WEBVTT - Avoiding Deep Fakes, Misinformation within AI

0:00:02.520 --> 0:00:09.160
<v Speaker 1>Bloomberg Audio Studios, podcasts, radio news. You're listening to Bloomberg

0:00:09.240 --> 0:00:13.760
<v Speaker 1>BusinessWeek with Carol Masser and Tim Steneveek on Bloomberg Radio.

0:00:14.080 --> 0:00:17.160
<v Speaker 2>As more and more companies adopt AI and more consumers

0:00:17.280 --> 0:00:19.599
<v Speaker 2>use it, the more we learn about the potential flaws

0:00:19.920 --> 0:00:23.040
<v Speaker 2>of the technology. For example, Carol, maybe you remember last

0:00:23.079 --> 0:00:27.960
<v Speaker 2>year when Google's AI suggested that everybody should just beat

0:00:28.040 --> 0:00:29.200
<v Speaker 2>a rock every day.

0:00:29.320 --> 0:00:30.240
<v Speaker 3>Not a good idea.

0:00:30.320 --> 0:00:33.479
<v Speaker 2>Well, they said, rocks have minerals and vitamins that are

0:00:33.560 --> 0:00:35.280
<v Speaker 2>quote important for digestive health.

0:00:35.400 --> 0:00:35.720
<v Speaker 1>Got it.

0:00:35.840 --> 0:00:38.400
<v Speaker 3>I understand that logic, but still not a good idea.

0:00:38.520 --> 0:00:40.000
<v Speaker 2>Or just a few weeks ago, in the Fable app,

0:00:40.040 --> 0:00:43.519
<v Speaker 2>which calls itself the social app for bookworms and binge watchers,

0:00:43.520 --> 0:00:46.240
<v Speaker 2>found itself having to apologize after some of its AI

0:00:46.320 --> 0:00:49.400
<v Speaker 2>generated reader summaries contained what the company ended up calling

0:00:49.479 --> 0:00:53.360
<v Speaker 2>quote very bigoted, racist language. Fable then remove these features

0:00:53.880 --> 0:00:57.160
<v Speaker 2>from the app. It's an evolving world, Yeah absolutely. Padma

0:00:57.200 --> 0:00:59.680
<v Speaker 2>Warrior is founder, president and a CEO of Fable. She's

0:00:59.680 --> 0:01:02.880
<v Speaker 2>also the former CTO of Motorola and Cisco, also the

0:01:02.880 --> 0:01:05.760
<v Speaker 2>former CEO of the US arm of the Chinese ev

0:01:05.880 --> 0:01:08.559
<v Speaker 2>maker Neo. She was on the board of Microsoft until

0:01:08.560 --> 0:01:11.280
<v Speaker 2>twenty twenty three. She's on the board of Spotify as well.

0:01:11.319 --> 0:01:14.720
<v Speaker 2>She joins us this afternoon from Palo Alto Padma. Good

0:01:14.720 --> 0:01:15.720
<v Speaker 2>to have you with us. How are you?

0:01:17.000 --> 0:01:18.919
<v Speaker 4>Thank you for having me. I'm doing well.

0:01:19.360 --> 0:01:22.120
<v Speaker 2>So I'm curious about Fable. A lot of people are

0:01:22.160 --> 0:01:26.920
<v Speaker 2>familiar with it because people use it for rating books,

0:01:27.680 --> 0:01:31.840
<v Speaker 2>finding books, same with stuff we watch on the bigger screen.

0:01:33.080 --> 0:01:35.319
<v Speaker 2>You're the lesson though. The takeaway from AI you guys

0:01:35.319 --> 0:01:38.440
<v Speaker 2>were so quick to come out and apologize for it.

0:01:38.880 --> 0:01:41.880
<v Speaker 2>Take action, be public about it. What were the What

0:01:42.240 --> 0:01:43.440
<v Speaker 2>did you learn from that incident?

0:01:44.880 --> 0:01:48.200
<v Speaker 4>Yeah? I know. Fable is a community platform.

