WEBVTT - A Conversation with Beena Ammanath 0:00:04.400 --> 0:00:07.800 Welcome to tex Stuff, a production from I Heart Radio. 0:00:12.200 --> 0:00:16.000 Hey therein Welcome to tex Stuff. I'm your host, Jonathan Strickland. 0:00:16.000 --> 0:00:18.919 I'm an executive producer with I Heart Radio. And how 0:00:19.040 --> 0:00:22.560 the tech are you? I gotta treat view folks. Today 0:00:22.600 --> 0:00:26.599 I had the opportunity to speak with Bena Amnath, the 0:00:26.640 --> 0:00:31.520 executive director of Deloitte AI Institute. Vina is an accomplished 0:00:31.560 --> 0:00:37.960 technologist and expert in artificial intelligence. She's a coder, she's, 0:00:38.560 --> 0:00:42.159 you know, an engineer, and she's a great communicator too. 0:00:42.240 --> 0:00:47.600 She has appeared on numerous shows and panels talking about AI. 0:00:47.680 --> 0:00:51.960 She's also the author of a book called Trustworthy AI. 0:00:52.600 --> 0:00:56.360 And this was a fantastic opportunity to speak with someone 0:00:56.400 --> 0:00:59.600 who actually has a deep amount of experience in the 0:00:59.640 --> 0:01:04.600 field and really talk about some of the big concepts 0:01:04.600 --> 0:01:07.200 in AI and get a little more perspective on them. 0:01:07.400 --> 0:01:13.520 And I have to admit Beena's responses really open up 0:01:13.560 --> 0:01:16.640 the blinders that I have on And of course I'm 0:01:16.680 --> 0:01:18.560 like a lot of people, right I I go through 0:01:18.600 --> 0:01:21.560 life thinking I have a pretty good handle on this. 0:01:21.720 --> 0:01:23.959 I think I know what's going on, and then I 0:01:23.959 --> 0:01:27.319 meet someone else who has had a you know, a 0:01:27.319 --> 0:01:30.880 different experience, and especially a different depth of experience in 0:01:30.880 --> 0:01:34.360 a particular field and realize, oh gosh, I hadn't even 0:01:34.400 --> 0:01:41.120 considered some of these specific scenarios, for example. So I 0:01:41.240 --> 0:01:45.240 really very much enjoyed my conversation with Bena. She's also 0:01:45.720 --> 0:01:50.240 incredibly good at putting things in a way that are 0:01:50.360 --> 0:01:54.040 easily understandable. A lot of technologists, when you start to 0:01:54.040 --> 0:01:58.880 talk with them, they get really heavy with jargon or 0:01:59.040 --> 0:02:03.040 or concepts. That makes sense if you've had experience working 0:02:03.080 --> 0:02:06.400 in that area, but if you haven't, your eyes kind 0:02:06.400 --> 0:02:09.799 of glaze over and you just trust that what they 0:02:09.800 --> 0:02:12.160 say makes sense. That was not the issue with Bina. 0:02:12.320 --> 0:02:16.600 She is really good at talking about this stuff on 0:02:16.680 --> 0:02:20.239 a level that that the average person can easily understand, 0:02:21.040 --> 0:02:25.840 and yet also really stressing how AI is a very 0:02:25.880 --> 0:02:28.839 important component today. I mean, we're seeing it rolled out 0:02:28.919 --> 0:02:33.520 in all sorts of different ways across all different sectors. 