How to keep human bias out of AI | Kriti Sharma — Transcript

Full transcript

1. 0:00Translator: Ivana Korom Reviewer: Joanna Pietrulewicz
2. 0:12How many decisions have been made about you today,
3. 0:16or this week or this year,
4. 0:19by artificial intelligence?
5. 0:22I build AI for a living
6. 0:24so, full disclosure, I'm kind of a nerd.
7. 0:27And because I'm kind of a nerd,
8. 0:30wherever some new news story comes out
9. 0:32about artificial intelligence stealing all our jobs,
10. 0:35or robots getting citizenship of an actual country,
11. 0:40I'm the person my friends and followers message
12. 0:43freaking out about the future.
13. 0:45We see this everywhere.
14. 0:47This media panic that our robot overlords are taking over.
15. 0:52We could blame Hollywood for that.
16. 0:56But in reality, that's not the problem we should be focusing on.
17. 1:01There is a more pressing danger, a bigger risk with AI,
18. 1:04that we need to fix first.
19. 1:07So we are back to this question:
20. 1:09How many decisions have been made about you today by AI?
21. 1:15And how many of these
22. 1:17were based on your gender, your race or your background?
23. 1:24Algorithms are being used all the time
24. 1:27to make decisions about who we are and what we want.
25. 1:32Some of the women in this room will know what I'm talking about
26. 1:35if you've been made to sit through those pregnancy test adverts on YouTube
27. 1:39like 1,000 times.
28. 1:41Or you've scrolled past adverts of fertility clinics
29. 1:44on your Facebook feed.
30. 1:47Or in my case, Indian marriage bureaus.
31. 1:50(Laughter)
32. 1:51But AI isn't just being used to make decisions
33. 1:54about what products we want to buy
34. 1:56or which show we want to binge watch next.
35. 2:01I wonder how you'd feel about someone who thought things like this:
36. 2:06"A black or Latino person
37. 2:08is less likely than a white person to pay off their loan on time."
38. 2:13"A person called John makes a better programmer
39. 2:16than a person called Mary."
40. 2:19"A black man is more likely to be a repeat offender than a white man."
41. 2:26You're probably thinking,
42. 2:28"Wow, that sounds like a pretty sexist, racist person," right?
43. 2:33These are some real decisions that AI has made very recently,
44. 2:37based on the biases it has learned from us,
45. 2:40from the humans.
46. 2:43AI is being used to help decide whether or not you get that job interview;
47. 2:48how much you pay for your car insurance;
48. 2:51how good your credit score is;
49. 2:52and even what rating you get in your annual performance review.
50. 2:57But these decisions are all being filtered through
51. 3:00its assumptions about our identity, our race, our gender, our age.
52. 3:08How is that happening?
53. 3:10Now, imagine an AI is helping a hiring manager
54. 3:14find the next tech leader in the company.
55. 3:16So far, the manager has been hiring mostly men.
56. 3:20So the AI learns men are more likely to be programmers than women.
57. 3:25And it's a very short leap from there to:
58. 3:28men make better programmers than women.
59. 3:31We have reinforced our own bias into the AI.
60. 3:35And now, it's screening out female candidates.
61. 3:40Hang on, if a human hiring manager did that,
62. 3:43we'd be outraged, we wouldn't allow it.
63. 3:46This kind of gender discrimination is not OK.
64. 3:49And yet somehow, AI has become above the law,
65. 3:54because a machine made the decision.
66. 3:57That's not it.
67. 3:59We are also reinforcing our bias in how we interact with AI.
68. 4:04How often do you use a voice assistant like Siri, Alexa or even Cortana?
69. 4:10They all have two things in common:
70. 4:13one, they can never get my name right,
71. 4:16and second, they are all female.
72. 4:20They are designed to be our obedient servants,
73. 4:23turning your lights on and off, ordering your shopping.
74. 4:27You get male AIs too, but they tend to be more high-powered,
75. 4:30like IBM Watson, making business decisions,
76. 4:33Salesforce Einstein or ROSS, the robot lawyer.
77. 4:38So poor robots, even they suffer from sexism in the workplace.
78. 4:42(Laughter)
79. 4:44Think about how these two things combine
80. 4:47and affect a kid growing up in today's world around AI.
81. 4:52So they're doing some research for a school project
82. 4:55and they Google images of CEO.
83. 4:58The algorithm shows them results of mostly men.
84. 5:01And now, they Google personal assistant.
85. 5:04As you can guess, it shows them mostly females.
86. 5:07And then they want to put on some music, and maybe order some food,
87. 5:11and now, they are barking orders at an obedient female voice assistant.
88. 5:19Some of our brightest minds are creating this technology today.
89. 5:24Technology that they could have created in any way they wanted.
90. 5:29And yet, they have chosen to create it in the style of 1950s "Mad Man" secretary.
91. 5:34Yay!
92. 5:36But OK, don't worry,
93. 5:38this is not going to end with me telling you
94. 5:40that we are all heading towards sexist, racist machines running the world.
95. 5:44The good news about AI is that it is entirely within our control.
96. 5:51We get to teach the right values, the right ethics to AI.
97. 5:56So there are three things we can do.
98. 5:58One, we can be aware of our own biases
99. 6:01and the bias in machines around us.
100. 6:04Two, we can make sure that diverse teams are building this technology.
