Bias in Technology. Edith Law & Mike Terry

Transcription

1 Bias in Technology Edith Law & Mike Terry

2 Bias in Technology A computer technology is biased if it systematically and unfairly discriminates against certain individuals or groups of individuals in favor of others. A technology discriminates unfairly if it denies an opportunity or a good, or if it assigns an undesirable outcome to an individual or group of individuals on grounds that are unreasonable or inappropriate. (Friedman et al, CHI 1995)

3 Bias Where are there biases in the design of technology? What are these biases? Who can be affected by bias in technology?

4 Biases in Technology: What Design of hardware Form factors Interaction modalities supported Voice input? Keyboard input? Mouse input? Screen reader-friendly? Auditory and visual feedback? Design of software Capabilities offered How functionality is offered to end-user Think of accessibility, localization needs These capabilities intended to address potential bias in designs

5 Biases in Technology: Who Biases can be related to: Physical, mental abilities/disabilities Age Gender Cultural background Education

6 Bias in Technology: Examples Voting systems Games and Gender Bias Educational software and Gender Bias Cultural Biases in Web Technology

7 Voting Systems

8 Voting Technology Voting systems need to be usable by every citizen of legal voting age Elderly Disabled Uneducated Poor users Rare instance of a very general purpose technology Failing to serve every citizen can lead to legal ramifications, inaccuracies in elections

9 Mechanical Voting Machines Bias found with respect to: People of shorter heights Text above eye level for shorter people Women, elderly women disproportionally represented by these characteristics Visual acuity Small type Poor spacing Use of uppercase letters Affects elderly, those with visual impairments Literacy Formal, complex language used on ballots For example, use of double negatives

10 Computer-Based Voting Computer-based touchscreen voting machines referred to as direct recording electronic (DRE) systems Many concerns surrounding reliability, validity of these systems

11 DRE Systems Fear of touchscreen device failures Inaccurate tracking Is a paper trail the answer? What happens when the paper trail printer jams (does happen) DRE systems leading to bias in voter turnout Study found that use of DRE systems linked to lower voter turnout for Hispanic population (Card & Moretti, 2006) Residual votes higher for jurisdictions with electronic voting systems (CalTech/MIT report) Residual votes are those votes that can t be used for one reason or another

12 Bias in DRE Interfaces Different screen layouts can lead to significantly different voting outcomes (work by Selker and others) Feedback seems to be one critical issue Do users know what selection they made? Can they detect errors, whether they make them or system makes them? Comprehensibility, error detection, and error recovery thus even more critical for these types of interfaces The Technology of Access: Allowing People of Age to Vote for Themselves (Selker 2007)

13 Gender Bias

14 Gender Bias Before James Beard hosted a cooking TV show in 1946, home cooking was (exclusively) a woman s job (Cassell) Once he popularized notion that men could cook, domestic devices for kitchen were no longer called appliances, but gadgets Coffee makers, grilling machines

15 But Will it Blend?

16 Gender Bias

17 Gender Bias

18

19 The All Male Cast Problem

20 Gender Stereotyping The problem with Laura is that she was designed by men for men. How do I know this, because if you genetically engineered a Laura shaped woman she would die within around 15 seconds, since there s no way her abdomen could house all her vital organs (Cal Jones, 1997, PC Gaming World) Anita Sarkeesian, a feminist cultural critic, has for months received death and rape threats from opponents of her recent work challenging the stereotypes of women in video games. Bomb threats for her public talks are now routine. One detractor created a game in which players can click their mouse to punch an image of her face. (NY Times, 10/15/2014)

21 Casual, Mobile, Social

22 Gender Bias within Computing Estimated that 1 in 4 jobs require computer literacy implications clear Kindergarteners judge computer to be boy s toy : Boys spend more time with computer Difference in time spent with computer increases with age Amongst 4-6th graders, heavy computer users: boys to girls = 4:1 60% boys enrol in computer classes in high school, 5% girls Justine Cassell s Genderizing HCI paper

23 Gender Bias within Computing Girls and boys tend to see computers differently: Boys see computer as a playful recreational tool, program it, play games on it Girls view computer as a tool, a means to accomplish task Educational participation: 19% of Bachelor's degrees in CS and CE awarded to women UW Fall 2010 Spring 2014: 12.5% 26% of MS 15% of Ph.D.s 14% of assistant professors 13% of associate professors 8% of full CS profs

24 Re-inventing Computer Science

25 Re-inventing Computer Science

26 Bias in Educational Software 1987 study (Huff and Cooper) asked people to design educational software for either boys, girls, or children Software was to teach how to use commas in sentences

27 Bias in Educational Software Software designed for boys designed as games with time pressures, competition key factors Example: Shoot the commas in sentences with a comma cannon Software designed for girls looked like learning tools Simple drill and practice programs Software designed for children looked like designs for boys 80% of the designers were female

28 Bias in Programming Environments Alice is a 3D programming environment developed at CMU Tool used extensively in secondary education to teach programming Students learn programming by programming a 3D environment How might there be gender bias in this tool?

29 Alice Alice s original design geared towards developing games In terms of functions provided, what is easy to do, what is more difficult to do Study found that girls prefer to create narratives Requires different set of high-level functions in the interface Same basic, core 3D engine But need different high-level affordances to effectively create narratives For example, ability to create multiple scenes

30 Storytelling Alice

31 Storytelling Alice Changed functionality to make it easier to create narratives Generic Alice Move Turn Roll Resize Play sound Move to Move toward Move away from Orient to Point at Set point of view to Set pose Move/Turn/Roll at speed New functions Say, think Play sound Walk to, walk offscreen, walk Move Sit on, lie on Kneel Fall down Stand up Straighten Look at, Look Turn to face, turn away from Turn Touch, Keep Touching

32 Storytelling Alice Findings Girls using this version of Alice spent 42% more time programming More than 3x more likely to spend extra time on their programs (Kelleher and Pausch, 2007)

33 Cultural Biases

34 Doodle Doodle is a tool for scheduling meetings. (Reinecke et al., CSCW 2013)

35 Doodle Found that people from collectivist countries respond earlier, agree to fewer options but find more consensus than people from individualist countries. (Reinecke et al., CSCW 2013)

36 Lab in the Wild

37 Visual Preference Study (Reinecke et al., CHI 2014)

38 Visual Preference Study (Reinecke et al., CHI 2014)

39 Visual Preference Study (Reinecke et al., CHI 2014) The study collected 2.4 million ratings, from almost 40k participants, who rated websites and provided information about their demographic background, applied to 430 websites. The study found that age, gender, country and education level all have significant effect on appeal.

40 Visual Preference Study (Reinecke et al., CHI 2014) < 20 years old > 50 years old

41 Visual Preference Study (Reinecke et al., CHI 2014) North America Asia

42 Visual Preference Study (Reinecke et al., CHI 2014) colorfulness complexity United States Germany Macedonia Hong Kong visual appeal colorfulness & complexity

43 Summary Design technology with many personas in mind (age, culture, education, language, gender, disabilities) Be aware that bias happens, with or without your knowledge/explicit action Be aware that bias can manifest itself in many ways From visual design to functionality provided in interface/system design Run studies with range of representative users to uncover biases in design Create diverse teams