0:02:34.120 --> 0:02:38.079 We mostly talked about business in this conversation, but clearly 0:02:38.680 --> 0:02:43.880 AI is everywhere. Whether we're talking about facial recognition technology 0:02:43.880 --> 0:02:46.880 that might be built directly into the camera on your phone, 0:02:47.560 --> 0:02:51.440 or maybe we're talking about a personal digital assistant, you know, 0:02:51.520 --> 0:02:55.040 something like the Amazon one. I won't say her name 0:02:55.240 --> 0:02:58.320 because some of you have her and she gets real 0:02:58.720 --> 0:03:02.680 like she she perks up when you say her name. Um, 0:03:02.720 --> 0:03:05.680 those sort of things obviously have components of AI built 0:03:05.680 --> 0:03:08.680 into them, but we were really looking at things like 0:03:09.680 --> 0:03:14.840 processes in business where you might need to use automation 0:03:14.919 --> 0:03:23.320 and artificial intelligence to make complicated processes more efficient and 0:03:23.840 --> 0:03:27.520 less human intensive. And so, yeah, this was a great 0:03:27.560 --> 0:03:30.320 conversation and I really feel like I learned a lot. 0:03:30.360 --> 0:03:34.200 I hope that you all enjoy it. And again, she 0:03:34.320 --> 0:03:37.440 does have a new book out. It's called Trustworthy AI, 0:03:37.520 --> 0:03:39.360 and there's a copy on the way to me, so 0:03:39.400 --> 0:03:43.680 I'm very eager to read it myself because just talking 0:03:43.680 --> 0:03:46.840 with Bina I felt like I was just scratching the surface. 0:03:47.000 --> 0:03:49.400 But you're gonna hear all that, So let's get to 0:03:49.640 --> 0:03:53.760 that interview. Bina. I want to welcome you to the show. 0:03:54.040 --> 0:03:57.119 I am so pleased to have an expert on AI. 0:03:57.160 --> 0:04:02.320 Trustworthy AI, no less, Welcome to tech stuff, Jonathan, thank 0:04:02.360 --> 0:04:04.720 you so much for having me on your show. I 0:04:04.800 --> 0:04:08.200 really enjoy your episodes, so I'm looking forward to having 0:04:08.240 --> 0:04:11.360 this conversation with you. I am as well. And one 0:04:11.400 --> 0:04:13.400 of the things that I like to do is kind 0:04:13.440 --> 0:04:18.800 of set some foundation for any kind of conversation around AI, 0:04:18.839 --> 0:04:22.480 because in my experience, and I'm sure you've experienced something 0:04:22.520 --> 0:04:26.000 similar chatting with people about AI, it seems like everyone 0:04:26.080 --> 0:04:30.680 has a different, sometimes very specific idea of what AI is. 0:04:31.279 --> 0:04:35.960 And I'm curious, how do you describe AI to people? Yeah, 0:04:36.040 --> 0:04:39.680 that's a great question to start with. So AI is 0:04:39.720 --> 0:04:43.520 a form of intelligence that uses machines to do things 0:04:43.600 --> 0:04:51.039 that traditionally required human intelligence. So it is artificial intelligence 0:04:51.040 --> 0:04:55.680 which is created artificially by machines. So now I like 0:04:55.800 --> 0:04:59.919 that description because it covers such a wide spectrum every 0:05:00.080 --> 0:05:04.039 thing from sort of the science fiction approach we've all 0:05:04.480 --> 0:05:08.000 seen about machines that seem to think like humans to 0:05:08.080 --> 0:05:10.200 a point where they usually become the threat. I mean, 0:05:10.240 --> 0:05:12.680 that's typically the way we look at it, which is 0:05:12.760 --> 0:05:14.360 I'm sure going to come into play when we talk 0:05:14.360 --> 0:05:17.120 about trustworthiness, because I'm sure a lot of people aren't 0:05:17.120 --> 0:05:22.320 aware how AI can sometimes be a danger, but not 0:05:22.400 --> 0:05:27.680 necessarily like sky Net from Terminator type danger. Let me 0:05:27.720 --> 0:05:31.920 elaborate on that description than a little bit more on 0:05:32.200 --> 0:05:35.640 you know that next level down on AI definition, Right, 0:05:36.080 --> 0:05:38.200 you know, the way I think about it. There are 0:05:38.200 --> 0:05:43.080 three types of AI. One is artificial narrow intelligence, which 0:05:43.160 --> 0:05:47.480 can do a very specific, narrow task that a human 0:05:47.680 --> 0:05:52.760 can do, like sort a bunch of photographs, Right, That's 0:05:52.760 --> 0:05:56.360 a very narrow specific task. So that's artificial narrow intelligence. 