1
00:00:00,050 --> 00:00:00,710
Case study.

2
00:00:00,710 --> 00:00:06,290
Mitigating AI bias diverse hires journey through fairness, legal compliance, and ethical AI.

3
00:00:06,320 --> 00:00:13,070
Bias in AI and non-discrimination laws intersect in complex ways, raising crucial questions about fairness,

4
00:00:13,100 --> 00:00:15,740
legality, and ethical responsibility.

5
00:00:16,310 --> 00:00:22,130
Consider the case of diversity Hire, a prominent tech company striving to create an inclusive workforce.

6
00:00:22,280 --> 00:00:28,400
The company invested significantly in an AI driven recruitment platform, AI hire, aimed at streamlining

7
00:00:28,400 --> 00:00:31,880
the hiring process and ensuring a diverse pool of candidates.

8
00:00:32,240 --> 00:00:35,090
The hope was that I would eliminate human biases.

9
00:00:35,090 --> 00:00:42,200
Yet unforeseen issues arose, presenting a practical example of how AI and non-discrimination laws interact.

10
00:00:42,710 --> 00:00:46,490
Diverse hires human resources team noticed a troubling pattern.

11
00:00:46,490 --> 00:00:52,910
Despite the AI systems promise, the demographic diversity of hired candidates did not reflect the applicant

12
00:00:52,910 --> 00:00:54,260
pools diversity.

13
00:00:54,560 --> 00:00:59,930
The system seemed to favor candidates from specific backgrounds, particularly white males.

14
00:00:59,960 --> 00:01:06,160
This observation led to an internal investigation of how the AI system was trained and functioned.

15
00:01:06,910 --> 00:01:12,730
The first question arising from this situation is how could the data set used to train AI hire contribute

16
00:01:12,730 --> 00:01:14,680
to biased hiring outcomes?

17
00:01:15,700 --> 00:01:21,520
The investigation revealed that the historical data fed into AI hire mostly consisted of resumes from

18
00:01:21,520 --> 00:01:26,560
previous successful hires, who predominantly belonged to specific demographic groups.

19
00:01:27,040 --> 00:01:32,890
This data set inherently carried the biases of past hiring practices, perpetuating the same bias.

20
00:01:32,890 --> 00:01:34,870
AI hire was meant to eliminate.

21
00:01:35,680 --> 00:01:41,590
Historical biases in data can be a significant source of discrimination in AI systems, as these models

22
00:01:41,590 --> 00:01:44,470
learn from past trends and may replicate them.

23
00:01:45,730 --> 00:01:50,110
Given these findings, Diversa hire sought to understand their legal obligations.

24
00:01:50,590 --> 00:01:56,620
The Civil Rights Act of 1964, particularly title seven, prohibits employment discrimination based

25
00:01:56,620 --> 00:02:00,640
on race, color, religion, sex, or national origin.

26
00:02:00,660 --> 00:02:05,250
This law applies to I in the same way it applies to human decision makers.

27
00:02:05,520 --> 00:02:11,190
The company's legal team reviewed EEOC guidelines emphasizing the need for transparency and fairness

28
00:02:11,220 --> 00:02:13,170
in AI driven decisions.

29
00:02:13,200 --> 00:02:19,500
They recognize the need to question what steps should Diversey hire take to ensure I hire complies with

30
00:02:19,500 --> 00:02:23,190
title VII and avoids perpetuating discriminatory practices?

31
00:02:24,180 --> 00:02:28,860
Firstly, Diversity Hire implemented a robust governance framework for AI hire.

32
00:02:29,250 --> 00:02:35,370
This included regular audits and assessments to evaluate the algorithm's fairness and identify biases

33
00:02:35,370 --> 00:02:36,420
in training data.

34
00:02:36,450 --> 00:02:42,150
By conducting these audits, the company aimed to uncover and rectify any factors leading to biased

35
00:02:42,180 --> 00:02:42,990
outcomes.

36
00:02:43,110 --> 00:02:48,750
For instance, more diverse and representative data sets were curated to train AI hire anew.

37
00:02:49,020 --> 00:02:55,020
This process required interdisciplinary collaboration involving computer scientists, ethicists, and

38
00:02:55,020 --> 00:02:59,100
social scientists to ensure a multifaceted approach to fairness.

39
00:03:00,150 --> 00:03:03,970
One key aspect of ensuring compliance is transparency.

40
00:03:04,210 --> 00:03:10,180
Diversity hire needed to make AI hires decision making process more transparent to the candidates and

41
00:03:10,180 --> 00:03:11,020
the public.

42
00:03:11,350 --> 00:03:17,050
This involved providing clear explanations of how the AI system evaluates candidates and what criteria

43
00:03:17,050 --> 00:03:17,770
it uses.

44
00:03:17,800 --> 00:03:23,440
Transparency helps build trust and allows for accountability, ensuring that the system operates within

45
00:03:23,440 --> 00:03:25,330
legal and ethical boundaries.

46
00:03:26,650 --> 00:03:33,070
But an essential question that followed was how can diversity hires audit process effectively identify

47
00:03:33,070 --> 00:03:35,710
and mitigate biases in AI hire?

48
00:03:36,430 --> 00:03:41,950
Analyzing the AI systems decisions through various metrics and comparing them to human made decisions

49
00:03:41,950 --> 00:03:43,750
helped identify discrepancies.

50
00:03:43,750 --> 00:03:50,200
For example, the company found that AI higher rated resumes with certain educational backgrounds higher

51
00:03:50,230 --> 00:03:52,810
even when not relevant to the job requirements.

