WEBVTT 1 00:00:00.090 --> 00:00:02.460 In this lesson, we will learn about 2 00:00:02.460 --> 00:00:05.820 ethical and governance considerations. 3 00:00:05.820 --> 00:00:09.840 Ethical and governance considerations are used to establish 4 00:00:09.840 --> 00:00:13.680 frameworks and guidelines to ensure AI, 5 00:00:13.680 --> 00:00:15.810 or artificial intelligence, 6 00:00:15.810 --> 00:00:18.600 is developed and used responsibly. 7 00:00:18.600 --> 00:00:21.780 And that it aligns with societal values 8 00:00:21.780 --> 00:00:24.300 and organizational principles. 9 00:00:24.300 --> 00:00:26.850 Ethical and governance considerations 10 00:00:26.850 --> 00:00:30.180 include the ethical governance of AI, 11 00:00:30.180 --> 00:00:34.500 and organizational policies on the use of AI. 12 00:00:34.500 --> 00:00:38.130 The ethical governance of AI refers to the structures 13 00:00:38.130 --> 00:00:40.380 and practices put in place 14 00:00:40.380 --> 00:00:43.500 to guide the ethical deployment of AI. 15 00:00:43.500 --> 00:00:47.010 Organizational policies on the use of AI 16 00:00:47.010 --> 00:00:51.090 are specific rules and guidelines created by companies 17 00:00:51.090 --> 00:00:56.090 to govern how AI is implemented within their organization. 18 00:00:56.190 --> 00:01:00.600 Local organizational policies ensure that AI use 19 00:01:00.600 --> 00:01:03.420 aligns with both legal requirements 20 00:01:03.420 --> 00:01:06.750 and an organization's ethical standards. 21 00:01:06.750 --> 00:01:09.210 Let's learn more about the ethical governance 22 00:01:09.210 --> 00:01:13.110 of artificial intelligence, and organizational policies 23 00:01:13.110 --> 00:01:16.080 on the use of artificial intelligence. 24 00:01:16.080 --> 00:01:19.860 First, we have the ethical governance of AI, 25 00:01:19.860 --> 00:01:22.230 or artificial intelligence. 26 00:01:22.230 --> 00:01:25.890 The ethical governance of AI establishes structures 27 00:01:25.890 --> 00:01:30.210 and practices to ensure AI is used responsibly, 28 00:01:30.210 --> 00:01:32.940 aligning with societal values. 29 00:01:32.940 --> 00:01:37.290 This governance prioritizes fairness, transparency, 30 00:01:37.290 --> 00:01:40.710 and accountability in AI applications, 31 00:01:40.710 --> 00:01:44.550 setting the foundation for trust and protection. 32 00:01:44.550 --> 00:01:47.820 A prominent example of ethical governance 33 00:01:47.820 --> 00:01:51.510 is the European Union's proposed AI Act, 34 00:01:51.510 --> 00:01:55.980 which categorizes AI systems into various risk levels 35 00:01:55.980 --> 00:01:59.070 and specifies guidelines for each. 36 00:01:59.070 --> 00:02:03.420 For high risk applications, such as facial recognition, 37 00:02:03.420 --> 00:02:07.020 the act requires comprehensive bias testing 38 00:02:07.020 --> 00:02:11.580 to detect and mitigate potential discriminatory outcomes. 39 00:02:11.580 --> 00:02:15.240 Alongside transparency through clear documentation 40 00:02:15.240 --> 00:02:17.940 of data sources, algorithms, 41 00:02:17.940 --> 00:02:22.350 and decision making processes for regulatory audits. 42 00:02:22.350 --> 00:02:24.540 Developers are also expected 43 00:02:24.540 --> 00:02:27.420 to integrate human oversight features, 44 00:02:27.420 --> 00:02:30.480 allowing intervention in the case of errors. 45 00:02:30.480 --> 00:02:35.480 And informing users of system capabilities and limitations. 46 00:02:35.490 --> 00:02:38.070 These aspects of ethical governance 47 00:02:38.070 --> 00:02:41.640 help guide responsible AI deployments, 48 00:02:41.640 --> 00:02:46.260 minimizing unintended harm, and promoting fairness. 49 00:02:46.260 --> 00:02:49.140 Conversely, a failure in ethical governance 50 00:02:49.140 --> 00:02:51.450 is demonstrated by COMPAS, 51 00:02:51.450 --> 00:02:53.430 the Correctional Offender Management 52 00:02:53.430 --> 00:02:56.340 Profiling for Alternative Sanctions. 53 00:02:56.340 --> 00:02:59.730 This is an algorithm used in the United States 54 00:02:59.730 --> 00:03:03.360 to assess recidivism, which is the likelihood 55 00:03:03.360 --> 00:03:07.800 of an individual re-offending after prison release. 