WEBVTT - Managing Risk in a World Where AI is Everywhere

0:00:00.160 --> 0:00:04.200
<v Speaker 1>This is Bloomberg Business Week with Carol Messer and Bloomberg

0:00:04.280 --> 0:00:08.400
<v Speaker 1>Quick Takes Tim Stinovic on Bloomberg Radio. Hey listen. AI.

0:00:08.520 --> 0:00:11.360
<v Speaker 1>Artificial intelligence not the thing of science fiction. It is

0:00:11.440 --> 0:00:14.240
<v Speaker 1>all over our world, already determines who we hire, what

0:00:14.320 --> 0:00:18.480
<v Speaker 1>we might invest in or buy, manager supply chains, provides governance.

0:00:18.600 --> 0:00:21.479
<v Speaker 1>It is everywhere. Last week I d C noting that

0:00:21.560 --> 0:00:24.119
<v Speaker 1>spending on AI solutions alone doubling in the US by

0:00:25.000 --> 0:00:28.640
<v Speaker 1>Tim two billion dollars. Pretty massive market and a massive

0:00:28.640 --> 0:00:31.720
<v Speaker 1>opportunity I think for a lot of companies. Bena Amanath

0:00:31.880 --> 0:00:35.400
<v Speaker 1>is Executive Director of Global Deloitte AI Institute, head of

0:00:35.520 --> 0:00:38.840
<v Speaker 1>Trustworthy AI and Ethical Tech at Deloitte as well. Joining

0:00:38.920 --> 0:00:42.040
<v Speaker 1>us now on the zoom from California. She's got a

0:00:42.080 --> 0:00:45.280
<v Speaker 1>new book out today. It's called Trustworthy AI, a Business

0:00:45.280 --> 0:00:49.440
<v Speaker 1>Guide for Navigating trust and Ethics in AI. So being

0:00:49.479 --> 0:00:51.600
<v Speaker 1>a give us an idea of you know, I think

0:00:51.600 --> 0:00:53.519
<v Speaker 1>before we actually get to the ethics AI that the

0:00:53.520 --> 0:00:55.560
<v Speaker 1>part of this that your book focuses on. I want

0:00:55.600 --> 0:00:59.160
<v Speaker 1>to talk a little bit about where where consumers experience AI,

0:00:59.280 --> 0:01:02.320
<v Speaker 1>Where are audience would actually interact with AI without even

0:01:02.320 --> 0:01:07.760
<v Speaker 1>knowing it? Everywhere if they're using a smartphone, they're using AI,

0:01:07.880 --> 0:01:10.560
<v Speaker 1>and which is the most prominent use of AI today

0:01:10.600 --> 0:01:13.800
<v Speaker 1>without your knowing that everything that they use, right from

0:01:13.880 --> 0:01:18.040
<v Speaker 1>their maps to the messaging to the social media, banking,

0:01:18.240 --> 0:01:22.360
<v Speaker 1>everything is powered by AI today and it's in their pockets,

0:01:22.400 --> 0:01:24.560
<v Speaker 1>in their phone right now. So it's a good thing.

0:01:25.040 --> 0:01:29.000
<v Speaker 1>I say that with a little sarcasm. I mean, how

0:01:29.000 --> 0:01:31.200
<v Speaker 1>do we like? It could be very helpful in supply

0:01:31.280 --> 0:01:34.720
<v Speaker 1>chain management if you think about it, in governance. I

0:01:34.760 --> 0:01:37.520
<v Speaker 1>remember doing some stories about venture capitalists being able to

0:01:37.560 --> 0:01:40.920
<v Speaker 1>look at uh, smaller deals that maybe wouldn't make sense

0:01:40.959 --> 0:01:44.040
<v Speaker 1>to spend time on, but using technology, using AI, they

0:01:44.040 --> 0:01:46.120
<v Speaker 1>can kind of screen through things. So how do you

0:01:46.240 --> 0:01:48.320
<v Speaker 1>think about AI? The good, the bad, and the ugly.

