AI Ethics and Consideration for Responsible Adoption — Transcript

Full transcript

1. 0:00We've talked about what AI is, how it's
2. 0:02[music] adopted, and what it takes to
3. 0:04lead through it. Now comes the harder
4. 0:06question.
5. 0:07>> [music]
6. 0:07>> What does this technology demand of us
7. 0:09ethically? Because AI doesn't just scale
8. 0:12benefits, it scales harm just as fast
9. 0:15[music] if you're not paying attention.
10. 0:16Start with job displacement, because
11. 0:18it's the fear people raise first.
12. 0:20>> [music]
13. 0:21>> In 2023, Hollywood's writers and actors
14. 0:24went on strike, largely over whether AI
15. 0:26would replace the people who create
16. 0:28stories. That reaction isn't new. Every
17. 0:31major technology shift produces a period
18. 0:33of genuine anxiety before society works
19. 0:36out new rules. This is what change looks
20. 0:39like from the inside, [music]
21. 0:40not a sign something's gone wrong.
22. 0:42People's individual reactions vary
23. 0:44predictably, too, the same way we
24. 0:46discussed with the heart-and-mind matrix
25. 0:48earlier. Some feel inspired, others
26. 0:50depressed, [music]
27. 0:52some just want to be equipped to keep
28. 0:53up. None of those reactions are
29. 0:55irrational. They're worth taking
30. 0:57seriously, not dismissing. [music]
31. 0:59Here's the uncomfortable truth about
32. 1:01bias.
33. 1:02AI systems in the US have historically
34. 1:04performed worse on images [music] of
35. 1:06people with darker skin, because
36. 1:08training data was built overwhelmingly
37. 1:11on images of white faces. Gender bias
38. 1:13shows up the same way, baked into the
39. 1:16data before a single user ever sees the
40. 1:18product. [music]
41. 1:19When you deploy a biased system to a
42. 1:21billion users, the bias scales with it.
43. 1:24My message to every leader,
44. 1:25>> [music]
45. 1:26>> this is not a problem for your lawyers.
46. 1:28It's a problem for you in how you design
47. 1:30the thing in the first [music] place.
48. 1:32Fairness isn't a compliance checkbox you
49. 1:35bolt on afterward, it has to be a first
50. 1:37principle commitment from the start. Two
51. 1:39more considerations, and you've actually
52. 1:42already seen both [music] in action.
53. 1:44Remember the reinforcing loop from
54. 1:45earlier? More users, more data, [music]
55. 1:48better product. That same data is
56. 1:51personal, preferences, location,
57. 1:54behavior, sometimes health information.
58. 1:57Every organization that benefits from
59. 1:59that loop also takes on a responsibility
60. 2:02to protect what it's collecting. [music]
61. 2:03And remember the Turkey-Syria earthquake
62. 2:06where survivors real footage saved
63. 2:08lives?
64. 2:09>> [music]
65. 2:09>> The same tools now let anyone fabricate
66. 2:11a convincing video of an event that
67. 2:13never happened.
68. 2:15That's not just a media problem, it's a
69. 2:17trust problem.
70. 2:18When fabrication becomes this easy, the
71. 2:21question stops being who has the tools
72. 2:23to report and becomes who decides what's
73. 2:27true and how. You don't have to work
74. 2:29this out from scratch. In 2021, UNESCO's
75. 2:33193 member states adopted [music] the
76. 2:35first global standard on AI ethics.
77. 2:38Grounded in principles including
78. 2:40fairness and non-discrimination,
79. 2:42the right to privacy and data
80. 2:44protection, transparency and
81. 2:46explainability, and human oversight,
82. 2:48meaning AI should never fully displace
83. 2:51human responsibility and accountability.
84. 2:54I want you to treat this as more than an
85. 2:56abstract ethics discussion. As you
86. 2:58integrate AI into your project,
87. 3:00proposal, and external work, don't just
88. 3:03ask whether a technology can help solve
89. 3:05your problem. Ask who it affects, what
90. 3:08risks it introduces, and what will
91. 3:10actually determine whether the people
92. 3:12you're serving trust it enough to use
93. 3:13it. None of this is a reason to hold
94. 3:15back. It's a reason to be deliberate.
95. 3:18Consider Duolingo Max, powered by
96. 3:20GPT-4O.
97. 3:22Users who engage with it complete three
98. 3:24times more conversations and report 78%
99. 3:27better real-world readiness to use the
100. 3:29language. Daily active users grew past
101. 3:3240 [music] million, up 51% after the AI
102. 3:35features launched. Roughly 2 billion
103. 3:38people worldwide have never been to
104. 3:40school.
105. 3:40>> [music]
106. 3:40>> Give someone a device and internet
107. 3:42access and you can give them a genuinely
108. 3:44capable instructor. The tools
109. 3:46[clears throat] to close real gaps
110. 3:48already exist.
111. 3:49>> [music]
112. 3:49>> Whether they get deployed responsibly at
113. 3:51the scale the problem demands, is a
114. 3:54choice. And [music] it's the leader's
115. 3:55choice to make. So, four things to carry
116. 3:58forward. Anticipate. Don't wait for
117. 4:01bias, privacy, or trust problems to
118. 4:04surface before you think about them.
119. 4:06Design responsibly from the start, not
120. 4:08as an afterthought. Stay engaged with
121. 4:11the people your plan affects, [music]
122. 4:12especially those most likely to be
123. 4:14harmed if you get it wrong. And hold the
124. 4:17ethical line even when it's
125. 4:18inconvenient.
126. 4:19>> [music]
127. 4:19>> Because the leaders who succeed in this
128. 4:21era will be the ones who create value
129. 4:23broadly, not just quickly. As you
130. 4:26finalize your own project plan, revisit
131. 4:28those three questions. Who's affected?
132. 4:31What's the risk? And what earns trust?