0:01:48.480 --> 0:01:52.480
<v Speaker 5>We focus on helping people find, by track, and share

0:01:52.600 --> 0:01:57.640
<v Speaker 5>all their entertainment interests in one personalized space. We have

0:01:57.880 --> 0:02:01.760
<v Speaker 5>users from all over the world onward and ninety five countries,

0:02:02.240 --> 0:02:05.800
<v Speaker 5>and our mission is to celebrate human creative tea in

0:02:05.840 --> 0:02:10.000
<v Speaker 5>the form of stories. So the incidents you are referring to,

0:02:10.280 --> 0:02:14.240
<v Speaker 5>we were using generative AI to create a summary for

0:02:14.320 --> 0:02:19.520
<v Speaker 5>readers based on their reading lists, and regrettably it served

0:02:19.520 --> 0:02:24.240
<v Speaker 5>a few readers with hurtful, unacceptable commentary related to race.

0:02:25.040 --> 0:02:27.200
<v Speaker 4>We apologized, as you mentioned.

0:02:27.000 --> 0:02:30.800
<v Speaker 5>Very quickly removed these gen ai features from our platform.

0:02:31.520 --> 0:02:34.400
<v Speaker 5>To me, they were very A couple of key lessons

0:02:34.440 --> 0:02:39.320
<v Speaker 5>that we learned. One is it's really important to have

0:02:39.360 --> 0:02:42.800
<v Speaker 5>a human in the loop. In these early days of

0:02:42.960 --> 0:02:48.359
<v Speaker 5>Generator Ai. It is essential to carry out really extensive

0:02:48.480 --> 0:02:50.760
<v Speaker 5>beta testing, and we did a lot of testing, by

0:02:50.760 --> 0:02:55.320
<v Speaker 5>the way, for our feature, but even with all the testing,

0:02:55.440 --> 0:02:57.919
<v Speaker 5>you're bound to have an issue. So I think it's

0:02:58.000 --> 0:03:01.079
<v Speaker 5>really really important to have human in the loop approach.

0:03:01.560 --> 0:03:07.280
<v Speaker 5>The second is transparency and listening to community feedback when

0:03:07.440 --> 0:03:11.120
<v Speaker 5>someone I think our mission, our philosophy at our company

0:03:11.200 --> 0:03:14.400
<v Speaker 5>is even when we do something with our product, which

0:03:14.440 --> 0:03:18.040
<v Speaker 5>generator wai, even if it hurts one person, we take

0:03:18.080 --> 0:03:21.400
<v Speaker 5>it very seriously. So I think that level of transparency

0:03:21.440 --> 0:03:24.840
<v Speaker 5>and being close to listening to the community and users

0:03:25.240 --> 0:03:28.520
<v Speaker 5>and having the ability to react quickly, those are two

0:03:28.560 --> 0:03:31.840
<v Speaker 5>big lessons that we learned and I hope other entrepreneurs

0:03:31.880 --> 0:03:32.360
<v Speaker 5>follow this.

0:03:33.600 --> 0:03:35.360
<v Speaker 3>How much is it that just we're kind of early

0:03:35.400 --> 0:03:37.480
<v Speaker 3>in on this process. I'm not excusing any of it

0:03:37.520 --> 0:03:41.880
<v Speaker 3>that goes, you know, it goes like it did for

0:03:41.920 --> 0:03:44.040
<v Speaker 3>you guys. But is it a case of it's only

0:03:44.040 --> 0:03:46.920
<v Speaker 3>as good as its input, and that is it a

0:03:46.960 --> 0:03:49.680
<v Speaker 3>fair assumption. How do you see you understand the tech world,

0:03:49.680 --> 0:03:51.800
<v Speaker 3>and I'm sure you're looking closely, you know at AI

0:03:51.920 --> 0:03:56.280
<v Speaker 3>right now that the more we put in better, cleaner information,

0:03:56.480 --> 0:03:57.760
<v Speaker 3>the better it gets.

0:03:58.600 --> 0:03:58.840
<v Speaker 4>Yeah.