0:01:49.560 --> 0:01:52.000
<v Speaker 1>That's a great question, Carol, And you know, I think

0:01:52.000 --> 0:01:54.600
<v Speaker 1>there's a lot of goodness that AI can bring to

0:01:54.720 --> 0:01:58.160
<v Speaker 1>humanity as a whole, but there are side effects as well.

0:01:58.240 --> 0:02:00.600
<v Speaker 1>And today we are in that era of it. Everybody

0:02:00.680 --> 0:02:03.840
<v Speaker 1>is focused on the value creation of AI, and we

0:02:03.920 --> 0:02:06.680
<v Speaker 1>see these high level headlines on all the bad things

0:02:06.720 --> 0:02:09.720
<v Speaker 1>AI is doing. The reality is that it's a balance.

0:02:10.000 --> 0:02:11.880
<v Speaker 1>There is a lot of goodness and there's a lot

0:02:11.880 --> 0:02:16.639
<v Speaker 1>of value positive value things what you mentioned, supply chain optimization, translation,

0:02:16.760 --> 0:02:19.959
<v Speaker 1>patient diagnosis. There's a lot of goodness that's coming out,

0:02:20.000 --> 0:02:23.760
<v Speaker 1>but there's also unintended consequences that's coming into the to

0:02:23.880 --> 0:02:26.760
<v Speaker 1>the forefront as AI is scaling out into more and

0:02:26.800 --> 0:02:28.600
<v Speaker 1>more applications. Okay, I want to get to some of

0:02:28.639 --> 0:02:30.880
<v Speaker 1>those unintended consequences, but I also have you know, I

0:02:30.919 --> 0:02:33.720
<v Speaker 1>think a lot of I come from a place that

0:02:33.960 --> 0:02:37.200
<v Speaker 1>a lot of our listeners and viewers can relate to.

0:02:37.360 --> 0:02:39.239
<v Speaker 1>But you know, the idea when you're on the phone

0:02:39.400 --> 0:02:42.119
<v Speaker 1>trying to tall an airline, for example, or perhaps get

0:02:42.160 --> 0:02:45.160
<v Speaker 1>something fixed, and you're you know, you have AI trying

0:02:45.160 --> 0:02:46.720
<v Speaker 1>to work with you, and like it's like, you know,

0:02:47.240 --> 0:02:50.799
<v Speaker 1>represent you just keep repeating representative over and over again,

0:02:50.919 --> 0:02:54.440
<v Speaker 1>right because the bond just isn't doing it for you.

0:02:54.560 --> 0:02:57.520
<v Speaker 1>And I think that's frustrating for a lot of consumers.

0:02:57.560 --> 0:03:00.120
<v Speaker 1>That's true, Tim, I'm with you. I had a lot

0:03:00.120 --> 0:03:03.920
<v Speaker 1>of challenge, especially with my accent, which doesn't fit into

0:03:03.919 --> 0:03:07.240
<v Speaker 1>the norm right to get communicate anything. And you know,

0:03:07.480 --> 0:03:11.519
<v Speaker 1>at some point you're just yelling representative, right. So A look,

0:03:11.680 --> 0:03:15.400
<v Speaker 1>the reality is that technology is still way in its infancy,

0:03:15.560 --> 0:03:18.080
<v Speaker 1>right because the research is still happening. It's not a

0:03:18.120 --> 0:03:21.040
<v Speaker 1>fully developed technology, but using it in the real world.

0:03:21.320 --> 0:03:25.680
<v Speaker 1>Since it's not fully developed, there are that unintended consequences

0:03:25.720 --> 0:03:28.880
<v Speaker 1>that not that's not well thought through. So you're using

0:03:29.200 --> 0:03:33.280
<v Speaker 1>think of it, you're using something that is not fully mature, right,

0:03:33.360 --> 0:03:37.720
<v Speaker 1>and it is growing along with you while you're using it.

