard Wahlund (Stockholm School of Economics)
 
Abstract: As technology becomes ever more ubiquitous, it is important to be able to understand the “smart” machines around us. But, despite growing concern about the potential prevalence of algorithmic bias, algorithms have proven hard to study due to a number of factors (proprietary code, complex structure, unknown training data, etc.). However, one approach to studying algorithms is by framing the problem in terms of behavior: By feeding the algorithm with different stimuli (i.e., instructions), and observing its corresponding response (i.e., output), inferences can be made about functional relations without having access to its underlying structure. Using this approach, two single-case experiments were conducted, where the ad delivery algorithms of Facebook and Google acted as study subjects. Each algorithm was instructed to spread a set of news videos according to an experimentally manipulated schedule, and the resulting behavior was observed. Results indicate that both algorithms had biased tendencies, distributing content to a disproportionally large share of men, which may be potentially harmful for democracy. This study highlights the need for future research to better understand machine behavior, and provides an example of how to use methods from behavior analysis as a valuable tool to study the algorithms involved in everyday life.
 
Arithmetic Decomposition: Early Intervention for a Behavioral Cusp in Mathematics
Area: EDC
Domain: Applied Research
ALLY PATTERSON (George Mason University)
 
Abstract: As early as first grade, girls in the United States are more likely than boys to perform arithmetic using inefficient, overt counting strategies. Decomposition strategies, which are used more frequently by boys, involve chains of behaviors that allow difficult problems to be expressed as multiple simpler problems. Children and adults who solve problems using decomposition strategies perform with higher accuracy during complex problem solving and demonstrate more approach-related behaviors related to mathematics. In the present set of experiments, the researcher designed and implemented novel early-intervention programs to teach component and target skills necessary for covert decomposition in addition (Experiment 1) and covert decomposition in subtraction (Experiment 2). First- and second- grade girls who relied on overt counting strategies at baseline were recruited for participation in the experiments. The interventions relied on behavior analytic techniques such as task analysis, chaining, errorless learning, and differential reinforcement. A functional relationship between the independent and dependent variables was determined through analysis of six features for single-subject designs. As a result of intervention procedures, all participants used a decomposition strategy to accurately and efficiently solve complex addition or subtraction problems. Broad implications of this research are relevant to increasing women’s participation in mathematics.
 
Autoshaping Touchscreen Responses in Two Non-Human Primate Species: The Role of Sex and Stimulus Movement
Area: EAB
Domain: Basic Research
TODD M MYERS (United States Army Medical Research Institute of Chemical Defense), Nathan Rice (USAMRICD), Jennifer Makar (U.S. Army Medical Research Institute of Chemical Defense)
 
Abstract: The stimulus-movement effect is a phenomenon in which stimulus discrimination or acquisition of a response is facilitated by moving stimuli. The effect has been found in monkeys, rats, and humans, but the experiments lacked adequate female representation to investigate potential sex differences. The current experiment analyzed acquisition of stimulus touching in cynomolgus monkeys (Macaca fascicularis) and African green monkeys (Chlorocebus aethiops sabeus) as a function of sex and stimulus movement. The cynomolgus monkeys were given a fixed order of classical conditioning procedures where stimulus correlation and temporal contiguity to food delivery was increased across conditions. Male cynomolgus monkeys acquired the response faster with a moving stimulus, whereas females acquired the response faster with a stationary stimulus. The African green monkeys were given a traditional autoshaping (positive automaintenance) procedure and males more often failed to acquire the stimulus-directed response than did female monkeys. Additionally, the stimulus-movement effects were less pronounced than in the cynomolgus monkeys. These results demonstrate that the stimulus-movement effect may be differentially affected by both sex and species, while also showing that additional experiments with females are needed to determine how sex interacts with behavioral phenomena discovered and elaborated almost exclusively using males.
 
 

BACK TO THE TOP

 

Back to Top
ValidatorError
  
Modifed by Eddie Soh
DONATE
{"isActive":false}