101. 6:09And three, we have to give it diverse experiences to learn from.
102. 6:14I can talk about the first two from personal experience.
103. 6:18When you work in technology
104. 6:19and you don't look like a Mark Zuckerberg or Elon Musk,
105. 6:23your life is a little bit difficult, your ability gets questioned.
106. 6:27Here's just one example.
107. 6:29Like most developers, I often join online tech forums
108. 6:33and share my knowledge to help others.
109. 6:36And I've found,
110. 6:37when I log on as myself, with my own photo, my own name,
111. 6:41I tend to get questions or comments like this:
112. 6:46"What makes you think you're qualified to talk about AI?"
113. 6:50"What makes you think you know about machine learning?"
114. 6:53So, as you do, I made a new profile,
115. 6:57and this time, instead of my own picture, I chose a cat with a jet pack on it.
116. 7:02And I chose a name that did not reveal my gender.
117. 7:05You can probably guess where this is going, right?
118. 7:08So, this time, I didn't get any of those patronizing comments about my ability
119. 7:15and I was able to actually get some work done.
120. 7:19And it sucks, guys.
121. 7:21I've been building robots since I was 15,
122. 7:23I have a few degrees in computer science,
123. 7:26and yet, I had to hide my gender
124. 7:28in order for my work to be taken seriously.
125. 7:31So, what's going on here?
126. 7:33Are men just better at technology than women?
127. 7:37Another study found
128. 7:39that when women coders on one platform hid their gender, like myself,
129. 7:44their code was accepted four percent more than men.
130. 7:48So this is not about the talent.
131. 7:51This is about an elitism in AI
132. 7:54that says a programmer needs to look like a certain person.
133. 7:59What we really need to do to make AI better
134. 8:02is bring people from all kinds of backgrounds.
135. 8:06We need people who can write and tell stories
136. 8:09to help us create personalities of AI.
137. 8:12We need people who can solve problems.
138. 8:15We need people who face different challenges
139. 8:18and we need people who can tell us what are the real issues that need fixing
140. 8:24and help us find ways that technology can actually fix it.
141. 8:29Because, when people from diverse backgrounds come together,
142. 8:33when we build things in the right way,
143. 8:35the possibilities are limitless.
144. 8:38And that's what I want to end by talking to you about.
145. 8:42Less racist robots, less machines that are going to take our jobs --
146. 8:46and more about what technology can actually achieve.
147. 8:50So, yes, some of the energy in the world of AI,
148. 8:53in the world of technology
149. 8:55is going to be about what ads you see on your stream.
150. 8:59But a lot of it is going towards making the world so much better.
151. 9:05Think about a pregnant woman in the Democratic Republic of Congo,
152. 9:09who has to walk 17 hours to her nearest rural prenatal clinic
153. 9:13to get a checkup.
154. 9:15What if she could get diagnosis on her phone, instead?
155. 9:19Or think about what AI could do
156. 9:21for those one in three women in South Africa
157. 9:24who face domestic violence.
158. 9:27If it wasn't safe to talk out loud,
159. 9:29they could get an AI service to raise alarm,
160. 9:32get financial and legal advice.
161. 9:35These are all real examples of projects that people, including myself,
162. 9:41are working on right now, using AI.
163. 9:45So, I'm sure in the next couple of days there will be yet another news story
164. 9:49about the existential risk,
165. 9:51robots taking over and coming for your jobs.
166. 9:54(Laughter)
167. 9:55And when something like that happens,
168. 9:57I know I'll get the same messages worrying about the future.
169. 10:01But I feel incredibly positive about this technology.
170. 10:07This is our chance to remake the world into a much more equal place.
171. 10:14But to do that, we need to build it the right way from the get go.
172. 10:19We need people of different genders, races, sexualities and backgrounds.
173. 10:26We need women to be the makers
174. 10:28and not just the machines who do the makers' bidding.
175. 10:33We need to think very carefully what we teach machines,
176. 10:37what data we give them,
177. 10:39so they don't just repeat our own past mistakes.
178. 10:44So I hope I leave you thinking about two things.
179. 10:48First, I hope you leave thinking about bias today.
180. 10:53And that the next time you scroll past an advert
181. 10:56that assumes you are interested in fertility clinics
182. 10:59or online betting websites,
183. 11:02that you think and remember
184. 11:04that the same technology is assuming that a black man will reoffend.
185. 11:09Or that a woman is more likely to be a personal assistant than a CEO.
186. 11:14And I hope that reminds you that we need to do something about it.
187. 11:20And second,
188. 11:22I hope you think about the fact
189. 11:24that you don't need to look a certain way
190. 11:26or have a certain background in engineering or technology
191. 11:30to create AI,
192. 11:31which is going to be a phenomenal force for our future.
193. 11:36You don't need to look like a Mark Zuckerberg,
194. 11:38you can look like me.
195. 11:41And it is up to all of us in this room
196. 11:44to convince the governments and the corporations
197. 11:46to build AI technology for everyone,
198. 11:49including the edge cases.
199. 11:52And for us all to get education
200. 11:54about this phenomenal technology in the future.
201. 11:58Because if we do that,
202. 12:00then we've only just scratched the surface of what we can achieve with AI.
203. 12:05Thank you.
204. 12:06(Applause)