0:03:58.840 --> 0:04:01.480
<v Speaker 5>Absolutely, And I think there is a lot of confusion.

0:04:01.680 --> 0:04:05.320
<v Speaker 5>People sometimes confuse AI the generative AI. I think it's

0:04:05.360 --> 0:04:09.320
<v Speaker 5>important to perhaps distinguish a little bit, at least broadly,

0:04:09.680 --> 0:04:13.560
<v Speaker 5>the difference between the two. You know, I think as

0:04:13.600 --> 0:04:16.120
<v Speaker 5>you guys probably know, and I'm not sure how much

0:04:16.320 --> 0:04:17.720
<v Speaker 5>how much people realize this.

0:04:17.839 --> 0:04:18.159
<v Speaker 4>Though.

0:04:18.600 --> 0:04:22.400
<v Speaker 5>AI processes data to analyze it and provide us with

0:04:22.560 --> 0:04:27.920
<v Speaker 5>predictions and insight, for example, performing tasks like identifying objects

0:04:27.920 --> 0:04:32.640
<v Speaker 5>in images, converting speech to tests, recommendations, all of this thing,

0:04:33.200 --> 0:04:37.400
<v Speaker 5>generative AI is different. It creates new content based on

0:04:37.920 --> 0:04:41.960
<v Speaker 5>the data that it has learned from. So generative AI

0:04:42.080 --> 0:04:46.240
<v Speaker 5>can create text, can create images, can create music videos.

0:04:45.839 --> 0:04:46.320
<v Speaker 4>All of this.

0:04:46.920 --> 0:04:50.880
<v Speaker 5>I find that generative AI it's developing very quickly and

0:04:51.040 --> 0:04:55.800
<v Speaker 5>very fast, and it's really really important because as I said,

0:04:56.160 --> 0:05:01.440
<v Speaker 5>even the best AI needs human oversight at the stage.

0:05:01.960 --> 0:05:04.679
<v Speaker 5>As you pointed out, it's only as good as the input.

0:05:04.760 --> 0:05:08.720
<v Speaker 5>So it's really important to train generative AI algorithms and

0:05:08.839 --> 0:05:12.679
<v Speaker 5>actually even AI on data sets that are culturally literate

0:05:12.880 --> 0:05:16.840
<v Speaker 5>and sensitive, because even as humans, we are evolving, right,

0:05:17.000 --> 0:05:20.080
<v Speaker 5>Like language that was used years ago is not acceptable

0:05:20.120 --> 0:05:25.000
<v Speaker 5>to us anymore, and it's really important to train our

0:05:25.120 --> 0:05:30.760
<v Speaker 5>algorithms and artificial intelligence so that it doesn't inadvertently promote

0:05:30.800 --> 0:05:32.800
<v Speaker 5>harmful stereotypes and biases.

0:05:34.440 --> 0:05:36.560
<v Speaker 3>So, okay and Tim and I are just kind of

0:05:36.600 --> 0:05:38.760
<v Speaker 3>like listening to So. I love first of all, your

0:05:38.880 --> 0:05:41.880
<v Speaker 3>distinction between generative AI and AI, and you're right to

0:05:41.880 --> 0:05:43.880
<v Speaker 3>make the distinction. AI has been around for a long

0:05:43.920 --> 0:05:48.599
<v Speaker 3>time in terms of what happened on your app, and

0:05:48.680 --> 0:05:54.200
<v Speaker 3>the information that certainly was not welcomed was that gen

0:05:54.279 --> 0:05:56.840
<v Speaker 3>ai or regular ai, that.

0:05:56.920 --> 0:05:57.600
<v Speaker 4>Was jen Ai.

0:05:58.240 --> 0:06:01.120
<v Speaker 5>We were saying gen Ai in three of our features

0:06:01.120 --> 0:06:04.000
<v Speaker 5>on the product, and we had issue with one of them,

0:06:04.080 --> 0:06:07.240
<v Speaker 5>but we actually pulled all three features back. We launched

0:06:07.279 --> 0:06:11.320
<v Speaker 5>two of them back without Genai and using human creativity,

0:06:11.880 --> 0:06:15.200
<v Speaker 5>so that our community actually is actually more happy now

0:06:15.240 --> 0:06:16.240
<v Speaker 5>with those features.