0:03:37.720 --> 0:03:41.320
<v Speaker 1>It's learning from you. It's learning my accent and adopting

0:03:41.320 --> 0:03:43.840
<v Speaker 1>it and in the next operation somebody with a similar

0:03:43.840 --> 0:03:46.720
<v Speaker 1>accents by it will be easier to recognize. So it

0:03:46.840 --> 0:03:49.440
<v Speaker 1>is learning and growing, and that's why there's an element

0:03:49.480 --> 0:03:51.839
<v Speaker 1>of patients that needs to come into play because it's

0:03:51.880 --> 0:03:55.360
<v Speaker 1>not being trained for every possible scenario. You want to

0:03:55.360 --> 0:03:58.800
<v Speaker 1>get back to her Guest Puma is executive director Global

0:03:58.840 --> 0:04:01.920
<v Speaker 1>Deloite AI Institute, Head of Trustworthy AI and Ethical Tech

0:04:01.960 --> 0:04:04.920
<v Speaker 1>at A Lloyd and the book she's got out Trustworthy,

0:04:05.040 --> 0:04:07.640
<v Speaker 1>a Business Guide for Navigating trust in Ethics and AI.

0:04:07.720 --> 0:04:09.920
<v Speaker 1>I want to get back to transparency or bring that up.

0:04:09.960 --> 0:04:12.520
<v Speaker 1>But you know, let me ask you, is AI what's

0:04:12.520 --> 0:04:14.720
<v Speaker 1>going to determine in a self driving car whether or

0:04:14.760 --> 0:04:16.599
<v Speaker 1>not you hit the dog, the cat, or the cyclist

0:04:18.120 --> 0:04:22.560
<v Speaker 1>or are none hopefully right? Thank you Tim, That's that's

0:04:22.560 --> 0:04:24.599
<v Speaker 1>a great question, and I think you know it is

0:04:24.960 --> 0:04:27.400
<v Speaker 1>at this point it is really up to that the

0:04:27.520 --> 0:04:30.600
<v Speaker 1>company who's creating the self driving car that was determining

0:04:30.600 --> 0:04:33.919
<v Speaker 1>it until we get regulations in place, And my book

0:04:34.000 --> 0:04:36.760
<v Speaker 1>really gives that company a structure to think about it

0:04:37.040 --> 0:04:41.400
<v Speaker 1>and offer potentially offer those options to the consumer or potentially,

0:04:41.440 --> 0:04:44.080
<v Speaker 1>you know, bring together a focus group to decide it.

0:04:44.160 --> 0:04:46.479
<v Speaker 1>But there needs to be a structured way of thinking it,

0:04:46.960 --> 0:04:49.880
<v Speaker 1>unlike just building that technology and deploying it out into

0:04:49.920 --> 0:04:52.400
<v Speaker 1>the world and then you know, seeing what happens. So

0:04:52.440 --> 0:04:54.720
<v Speaker 1>I think there needs to be more mindfulness, more thought

0:04:54.800 --> 0:04:58.040
<v Speaker 1>put into what are the possible scenarios and how do

0:04:58.080 --> 0:05:00.960
<v Speaker 1>you mitigate those? What are the questions that executives come

0:05:01.000 --> 0:05:02.640
<v Speaker 1>to you with when it comes to AI, like what

0:05:02.680 --> 0:05:07.200
<v Speaker 1>are their goals uh in sort of building out projects

0:05:07.279 --> 0:05:10.279
<v Speaker 1>that rely on AI technology? And then what do you

0:05:10.320 --> 0:05:12.120
<v Speaker 1>have to tell them in terms of guard rails and

0:05:12.120 --> 0:05:13.840
<v Speaker 1>in terms of making sure they don't make mistakes of

0:05:13.920 --> 0:05:17.039
<v Speaker 1>the past. That's a great question, Jim, and you know

0:05:17.760 --> 0:05:20.120
<v Speaker 1>many of them are just looking for how can I

0:05:20.200 --> 0:05:22.840
<v Speaker 1>get value from AI from my business? And there are

0:05:22.839 --> 0:05:26.719
<v Speaker 1>two options. One is either through new revenue opportunities, meaning

0:05:26.920 --> 0:05:31.000
<v Speaker 1>building new AI products that couldn't be done with existing technology.