52
00:03:52,840 --> 00:03:58,900
This insight allowed them to adjust the system's weighting mechanisms to focus on skills and experiences

53
00:03:58,900 --> 00:04:00,220
pertinent to the job.

54
00:04:01,470 --> 00:04:06,000
Moreover, Diversity Hire looked at external oversight and regulatory compliance.

55
00:04:06,030 --> 00:04:12,150
They adhered to guidelines similar to those outlined by the GDPR, which requires organisations to provide

56
00:04:12,150 --> 00:04:18,150
individuals with information about how their data is used and implement measures to prevent discriminatory

57
00:04:18,150 --> 00:04:18,960
outcomes.

58
00:04:19,950 --> 00:04:26,070
By adopting these principles, Diversity Hire aim to ensure that AI hires operation was not only fair

59
00:04:26,070 --> 00:04:27,960
but also legally sound.

60
00:04:29,460 --> 00:04:33,000
Collaboration with external experts also proved beneficial.

61
00:04:33,030 --> 00:04:39,840
Diversity invited consultants specialising in ethics, law and social sciences to review AI hire and

62
00:04:39,870 --> 00:04:41,400
offer recommendations.

63
00:04:41,640 --> 00:04:47,100
This interdisciplinary approach ensured that the AI system considered diverse perspectives and did not

64
00:04:47,100 --> 00:04:49,560
inadvertently disadvantage certain groups.

65
00:04:49,740 --> 00:04:56,400
For instance, sociologists provided insights into social structures and biases, while legal experts

66
00:04:56,400 --> 00:04:59,520
ensured compliance with non-discrimination laws.

67
00:05:00,480 --> 00:05:03,920
A thought provoking scenario emerged when diversity hire expanded.

68
00:05:03,950 --> 00:05:08,000
IE hires application to loan approvals for its financial branch.

69
00:05:08,450 --> 00:05:10,910
This highlighted another critical question.

70
00:05:11,300 --> 00:05:16,190
What safeguards should be in place to prevent AI bias in financial decision making?

71
00:05:16,730 --> 00:05:22,910
The company recognized that financial data could also carry historical biases leading to discriminatory

72
00:05:22,910 --> 00:05:24,170
lending practices.

73
00:05:24,800 --> 00:05:30,320
By applying similar audit and transparency measures as in hiring diversity hire, aim to ensure that

74
00:05:30,320 --> 00:05:32,900
loan approvals were fair and unbiased.

75
00:05:33,740 --> 00:05:39,230
Addressing bias in financial AI required understanding the nuances of different demographic groups,

76
00:05:39,230 --> 00:05:41,390
financial behaviors, and histories.

77
00:05:41,900 --> 00:05:47,990
This involved training the AI model on diverse data sets representing various financial backgrounds,

78
00:05:47,990 --> 00:05:53,000
and ensuring that the system did not prioritize certain demographics over others.

79
00:05:53,630 --> 00:05:59,900
Analyzing the outcomes of loan approvals across different groups helped identify and rectify any biases.

80
00:06:00,900 --> 00:06:07,080
To further illustrate the importance of addressing bias in AI, consider the case of facial recognition

81
00:06:07,080 --> 00:06:08,010
technology.

82
00:06:08,520 --> 00:06:14,670
Diverse hire explored using AI for security and attendance, but studies showed that facial recognition

83
00:06:14,670 --> 00:06:18,180
systems could be biased against people of color and women.

84
00:06:18,600 --> 00:06:23,640
These systems were significantly less accurate in identifying individuals from these groups compared

85
00:06:23,640 --> 00:06:28,650
to white males, leading to false identifications and discriminatory practices.

86
00:06:29,430 --> 00:06:35,040
Diversity hire decided to halt the implementation of facial recognition technology until these biases

87
00:06:35,040 --> 00:06:36,660
could be effectively mitigated.

88
00:06:37,320 --> 00:06:37,920
Births.

89
00:06:38,400 --> 00:06:44,040
In response to these findings, some jurisdictions implemented bans or strict regulations on facial

90
00:06:44,040 --> 00:06:49,500
recognition technology used by law enforcement and other entities, illustrating the importance of ongoing

91
00:06:49,530 --> 00:06:52,050
vigilance and regulatory oversight.

92
00:06:52,770 --> 00:06:58,800
Diverse hire adopted a similar stance, emphasizing that ethical considerations must guide technological

93
00:06:58,800 --> 00:07:02,480
deployment, especially in sensitive areas like security.

94
00:07:04,280 --> 00:07:10,790
In conclusion, the case of Diversifier demonstrates the intricate relationship between AI and non-discrimination

95
00:07:10,790 --> 00:07:11,540
laws.

96
00:07:11,630 --> 00:07:18,080
The company's proactive approach in identifying and mitigating biases in AI hire showcases the importance

97
00:07:18,080 --> 00:07:24,830
of regular audits, transparency, interdisciplinary collaboration, and adherence to legal standards.

98
00:07:25,160 --> 00:07:31,460
Addressing these challenges head on allows organizations to harness AI's potential while promoting fairness

99
00:07:31,460 --> 00:07:32,390
and equality.

100
00:07:33,380 --> 00:07:39,320
Diverse hires experience provides valuable lessons for any organization integrating AI into decision

101
00:07:39,350 --> 00:07:40,640
making processes.

102
00:07:41,120 --> 00:07:47,270
By continuously evaluating and refining AI systems, companies can ensure compliance with non-discrimination

103
00:07:47,270 --> 00:07:49,970
laws and uphold ethical principles.

104
00:07:50,150 --> 00:07:55,460
Ultimately, this contributes to creating a more inclusive and just society where technology serves

105
00:07:55,460 --> 00:08:00,050
as a tool for progress rather than perpetuating existing inequalities.