56 00:03:07.800 --> 00:03:10.770 Deployed without requirements for bias testing 57 00:03:10.770 --> 00:03:14.940 or transparency, COMPAS operated with limited scrutiny 58 00:03:14.940 --> 00:03:17.400 over its impact on users. 59 00:03:17.400 --> 00:03:20.100 And while it was designed to support 60 00:03:20.100 --> 00:03:22.860 pretrial and parole decisions, 61 00:03:22.860 --> 00:03:25.830 rather than determine sentences directly, 62 00:03:25.830 --> 00:03:29.790 studies revealed that COMPAS tended to overestimate 63 00:03:29.790 --> 00:03:33.840 the risk of recidivism for minority defendants, 64 00:03:33.840 --> 00:03:36.870 leading to racial disparities in outcomes. 65 00:03:36.870 --> 00:03:40.230 In this case, an ethical governance framework 66 00:03:40.230 --> 00:03:44.100 could have mandated bias audits and transparency measures, 67 00:03:44.100 --> 00:03:47.490 allowing stakeholders to monitor and address 68 00:03:47.490 --> 00:03:50.400 any unintended discriminatory patterns 69 00:03:50.400 --> 00:03:52.560 in the algorithm's outputs. 70 00:03:52.560 --> 00:03:56.310 This underscores the importance of governance structures 71 00:03:56.310 --> 00:04:00.270 in high stakes decisions to prevent unfair treatment 72 00:04:00.270 --> 00:04:02.580 stemming from biased data. 73 00:04:02.580 --> 00:04:06.510 Now, for AI governance frameworks to be effective, 74 00:04:06.510 --> 00:04:10.710 they must prioritize transparency and accountability. 75 00:04:10.710 --> 00:04:14.940 Transparency enables users and stakeholders to understand 76 00:04:14.940 --> 00:04:17.820 the AI decision making processes, 77 00:04:17.820 --> 00:04:20.250 especially when these decisions 78 00:04:20.250 --> 00:04:22.980 could affect individual rights. 79 00:04:22.980 --> 00:04:25.410 By setting transparency standards, 80 00:04:25.410 --> 00:04:28.950 organizations can make available clear explanations 81 00:04:28.950 --> 00:04:32.460 of the AI's decision logic, data sources, 82 00:04:32.460 --> 00:04:34.740 and any potential biases, 83 00:04:34.740 --> 00:04:39.060 building trust and ensuring informed AI use. 84 00:04:39.060 --> 00:04:42.360 Accountability is equally as important, 85 00:04:42.360 --> 00:04:44.880 as it establishes responsibility 86 00:04:44.880 --> 00:04:47.910 for the outcomes of AI systems. 87 00:04:47.910 --> 00:04:51.030 With accountability embedded into governance, 88 00:04:51.030 --> 00:04:56.010 organizations have protocols to address errors or biases, 89 00:04:56.010 --> 00:04:58.620 and correct issues properly. 90 00:04:58.620 --> 00:05:01.770 This ensures that if an AI system produces 91 00:05:01.770 --> 00:05:06.270 an inaccurate or a harmful result, there is a clear process 92 00:05:06.270 --> 00:05:09.540 to investigate and prevent future errors, 93 00:05:09.540 --> 00:05:12.570 enabling organizations to manage risks. 94 00:05:12.570 --> 00:05:14.730 And support the ethical deployment 95 00:05:14.730 --> 00:05:17.910 of AI in critical applications. 96 00:05:17.910 --> 00:05:22.910 Second, we have organizational policies on the use of AI. 97 00:05:23.190 --> 00:05:28.080 Organizational policies on AI use establish internal rules 98 00:05:28.080 --> 00:05:32.970 and guidelines to control how AI is used within a company, 99 00:05:32.970 --> 00:05:35.670 ensuring it aligns with ethical standards 100 00:05:35.670 --> 00:05:37.800 and legal obligations. 101 00:05:37.800 --> 00:05:42.360 Microsoft, for example, has AI ethics principles 102 00:05:42.360 --> 00:05:45.930 that support reviews for high impact systems 103 00:05:45.930 --> 00:05:48.060 involving expert oversight, 104 00:05:48.060 --> 00:05:52.500 to help ensure technology is deployed responsibly. 105 00:05:52.500 --> 00:05:56.820 This policy safeguards users from potential harm, 106 00:05:56.820 --> 00:06:00.030 and standardizes responsible AI use 107 00:06:00.030 --> 00:06:02.250 across the organization. 