0:05:31.080 --> 0:05:34.760
<v Speaker 1>So there's put a brand new revenue opportunities. Or the

0:05:34.760 --> 0:05:37.960
<v Speaker 1>other is optimizing or cost savings. Right, how do you

0:05:38.360 --> 0:05:41.360
<v Speaker 1>optimize your current supply chain process for example to get

0:05:41.400 --> 0:05:44.640
<v Speaker 1>better value for the goods that you buy from your sourcers?

0:05:44.680 --> 0:05:48.640
<v Speaker 1>How do you optimize your document management for example? So

0:05:48.680 --> 0:05:50.839
<v Speaker 1>there are those are the two you know ways you

0:05:50.880 --> 0:05:54.440
<v Speaker 1>can get value for from AI for your business today,

0:05:55.040 --> 0:05:58.040
<v Speaker 1>and most of them at that point are not thinking

0:05:58.080 --> 0:06:01.360
<v Speaker 1>about guard rails or the ways could go wrong. So

0:06:01.520 --> 0:06:04.760
<v Speaker 1>really it is up to us as the general audience,

0:06:04.839 --> 0:06:07.760
<v Speaker 1>as the public, and definitely mean my role is to

0:06:07.880 --> 0:06:11.200
<v Speaker 1>raise that awareness. Look, you're focusing on all the positive

0:06:11.279 --> 0:06:14.880
<v Speaker 1>value creation, but there are risks associated with it. The

0:06:14.920 --> 0:06:18.120
<v Speaker 1>obvious ones that the brand and reputational risk, but there

0:06:18.160 --> 0:06:21.080
<v Speaker 1>are really bad things that can happen if you do

0:06:21.160 --> 0:06:24.680
<v Speaker 1>not think about the ethical implications up front and put

0:06:24.680 --> 0:06:27.720
<v Speaker 1>in those guard rails. Like like, what's the thing that

0:06:27.760 --> 0:06:30.359
<v Speaker 1>you that you worry most about in terms of AI

0:06:30.480 --> 0:06:34.440
<v Speaker 1>and maybe the misuse or the unethical impact of it.

0:06:35.440 --> 0:06:38.479
<v Speaker 1>The thing I worry most about is that we are

0:06:38.640 --> 0:06:40.800
<v Speaker 1>not thinking about the ways it could go wrong, and

0:06:40.920 --> 0:06:43.680
<v Speaker 1>you can take any scenario right and it is. The

0:06:43.800 --> 0:06:46.880
<v Speaker 1>challenge is that it's not a one size fits all.

0:06:46.960 --> 0:06:49.480
<v Speaker 1>There is no single answer. Like you think of something

0:06:49.600 --> 0:06:53.160
<v Speaker 1>as you know, as simple as personalized marketing, right where

0:06:53.240 --> 0:06:56.599
<v Speaker 1>you are taking consumer data, matching it with products and

0:06:56.640 --> 0:07:01.880
<v Speaker 1>providing personalized ads. Right, But that same person, that same technology,

0:07:02.000 --> 0:07:06.640
<v Speaker 1>the AI can be used for providing for potentially personalized healthcare.

0:07:06.960 --> 0:07:09.840
<v Speaker 1>Now when you think about bias in this scenario, the

0:07:10.360 --> 0:07:13.840
<v Speaker 1>bias in the personalized marketing world means you know, the

0:07:13.880 --> 0:07:16.960
<v Speaker 1>wrong add to the wrong person, but in the healthcare

0:07:17.000 --> 0:07:19.880
<v Speaker 1>space it means a wrong diagnosis to the wrong person,

0:07:19.960 --> 0:07:22.440
<v Speaker 1>which is terrible, right, So I think it has to

0:07:22.480 --> 0:07:24.560
<v Speaker 1>be weighted. So what I try to make sure is

0:07:24.600 --> 0:07:28.120
<v Speaker 1>you think about the nuances the context, because AI as

0:07:28.160 --> 0:07:33.040
<v Speaker 1>a whole is about intelligence, and intelligence is different based

0:07:33.080 --> 0:07:36.680
<v Speaker 1>on the scenario, the solution, the industry that you're in.

