Mark Zuckerberg: We Need To Rely On And Build AI Tools To Help Flag Certain Content | CNBC — Transcript

Full transcript

1. 0:00mr. Zuckerberg I know we both wish we
2. 0:03met under a different set of
3. 0:04circumstances the story broke were
4. 0:07quoted as saying I started Facebook I
5. 0:11run it I'm responsible for what happens
6. 0:14here and quote you said those same words
7. 0:17your opening statement but hour and a
8. 0:19half ago I know you believe that in your
9. 0:22heart it's not just some talking points
10. 0:24some canned speech because by four years
11. 0:28five I'm sorry nine years in the Navy I
12. 0:30know the best commanding officers the
13. 0:32best skippers the best CEOs have that
14. 0:35exact same attitude if Facebook was
15. 0:39they'd be shipped your privacy has taken
16. 0:43a direct hit your trust is severely
17. 0:47damaged you're taking on water and your
18. 0:51future made me a fine with the number
19. 0:52poor the Washington Post with four
20. 0:54commas in it today over a billion
21. 0:58dollars in fines coming your way as you
22. 1:01know you have to reinforce your words
23. 1:03with actions and a few question about
24. 1:06some Alam B's have happen in the past
25. 1:09first of all back in 2012
26. 1:12apparently Facebook did experiment on
27. 1:16six hundred eighty nine thousand three
28. 1:17Facebook users you reduce positive posts
29. 1:22from users friends and lipids so-called
30. 1:25downer posts from other friends they see
31. 1:28fat positive information one group
32. 1:30another group negative information the
33. 1:33goal was to see how the tone of these
34. 1:36posts would affect behavior I will get
35. 1:39in this Forbes article the LA Times
36. 1:41about unlegal illegal human
37. 1:43experimentation up permission I won't
38. 1:45talk about that but it seemed that this
39. 1:47is disconnecting people and stark
40. 1:50contrast your mission to connect people
41. 1:52explain to us how you guys thought this
42. 1:54idea was a good idea explain with people
43. 1:58give them more negative information
44. 1:59positive information well congressman I
45. 2:02view our responsibility as not just
46. 2:04building services that people like to
47. 2:06use but making sure that those services
48. 2:08are also good for people and good for
49. 2:10society overall at the time
50. 2:13were a number of questions about whether
51. 2:15people seeing content that was either
52. 2:17positive or negative on social networks
53. 2:19was affecting their mood and we felt
54. 2:23like we had a responsibility to
55. 2:24understand whether that was the case
56. 2:27because we don't want to have that
57. 2:30effect right we don't want to have it so
58. 2:32that we want use of social media and our
59. 2:34products to be good for people's
60. 2:35well-being I mean we continually make
61. 2:37changes to to that effect including just
62. 2:41recently this year we did a number of
63. 2:44research projects that showed that when
64. 2:48social media is used for building
65. 2:50relationships but so when you're
66. 2:51interacting with people it's associated
67. 2:54with a lot of positive effects of
68. 2:56well-being that you'd expected it makes
69. 2:58you feel more connected less lonely it
70. 3:00correlates with long-term measures of
71. 3:02happiness and health whereas if you're
72. 3:04using social media or the internet just
73. 3:05to passively consume content then that
74. 3:08doesn't have those same positive effects
75. 3:10or can even be negative so we've tried
76. 3:11to shift the product more towards
77. 3:13helping people interact with friends and
78. 3:16family as a result of that so that's the
79. 3:17kind of an example of the kind of work
80. 3:19that we that we do well that was
81. 3:21questioned I believe I've heard you
82. 3:23employ 27,000 people there abouts is
83. 3:26that correct
84. 3:26yes I've also been told that about 20
85. 3:30thousands people including contractors
86. 3:32do work on data security is that correct
87. 3:34yes the 27,000 number is full-time
88. 3:37employees and the security and content
89. 3:40review includes contractors of which
90. 3:42there are tens of thousands ok these
91. 3:45employees are dedicated to security
92. 3:48practices
93. 3:49how can Cambridge and local happen was
94. 3:51so much your work force dedicated that
95. 3:53these perf these causes how'd that
96. 3:55happen well congressman the the issue of
97. 3:59Cambridge analytic and Alexander Cogan
98. 4:01happened before we ramped those programs
99. 4:03up dramatically but one thing that I
100. 4:05think is important to understand overall
101. 4:07is just the sheer volume of content on
102. 4:09Facebook makes its that we can no amount
103. 4:12of people that we can hire will be
104. 4:14enough to review all of the content we
105. 4:17need to rely on and build sophisticated
106. 4:19AI tools that can help us flag certain
107. 4:22content and we're getting good in
108. 4:23certain areas one of the areas that I
109. 4:25mentioned earlier
110. 4:26some terrorist content for example where
111. 4:29we now have AI systems that can identify
112. 4:33and take down 99% of the al-qaeda and
113. 4:36Isis related content in our system
114. 4:37before someone a human even Flags it to
115. 4:40us generally think we need to do more of
116. 4:41that hey there thanks for checking out
117. 4:43CNBC on YouTube be sure to subscribe to
118. 4:46stay up to date on all of the day's
119. 4:48biggest stories you can also click on
120. 4:49any of the videos around me to watch the
121. 4:51latest from CNBC thanks for watching