0:05:56.720 --> 0:06:00.000 And then there is a form of artificial general intelligence, 0:06:00.440 --> 0:06:03.239 which is a form of AI that can do any 0:06:03.360 --> 0:06:06.680 task that human beings can do, right, So it is 0:06:06.720 --> 0:06:09.840 pretty much everything that a human being can do. And 0:06:09.880 --> 0:06:12.719 then I think of a third category, which is artificial 0:06:12.920 --> 0:06:17.000 super intelligence, which is a form of intelligence which is 0:06:17.120 --> 0:06:22.239 smarter than all human beings combined and can do more 0:06:22.360 --> 0:06:26.920 things that human intelligence couldn't do. So when we talk 0:06:27.000 --> 0:06:32.839 about AI in the business world or in reality, it 0:06:33.000 --> 0:06:35.479 is where we are with AI. It's very much in 0:06:35.520 --> 0:06:39.440 that artificial narrow intelligence space. But when we hear a 0:06:39.480 --> 0:06:43.200 lot about the in the media or the hype and 0:06:43.320 --> 0:06:46.880 the fear, you know, it's talking really about that super 0:06:46.960 --> 0:06:50.160 intelligence phase, which is a form of AI that is 0:06:50.200 --> 0:06:55.040 smarter than all human beings combined and has more capabilities 0:06:55.080 --> 0:06:57.679 that human intelligence. And I think there's a big gap 0:06:57.760 --> 0:07:02.200 between reality and between where we you know, we are 0:07:02.240 --> 0:07:06.159 anticipating things to be right and and you know. Part 0:07:06.160 --> 0:07:09.320 of the reason is, you know, actual super intelligence is 0:07:09.360 --> 0:07:12.800 a lot of our human imagination, which is where AI 0:07:13.120 --> 0:07:17.240 was when I was studying years ago, right, So I 0:07:17.320 --> 0:07:19.720 do think there is value in imagination. I do think 0:07:19.720 --> 0:07:22.200 there is value in thinking of worst case scenarios so 0:07:22.240 --> 0:07:25.040 that you can address it. But the reality is we 0:07:25.160 --> 0:07:28.440 still today don't have the tools or the capabilities to 0:07:28.520 --> 0:07:32.640 build out artificial general intelligence or super intelligence, and we 0:07:32.720 --> 0:07:36.320 do not have that capability today. I see parallels in 0:07:36.320 --> 0:07:40.160 this as well, like I have a very similar description 0:07:40.440 --> 0:07:44.400 of autonomous cars, for example, like people talk about autonomous 0:07:44.480 --> 0:07:48.240 cars like we've reached level five autonomy, when really I 0:07:48.280 --> 0:07:51.760 would argue we're still around level two creeping into level three, 0:07:51.880 --> 0:07:54.280 but we we are not close to level four or five. 0:07:55.080 --> 0:07:57.960 And this is why I like having this kind of 0:07:57.960 --> 0:08:00.720 conversation right up front, so that people kind of set 0:08:00.760 --> 0:08:06.640 their expectations, because artificial intelligence can already do incredible things 0:08:06.640 --> 0:08:12.600 in these very narrow, narrow uses, and I'm blown away 0:08:12.600 --> 0:08:15.480 by it whenever I learned about that. But I do 0:08:15.640 --> 0:08:19.840 also see the the allure and sometimes the trap of 0:08:20.360 --> 0:08:25.000 extrapolating that beyond the narrow cases and thinking what happens 0:08:25.080 --> 0:08:29.239 when this goes beyond that, which it could very well happen, 0:08:29.480 --> 0:08:33.600 but we're not we're not at that stage yet. Um. 0:08:33.720 --> 0:08:37.319 But but I