108 00:06:02.250 --> 00:06:05.430 A notable case underscoring the importance 109 00:06:05.430 --> 00:06:10.430 of strong policies is Amazon's discontinued AI hiring tool, 110 00:06:11.460 --> 00:06:14.640 which was found to exhibit gender bias 111 00:06:14.640 --> 00:06:17.370 due to being trained on historical data 112 00:06:17.370 --> 00:06:19.800 that favored male candidates. 113 00:06:19.800 --> 00:06:24.030 This issue highlights why internal policies should mandate 114 00:06:24.030 --> 00:06:29.030 fairness checks to prevent AI from replicating biases, 115 00:06:29.310 --> 00:06:32.700 especially in sensitive areas like hiring 116 00:06:32.700 --> 00:06:36.720 that impact people's careers and livelihoods. 117 00:06:36.720 --> 00:06:39.750 Okay, we have a good and bad example 118 00:06:39.750 --> 00:06:42.750 of organizational policies on AI now. 119 00:06:42.750 --> 00:06:45.870 So, what components make a good one? 120 00:06:45.870 --> 00:06:49.290 Transparency, training and education, 121 00:06:49.290 --> 00:06:51.330 as well as clear guidelines 122 00:06:51.330 --> 00:06:54.480 and continuous training is the answer. 123 00:06:54.480 --> 00:06:58.800 Transparent policies establish clear ethical boundaries, 124 00:06:58.800 --> 00:07:01.920 helping employees understand how AI 125 00:07:01.920 --> 00:07:04.410 will be applied within their roles. 126 00:07:04.410 --> 00:07:07.080 And defining acceptable practices 127 00:07:07.080 --> 00:07:10.740 for data collection and AI applications, 128 00:07:10.740 --> 00:07:14.340 aligned with ethical and privacy standards. 129 00:07:14.340 --> 00:07:18.840 Next, training and education policies ensure employees 130 00:07:18.840 --> 00:07:22.500 are equipped with knowledge about AI ethics. 131 00:07:22.500 --> 00:07:26.100 A structured ethics training program, for instance, 132 00:07:26.100 --> 00:07:28.950 could cover topics like data privacy, 133 00:07:28.950 --> 00:07:33.630 algorithmic bias, and transparency in decision making. 134 00:07:33.630 --> 00:07:37.680 Finally, clear guidelines and continuous training 135 00:07:37.680 --> 00:07:42.450 reinforce accountability at all levels of AI interaction, 136 00:07:42.450 --> 00:07:45.360 ensuring that employees use AI 137 00:07:45.360 --> 00:07:48.930 in ways that align with organizational values. 138 00:07:48.930 --> 00:07:51.570 Understanding both the technical 139 00:07:51.570 --> 00:07:54.480 and ethical frameworks of AI 140 00:07:54.480 --> 00:07:58.950 helps employees engage with the technology responsibly, 141 00:07:58.950 --> 00:08:02.040 building trust in AI as it becomes 142 00:08:02.040 --> 00:08:06.330 more deeply integrated into business processes. 143 00:08:06.330 --> 00:08:10.950 So remember, ethical and governance frameworks 144 00:08:10.950 --> 00:08:14.010 for artificial intelligence, or AI, 145 00:08:14.010 --> 00:08:17.310 ensure the technology is used responsibly. 146 00:08:17.310 --> 00:08:21.210 And aligns with societal and organizational values. 147 00:08:21.210 --> 00:08:24.780 First, ethical governance provides the structure 148 00:08:24.780 --> 00:08:29.780 for deploying AI fairly, transparently, and accountably. 149 00:08:30.270 --> 00:08:34.620 This builds trust and minimizes unintended harm. 150 00:08:34.620 --> 00:08:39.620 Next, organizational policies define specific rules 151 00:08:39.660 --> 00:08:44.580 for how AI is applied with companies, ensuring its use 152 00:08:44.580 --> 00:08:49.200 aligns with both ethical standards and legal requirements. 153 00:08:49.200 --> 00:08:52.680 Finally, transparency and accountability 154 00:08:52.680 --> 00:08:54.990 are key to these frameworks, 155 00:08:54.990 --> 00:08:58.620 helping stakeholders understand AI decisions 156 00:08:58.620 --> 00:09:01.050 and manage risks responsibly. 157 00:09:01.050 --> 00:09:04.050 Together, these approaches support a safe 158 00:09:04.050 --> 00:09:08.850 and ethical integration of AI, fostering a culture 159 00:09:08.850 --> 00:09:13.383 of responsible technology use within organizations.