0:06:16.520 --> 0:06:20.039
<v Speaker 2>Do you think you can relaunch the genai feature eventually?

0:06:21.640 --> 0:06:22.800
<v Speaker 4>I think it'll take us.

0:06:23.160 --> 0:06:26.280
<v Speaker 5>I think for us we need to be really cautious because,

0:06:26.320 --> 0:06:29.160
<v Speaker 5>as I said, if you look at the statistics, right,

0:06:29.240 --> 0:06:33.680
<v Speaker 5>it was like ten possible issues out of thirty million

0:06:33.839 --> 0:06:36.240
<v Speaker 5>that we did. But even that small ratio, I think

0:06:36.279 --> 0:06:39.440
<v Speaker 5>the interesting thing for me, at least with genai, the

0:06:39.520 --> 0:06:42.520
<v Speaker 5>testing can't just be quantified and say we had a

0:06:42.600 --> 0:06:46.200
<v Speaker 5>very small percentage of failure, so we should release this feature,

0:06:46.839 --> 0:06:50.120
<v Speaker 5>because as I said, even one harmful comment and one

0:06:50.200 --> 0:06:54.080
<v Speaker 5>hurtful comment is bad, and so how do you find

0:06:54.120 --> 0:06:57.080
<v Speaker 5>that in millions and millions that you're creating. So I

0:06:57.120 --> 0:07:00.000
<v Speaker 5>think at this stage we will probably not use genai

0:07:00.160 --> 0:07:03.440
<v Speaker 5>for those sorts of features, and I think we will

0:07:03.440 --> 0:07:06.200
<v Speaker 5>continue to work on it to train our models and

0:07:06.240 --> 0:07:08.479
<v Speaker 5>algorithms in the right way.

0:07:09.160 --> 0:07:13.680
<v Speaker 3>Do you think that the government should be more involved

0:07:13.800 --> 0:07:18.200
<v Speaker 3>or should be very closely involved with oversight of AI.

0:07:20.160 --> 0:07:20.640
<v Speaker 4>I'm not sure.

0:07:21.160 --> 0:07:24.040
<v Speaker 5>Yeah, I think it's difficult, I think, but so much

0:07:24.160 --> 0:07:27.920
<v Speaker 5>broad usage, I don't know if government should be involved,

0:07:27.920 --> 0:07:30.640
<v Speaker 5>but there should certainly be guardrailed. I think it's up

0:07:30.720 --> 0:07:34.840
<v Speaker 5>to technology development teams and companies to put those guardrails

0:07:34.840 --> 0:07:38.040
<v Speaker 5>and safeguards because there's broad sets of use cases. They're

0:07:38.120 --> 0:07:41.960
<v Speaker 5>very specific sets of use cases. But yeah, definitely there

0:07:42.000 --> 0:07:45.360
<v Speaker 5>should be policy that guides us at an overarching level

0:07:46.120 --> 0:07:49.440
<v Speaker 5>so that we could use that like anything we govern

0:07:49.560 --> 0:07:52.560
<v Speaker 5>right with technology. I think that is the role that

0:07:52.640 --> 0:07:53.600
<v Speaker 5>government should play.

0:07:54.920 --> 0:07:56.760
<v Speaker 2>What about when it comes to TikTok. It's in the

0:07:56.800 --> 0:08:00.520
<v Speaker 2>news right now, obviously because of that Supreme Court ruling

0:08:00.520 --> 0:08:02.880
<v Speaker 2>that we got. The future of the fate of the

0:08:02.920 --> 0:08:05.200
<v Speaker 2>app is uncertain. What are your thoughts on it.