think of things like, yes, I think 0:08:37.320 --> 0:08:40.600 of things like like the image recognition like that to 0:08:40.679 --> 0:08:45.520 me is still an amazing thing to see developed, Like 0:08:45.960 --> 0:08:48.840 you know it is. Ever since I started covering tech, 0:08:49.360 --> 0:08:54.480 the ability has grown so fast. Like I remember when, 0:08:54.679 --> 0:08:58.160 at least on the consumer side, you might see something 0:08:58.200 --> 0:09:01.640 that was like detecting a ace, not recognizing a face, 0:09:01.679 --> 0:09:05.720 but detecting the structure that makes a face and for 0:09:05.760 --> 0:09:09.320 a camera, And now you know that that looks like 0:09:09.360 --> 0:09:14.360 stone age technology by comparison of what we're seeing today. Yes, Jarthan, 0:09:14.520 --> 0:09:16.880 And you I know you've been cowering tech for a 0:09:16.920 --> 0:09:23.160 long time. You know you've certainly seen seen the early evolution, right, 0:09:23.240 --> 0:09:26.880 but you know, and look, when I studied computer science, 0:09:27.320 --> 0:09:31.240 I did program assembly language programming basically using zeros and 0:09:31.400 --> 0:09:34.760 ones right that level. And you know the languages that 0:09:34.840 --> 0:09:38.640 I use, like Pascal and Fortran lots, you know those 0:09:38.640 --> 0:09:42.280 don't even exist today, right, So there's a whole evolution happening. 0:09:42.600 --> 0:09:45.600 And I do think that is a big component to 0:09:45.840 --> 0:09:49.600 imagine the future so that we can at least go 0:09:49.760 --> 0:09:53.240 towards take care of the risk, and focus on all 0:09:53.240 --> 0:09:56.880 the good things that you know AI and technology can do. Right, 0:09:56.960 --> 0:10:00.000 So imagination is a good thing, but not the fear 0:10:00.000 --> 0:10:03.080 a part. And also I think I think what you 0:10:03.440 --> 0:10:06.800 said kind of is a great message to anyone who's 0:10:06.840 --> 0:10:10.720 interested in really focusing on AI. The fact that you 0:10:10.800 --> 0:10:15.040 were working in assembly so close such a low level language, 0:10:15.520 --> 0:10:19.320 you get you get a real familiarity with what these 0:10:19.520 --> 0:10:24.920 machines can do and their their potential that I think, uh, 0:10:25.240 --> 0:10:27.719 you you almost lose when you start working on high 0:10:27.840 --> 0:10:33.000 level uh programming languages Like you you get so focused 0:10:33.080 --> 0:10:35.880 on what the programming language lets you do. But if 0:10:35.880 --> 0:10:38.080 you've worked out at that low level, you're like, hey, no, 0:10:38.280 --> 0:10:41.920 I know circuits and wires, Okay, I am, I am, 0:10:41.960 --> 0:10:44.920 I am one step away from this machine. Yeah, we 0:10:45.080 --> 0:10:47.120 we have to realize, you know, just like you know, 0:10:47.200 --> 0:10:49.520 when you talk about it today, will talk mostly about 0:10:49.520 --> 0:10:52.480 the software, but the hardware is also evolving. Right. We 0:10:52.559 --> 0:10:56.160 certainly don't have wax of those massive back from pure 0:10:56.320 --> 0:11:00.520 systems that we had to program, right. I think there 0:11:00.640 --> 0:11:05.520 is evolution happening in every dimension, and you know, it's 0:11:05.559 --> 0:11:08.960 it's part of the growth of AI or any technology 0:11:09.000 --> 0:11:11.880 if you think about it. M hmm, Well, I also 0:11:11.960 --> 0:11:14.480 want to know what you mean when you when you 0:11:14.600 --> 0:11:18.560 use the phrase trustworthy AI. So what is it that 0:11:18.679 --> 0:11:23.439 makes AI trustworthy? And what what's the what's the alternative? 