0:07:37.040 --> 0:07:39.040
<v Speaker 1>It is not going to be a one size fit

0:07:39.040 --> 0:07:41.440
<v Speaker 1>at all. So how do you weigh it as a

0:07:41.480 --> 0:07:44.720
<v Speaker 1>CEO to say which other risks were I'm willing to

0:07:44.760 --> 0:07:46.840
<v Speaker 1>take and which other is I'm not willing to take?

0:07:46.880 --> 0:07:49.760
<v Speaker 1>And that I need to think more about. I'm interested

0:07:49.800 --> 0:07:52.480
<v Speaker 1>in when you know you you have a background in

0:07:52.480 --> 0:07:57.800
<v Speaker 1>this is you know that engineering um background in obviously

0:07:57.840 --> 0:08:01.040
<v Speaker 1>in technology. And I'm wondering from consumer side, not from

0:08:01.040 --> 0:08:04.520
<v Speaker 1>the consultant side. When you've been out in the world

0:08:04.960 --> 0:08:08.440
<v Speaker 1>and you've encountered AI and you've thought to yourself, this

0:08:08.560 --> 0:08:11.880
<v Speaker 1>is exactly the right way that this company should be

0:08:11.960 --> 0:08:15.840
<v Speaker 1>using AI. What's an example of that? Well, you know,

0:08:15.960 --> 0:08:18.920
<v Speaker 1>I think the ones that that I struggle with most

0:08:19.080 --> 0:08:21.680
<v Speaker 1>are the you know, where there is a human impact.

0:08:22.000 --> 0:08:24.280
<v Speaker 1>And I'll give you two scenarios. You know, let's take

0:08:24.320 --> 0:08:27.600
<v Speaker 1>facial recognition. Everybody has heard about it and has heard

0:08:27.960 --> 0:08:31.120
<v Speaker 1>all the ways it could go back. Right, So, facial

0:08:31.160 --> 0:08:34.760
<v Speaker 1>recognition as as a technology, when it's biased and used

0:08:34.760 --> 0:08:38.920
<v Speaker 1>in a law enforcement scenario, it's a terrible thing. But

0:08:39.440 --> 0:08:44.200
<v Speaker 1>there is also facial recognition being used by startups to

0:08:44.360 --> 0:08:47.880
<v Speaker 1>recognize human trafficking victims. Right. And look, I gotta tell

0:08:47.920 --> 0:08:50.920
<v Speaker 1>you too. Recently, My parents recently flew internationally and I

0:08:50.960 --> 0:08:52.600
<v Speaker 1>talked to them and they told me that in order

0:08:52.640 --> 0:08:55.080
<v Speaker 1>to board the plane, to order to board the plane,

0:08:55.080 --> 0:08:57.240
<v Speaker 1>they just walked on because they didn't have to give

0:08:57.240 --> 0:09:02.559
<v Speaker 1>their boarding past to any When the facial recognition recognize them.

0:09:00.120 --> 0:09:05.640
<v Speaker 1>I think exactly cool. That's a great scenario that who

0:09:05.760 --> 0:09:09.880
<v Speaker 1>we're creepy, Like they don't like creepy stuff right there,

0:09:09.920 --> 0:09:11.640
<v Speaker 1>like you know, kind of late adopters. They like the

0:09:11.679 --> 0:09:15.600
<v Speaker 1>ease of it. Yeah, they like the ease of it. Yes.

0:09:15.760 --> 0:09:18.160
<v Speaker 1>The coolest one I've heard with facial recognition is when

0:09:18.160 --> 0:09:21.520
<v Speaker 1>you're using it at traffic light to recognize human trafficking victims.