0:08:06.520 --> 0:08:11.280
<v Speaker 5>We actually are a welcoming community to a big part

0:08:11.320 --> 0:08:13.760
<v Speaker 5>of TikTok called book talk. Book talk is one of

0:08:13.760 --> 0:08:17.120
<v Speaker 5>the largest communities on TikTok, and the last report I

0:08:17.200 --> 0:08:22.240
<v Speaker 5>read it said hashtag book talk is used by billions

0:08:22.280 --> 0:08:23.440
<v Speaker 5>of times across the board.

0:08:23.480 --> 0:08:25.000
<v Speaker 4>It's a very large community.

0:08:25.760 --> 0:08:29.080
<v Speaker 5>We definitely feel we're a safe haven for book talkers

0:08:29.120 --> 0:08:32.960
<v Speaker 5>on fable. Actually, people we do have quite a lot

0:08:32.960 --> 0:08:35.839
<v Speaker 5>of book talk influences today on Fable and a lot of.

0:08:35.760 --> 0:08:37.559
<v Speaker 4>People who use both platforms.

0:08:38.200 --> 0:08:41.080
<v Speaker 5>Many people that are currently on TikTok in book talk

0:08:41.160 --> 0:08:45.000
<v Speaker 5>call us the coziest, called fable as the cosiest corner

0:08:45.040 --> 0:08:47.520
<v Speaker 5>of the Internet. So we hope to be that and

0:08:47.559 --> 0:08:49.920
<v Speaker 5>we hope to welcome communities like that. I think the

0:08:49.960 --> 0:08:51.959
<v Speaker 5>good thing about TikTok, and I think what it has

0:08:52.040 --> 0:08:56.000
<v Speaker 5>shown us is that there is a power to creation

0:08:56.360 --> 0:09:00.320
<v Speaker 5>and community and that is sort of like neededeople to

0:09:00.400 --> 0:09:02.000
<v Speaker 5>express themselves, belong in.

0:09:01.920 --> 0:09:04.920
<v Speaker 4>A community, and be with like minded people.

0:09:05.640 --> 0:09:08.000
<v Speaker 5>So we definitely hope we will be the haven for

0:09:08.080 --> 0:09:11.559
<v Speaker 5>people that are currently on TikTok if it gets banned

0:09:11.640 --> 0:09:12.959
<v Speaker 5>or it is not allowed to.

0:09:12.880 --> 0:09:15.360
<v Speaker 2>Be used in You have two million users around the world.

0:09:15.360 --> 0:09:17.240
<v Speaker 2>Have you seen an uptake in the last couple of weeks.

0:09:17.240 --> 0:09:21.400
<v Speaker 2>Have people come into the platform in anticipation of this, Yes, we've.

0:09:21.240 --> 0:09:24.840
<v Speaker 5>Seen a huge uptick in the last four weeks, actually

0:09:24.840 --> 0:09:26.400
<v Speaker 5>even before when you started.

0:09:27.040 --> 0:09:28.800
<v Speaker 4>So we are just really hoping.

0:09:28.520 --> 0:09:32.800
<v Speaker 5>That this continues and we can host and welcome these

0:09:32.840 --> 0:09:34.160
<v Speaker 5>communities onto Fable.

0:09:34.240 --> 0:09:35.960
<v Speaker 2>So it sounds like a ban of TikTok would be

0:09:35.960 --> 0:09:37.240
<v Speaker 2>really good for Fable.

0:09:38.600 --> 0:09:41.120
<v Speaker 5>I think either way, like I said, we were we

0:09:41.360 --> 0:09:44.400
<v Speaker 5>had a lot of book talkers come to Fable.

0:09:44.480 --> 0:09:45.800
<v Speaker 4>Even with book talk, it is.

0:09:45.880 --> 0:09:49.200
<v Speaker 5>Very complimentary in the use case, so it's not like

0:09:49.280 --> 0:09:51.480
<v Speaker 5>we are routing for that to be banned or anything.

0:09:52.520 --> 0:09:56.200
<v Speaker 5>I think for us, TikTok continues, that's great because that

0:09:56.280 --> 0:09:59.880
<v Speaker 5>community will thrive and we will thrive with it. Unfortunately

0:10:00.160 --> 0:10:02.880
<v Speaker 5>that community goes away. We want to be the alternative

0:10:02.960 --> 0:10:03.960
<v Speaker 5>for people who love.