0:11:23.480 --> 0:11:28.880 What is the untrustworthy side? Yeah, that's that's a great question, 0:11:28.960 --> 0:11:31.960 and that's something that you know. As a technologist, I'm 0:11:32.160 --> 0:11:35.760 enamored by all the cool things that AI can do 0:11:36.120 --> 0:11:39.400 because I just focus on all the value creation. But 0:11:39.559 --> 0:11:43.120 over the past a few years, as AI started becoming real, 0:11:43.200 --> 0:11:45.760 I also realized that there are you know, with all 0:11:45.800 --> 0:11:49.640 the good things that can do, there are negative consequences 0:11:49.720 --> 0:11:54.079 to it, right, and so I put that negative consequences 0:11:54.240 --> 0:11:59.199 in the under the bucket of untrusted untrustworthiness. Ethics is 0:11:59.240 --> 0:12:01.640 a big composed under of it, right, Whether the AI 0:12:01.800 --> 0:12:05.800 is fair or biased or transparent explainable, but also things 0:12:05.840 --> 0:12:09.480 like is it compliant with local regulations, does it have 0:12:09.679 --> 0:12:12.800 controls in place? Does it have governance in place to 0:12:12.920 --> 0:12:17.160 continuously monitor for it going wrongue because you know, Jonathan, 0:12:17.200 --> 0:12:20.480 today AI is mostly machine learning, so that it's learning 0:12:20.480 --> 0:12:23.400 and evolving. It's not that era when we developed code 0:12:23.440 --> 0:12:25.800 put it out there and the code states static and 0:12:25.840 --> 0:12:30.280 its behavior was very predictable. With AI, the outputs can 0:12:30.360 --> 0:12:33.360 change depending on inputs your feed and it's impossible to 0:12:33.480 --> 0:12:37.760 trade on all possible inputs. So trustworthy for me is 0:12:37.920 --> 0:12:42.079 you know, really, or is when you have addressed, when 0:12:42.120 --> 0:12:45.640 you have thought about and addressed all the possible negative 0:12:45.760 --> 0:12:49.280 things that this AI solution can cause. Well, I would 0:12:49.320 --> 0:12:51.320 love to kind of dive into a little bit more 0:12:51.360 --> 0:12:53.760 of that because one of the things that you said 0:12:53.760 --> 0:12:57.280 that really resonated with me was the idea of transparency, 0:12:57.679 --> 0:13:01.480 because I have covered this past episodes of tech stuff, 0:13:01.520 --> 0:13:05.160 but the sort of the black box problem of creating 0:13:05.200 --> 0:13:09.280 a system, for example, a machine learning system, and you 0:13:09.360 --> 0:13:13.719 have this this machine that's training itself over and over 0:13:13.760 --> 0:13:16.200 and over. Maybe it's adversarial training, maybe you actually have 0:13:16.240 --> 0:13:20.120 two systems that are set against each other and you're training, 0:13:20.920 --> 0:13:25.320 and the issues that can arise if you have distanced 0:13:25.360 --> 0:13:28.959 yourself so far from what the machine is doing that 0:13:29.120 --> 0:13:33.120 you are unable to determine the process by which it 0:13:33.760 --> 0:13:37.199 arrives at its conclusions. And that to me is one 0:13:37.200 --> 0:13:41.120 of those those pitfalls. Yes, but I would also challenge 0:13:41.120 --> 0:13:44.920 it a little bit, Jonathan, because that a whole synthesis 0:13:44.960 --> 0:13:48.280 of my book is that it depends on the use case. 0:13:48.760 --> 0:13:51.880 It is not a one size fit soul. So depending 0:13:52.000 --> 0:13:55.920 on where and to solve what problem are you using 0:13:55.920 --> 0:13:59.840 that DAIR solution, it is for that organization that teams 0:14:00.080 --> 