0:09:22.120 --> 0:09:25.000
<v Speaker 1>And that's why that nuance comes in. Right, Yes, the

0:09:25.040 --> 0:09:27.800
<v Speaker 1>technology might be biased, but if it is helping you

0:09:27.920 --> 0:09:33.120
<v Speaker 1>rescue eighty percent more victims then you could without AI,

0:09:33.320 --> 0:09:36.719
<v Speaker 1>then that's a decision that you know, the organization and

0:09:36.760 --> 0:09:40.120
<v Speaker 1>we need to make right. It is biased, but it

0:09:40.200 --> 0:09:44.320
<v Speaker 1>is still helping rescue eighty person more with Without it,

0:09:44.480 --> 0:09:47.719
<v Speaker 1>you would not get that eighty person of the victims rescue, right,

0:09:47.760 --> 0:09:50.880
<v Speaker 1>So it's an informed decision on Yes, we know it's biased,

0:09:51.080 --> 0:09:53.199
<v Speaker 1>but we're still going to use it in this scenario

0:09:53.280 --> 0:09:55.880
<v Speaker 1>because it's helping us. But you know, in a law

0:09:55.960 --> 0:09:59.320
<v Speaker 1>enforcement scenario, even if it's just eighty percent, you know,

0:10:00.080 --> 0:10:02.959
<v Speaker 1>curate it's a terrible thing. You cannot use it, right,

0:10:03.000 --> 0:10:05.280
<v Speaker 1>So it depends on the scenario in which you're using

0:10:05.280 --> 0:10:07.560
<v Speaker 1>it the example you gave about your parents being able

0:10:07.600 --> 0:10:09.920
<v Speaker 1>to walk in right, it's the great rus right, what's

0:10:09.920 --> 0:10:13.880
<v Speaker 1>the worst case that can happen? Pebody, Yeah, it just

0:10:13.960 --> 0:10:17.480
<v Speaker 1>kind about fifty seconds left here. I mean, we've already

0:10:17.520 --> 0:10:20.640
<v Speaker 1>given up a fair amount of privacy right because of AI,

0:10:20.760 --> 0:10:24.880
<v Speaker 1>and we probably will continue to do so. Yes, Yes,

0:10:25.080 --> 0:10:28.520
<v Speaker 1>And the reality is the definition of privacy has really

0:10:28.600 --> 0:10:31.440
<v Speaker 1>changed in the last two and a half years. Before UNIT,

0:10:31.440 --> 0:10:34.280
<v Speaker 1>it was all about all data sharing, data usages terrible.

0:10:34.760 --> 0:10:37.199
<v Speaker 1>You know, privacy needs to be protected. But what has

0:10:37.240 --> 0:10:40.280
<v Speaker 1>happened with contact tracing? If you don't share your data,

0:10:40.360 --> 0:10:44.599
<v Speaker 1>it's actually bad, right. So I think the evolving definitions

0:10:44.640 --> 0:10:48.439
<v Speaker 1>of some of these dimensions of trustworthy AI is going

0:10:48.480 --> 0:10:51.440
<v Speaker 1>to continue to evolve and shape. So we just have

0:10:51.600 --> 0:10:54.840
<v Speaker 1>to be agile about it. Wow, a lot of issues

0:10:54.880 --> 0:10:57.440
<v Speaker 1>and certainly something we've scratched the surface that feels like

0:10:57.480 --> 0:10:59.960
<v Speaker 1>and certainly something that will be a bigger part of

0:11:00.000 --> 0:11:02.800
<v Speaker 1>our conversation as I just kind of invades all parts

0:11:02.800 --> 0:11:04.960
<v Speaker 1>of our world in a good way, But there's also

0:11:05.080 --> 0:11:08.360
<v Speaker 1>things to be concerned about. As you just laid out. Beina,

0:11:08.360 --> 0:11:11.319
<v Speaker 1>I'm a math. She is executive director at Global Deloitte

0:11:11.360 --> 0:11:14.760
<v Speaker 1>AI Institute, Head of Trustworthy AI and Ethical Tech at Deloitte.

0:11:14.880 --> 0:11:18.800
<v Speaker 1>Her new book is Trustworthy AI, a Business Guide for

0:11:18.920 --> 0:11:21.400
<v Speaker 1>navigating trust and ethics and AI. A lot of really,

0:11:21.440 --> 0:11:23.840
<v Speaker 1>a lot of fascinating conversations, like conversation to be had

0:11:23.880 --> 0:11:26.160
<v Speaker 1>around AI and treading and to understand like where we're

0:11:26.200 --> 0:11:28.040
<v Speaker 1>using it in the background right exactly