0:10:03.840 --> 0:10:06.760
<v Speaker 4>Books and TV shows and stories in general.

0:10:07.679 --> 0:10:11.800
<v Speaker 3>What would people maybe find surprising about your app and

0:10:12.360 --> 0:10:15.000
<v Speaker 3>the kind of activity that goes on. What is it

0:10:15.360 --> 0:10:18.920
<v Speaker 3>that you think might be a bit surprising A.

0:10:19.240 --> 0:10:21.959
<v Speaker 5>You know, I think that's a great question. I started

0:10:22.040 --> 0:10:25.800
<v Speaker 5>this company because I really wanted to bring the technology

0:10:25.840 --> 0:10:29.400
<v Speaker 5>world and the creative world together. At Fable, we say

0:10:29.440 --> 0:10:31.600
<v Speaker 5>we want to build a tech company with the soul

0:10:31.679 --> 0:10:35.160
<v Speaker 5>of an artist. So I feel we are very proud

0:10:35.200 --> 0:10:37.600
<v Speaker 5>of our design. We are very proud of the UI

0:10:37.800 --> 0:10:43.880
<v Speaker 5>that we create. It's beautiful. We celebrate human creation. Yet

0:10:43.920 --> 0:10:46.560
<v Speaker 5>we are a tech company. We're based, you know, I'm

0:10:46.600 --> 0:10:50.240
<v Speaker 5>based in Silicon Valley, our team is distributed. I think

0:10:50.240 --> 0:10:53.439
<v Speaker 5>people would find us to be unique in being able

0:10:53.480 --> 0:10:57.240
<v Speaker 5>to bring both of those together. The other thing that

0:10:57.320 --> 0:11:00.440
<v Speaker 5>I feel we are unique is we truly are a

0:11:00.559 --> 0:11:03.640
<v Speaker 5>platform if you love and every human loves stories. Right

0:11:03.679 --> 0:11:05.400
<v Speaker 5>at the end of the day, human beings are all

0:11:05.440 --> 0:11:08.240
<v Speaker 5>about the story we tell. If you love stories, we

0:11:08.320 --> 0:11:09.960
<v Speaker 5>are the destination for you.

0:11:10.000 --> 0:11:12.760
<v Speaker 3>Just got about twenty five seconds left here. How do

0:11:12.800 --> 0:11:14.760
<v Speaker 3>you make money and are you profitable? Just quickly?

0:11:15.760 --> 0:11:17.920
<v Speaker 4>We are very still early. We're a startup.

0:11:18.840 --> 0:11:22.160
<v Speaker 5>We you know, we are not yet quite monetizing the platform.

0:11:22.200 --> 0:11:23.560
<v Speaker 4>We'll begin to do that effort.

0:11:23.600 --> 0:11:26.360
<v Speaker 5>Today, the platform is free, which makes it easy for

0:11:26.440 --> 0:11:27.199
<v Speaker 5>people to join.

0:11:28.080 --> 0:11:31.199
<v Speaker 3>Will it be the subscription based ultimately or is that

0:11:31.280 --> 0:11:34.000
<v Speaker 3>the that's the revenue stream going forward quickly?

0:11:34.559 --> 0:11:38.360
<v Speaker 5>Yes. Yes, we will launch a subscription here soon and

0:11:38.440 --> 0:11:41.640
<v Speaker 5>hope to add traditional value added features for people to

0:11:41.720 --> 0:11:45.319
<v Speaker 5>track their links, to track of their progress, you know,

0:11:45.480 --> 0:11:47.880
<v Speaker 5>keep track of what interests their life, et cetera.

0:11:48.160 --> 0:11:50.920
<v Speaker 3>Got it come back? We'd love to hear more over

0:11:50.920 --> 0:11:54.360
<v Speaker 3>the year. Padna Warrior, founder president CEO of Fable joining

0:11:54.440 --> 0:11:57.120
<v Speaker 3>us on this Friday. This is Bloomberg Business Week