0:14:04.160 decide if transparency is crucial. Right. If if your AI 0:14:04.280 --> 0:14:09.800 solution is being used to for patient care in a 0:14:09.920 --> 0:14:15.840 hospital system, then transparency is absolutely crucial, right. But if 0:14:15.880 --> 0:14:19.680 you are using the AI solution to predict when an 0:14:19.800 --> 0:14:23.040 X ray machine might fail, and you're able to predict 0:14:23.040 --> 0:14:26.680 at accuracy rate that this machine is going to fail 0:14:26.680 --> 0:14:28.960 in the next forty eight hours of call a technician, 0:14:29.760 --> 0:14:33.160 transparency may not be as crucial, right. So I think 0:14:33.920 --> 0:14:37.360 transparency is crucial depending on the use case. And that's 0:14:37.360 --> 0:14:40.440 true for all the other dimensions as well, even fairness 0:14:40.440 --> 0:14:42.840 and bias, which we hear a lot about. So it 0:14:43.000 --> 0:14:46.080 really depends on the use case that that you're using 0:14:46.120 --> 0:14:49.760 the I fall Hey there, Jonathan back at the home studio, 0:14:49.920 --> 0:14:52.560 just here to say we are going to take a 0:14:52.640 --> 0:14:55.520 quick break, but we'll be back with more with Bena Amanath, 0:14:55.640 --> 0:15:08.480 the executive director of Deloitte AI Institute. Bias doesn't necessarily 0:15:08.560 --> 0:15:11.800 mean negative, depending upon the use case of the technology. 0:15:12.640 --> 0:15:15.000 In some cases you need to have a biased system 0:15:15.040 --> 0:15:18.440 because it's specifically meant to be weighted to do one 0:15:18.480 --> 0:15:21.440 thing versus another, and without the bias it doesn't do that. 0:15:22.400 --> 0:15:25.360 But the way we typically hear about bias is when 0:15:25.400 --> 0:15:28.840 it is making a negative impact, when it's something for 0:15:28.960 --> 0:15:32.720 it's like like the facial recognition technologies. We've heard plenty 0:15:32.800 --> 0:15:36.920 about that. So it is interesting to me, And uh, 0:15:37.080 --> 0:15:40.160 I'm curious, like, what are what are some of the 0:15:40.320 --> 0:15:44.560 uses of AI you're seeing in technology now that you 0:15:44.600 --> 0:15:49.560 find really exciting. Yeah, no, I think you know, we're 0:15:49.600 --> 0:15:54.120 still very early on in this technology evolution and there 0:15:54.160 --> 0:15:56.480 are still so many use cases to be solved, so 0:15:56.520 --> 0:16:00.360 many industries to take a I too right to point 0:16:00.400 --> 0:16:04.520 about bias and its relevance. I completely agree with you that, 0:16:04.600 --> 0:16:06.600 you know, it depends on the use case, and it 0:16:06.680 --> 0:16:09.560 goes back to that first question we talked about, right, 0:16:09.640 --> 0:16:14.280 how AI is really emulating human intelligence, which means that 0:16:14.440 --> 0:16:16.680 it is going to carry over the biases of the 0:16:16.800 --> 0:16:20.640 humans that are building it. Right, But as as a 0:16:20.680 --> 0:16:24.320 business or as an organization who's looking to use an 0:16:24.320 --> 0:16:27.520 air solution, who's looking to develop an a solution, they 0:16:27.560 --> 0:16:30.920 really have to, you know, bring together the stakeholders to 0:16:31.240 --> 0:16:36.640 discuss and decide how crucial is fairness or unbiased ness 0:16:36.680 --> 0:16:41.200 important in this particular AI use case. And easy one 0:16:41.240 --> 0:16:44.080 out is if it doesn't involve human data, then you 0:16:44.200 --> 0:16:46.960 probably don't have to worry as biased as a factor 0:16:47.040 --> 0:16:50.360 and address it. And if it does involve human data, 0:16:50.760 --> 0:16:53.800 then again there is weightage in what right if there 0:16:53.920 --> 0:16:59.200 is biased at in an algorithm that is providing personalized marketing, 0:17:00.120 --> 0:17:02.160 that you know that there is a weight to it. 0:17:02.320 --> 0:17:04.160 And if it is if there is biased in an 0:17:04.200 --> 0:17:09.359 algorithm that is supporting law enforcement decisions, that's a higher rate, right. 0:17:09.440 --> 0:17:12.560 And it's really about rating it, you know, weighing it 0:17:12.680 --> 0:17:16.800 and deciding which ones are the one where biases acceptable 0:17:16.880 --> 0:17:19.639 and you can still proceed and get value from the 0:17:19.680 --> 0:17:22.280 AI solution, and which are the ones where it is 0:17:22.359 --> 0:17:26.359 absolutely not acceptable and you need to stop and figure 0:17:26.400 --> 0:17:29.879 out and alternate way to solve for that problem. It's 0:17:29.920 --> 0:17:32.600 fascinating because it to me this is starting to sound 0:17:32.880 --> 0:17:35.879 and I agree with you, like the machines we build 0:17:36.440 --> 0:17:39.760 are in large part reflections upon ourselves, especially when we're 0:17:39.800 --> 0:17:44.000 talking about coding and software. I mean obviously that's going 0:17:44.080 --> 0:17:47.040 that's a creative process. I don't know that everybody views 0:17:47.080 --> 0:17:49.560 it that way, but I think of it very similar 0:17:49.600 --> 0:17:53.200 to creating any kind of creative work. It's a reflection 0:17:53.680 --> 0:17:56.600 of your process and your you know, the things that 0:17:56.640 --> 0:18:00.399 are important to you, the things you've prioritized. And it 0:18:00.440 --> 0:18:02.920 makes me think of how we're in an era now 0:18:03.000 --> 0:18:06.240 where I'm getting a little in the weeds here, but 0:18:06.240 --> 0:18:09.360 we're in an era where we're more likely to address 0:18:09.440 --> 0:18:12.440 things like, uh, mental health and the fact that we 0:18:12.520 --> 0:18:15.320 need to be mindful and we need to improve ourselves. 0:18:15.359 --> 0:18:19.119 And it's almost like taking that same approach, but applying 0:18:19.280 --> 0:18:22.280 that sort of thinking to designing a system so that 0:18:22.680 --> 0:18:25.560 we are being mindful to create the best system for 0:18:25.680 --> 0:18:31.400 whatever purpose it is it's intended to address. Yeah, you've 0:18:31.400 --> 0:18:34.239 got it exactly right. The way I think about it is, 0:18:34.280 --> 0:18:39.320 how can we reduce the unintended consequences? Right? We know 0:18:39.480 --> 0:18:42.240 there are going to be risk associated with it, How 0:18:42.280 --> 0:18:45.280 are we going to have a discussion prior to putting 0:18:45.280 --> 0:18:48.199 the solution out into the world and then you know, 0:18:48.280 --> 0:18:51.120 see all the negative impacts. Can we have a proactive 0:18:51.119 --> 0:18:54.680 discussion as part of your project planning meeting or your 0:18:54.760 --> 0:18:58.760 design meeting right to proactively identify what are the ways 0:18:58.760 --> 0:19:01.080 this could go wrong and fix it. Johnathan, you know, 0:19:01.160 --> 0:19:04.160 the easiest example that I can give is we're living 0:19:04.160 --> 0:19:08.199 in this very interesting era where you know, AI as 0:19:08.240 --> 0:19:11.960 a core technology is developing and you know, there are 0:19:11.960 --> 0:19:14.439 all these the value that you're getting from it, and 0:19:14.440 --> 0:19:17.600 then there are all these negative things that can happen. 0:19:17.680 --> 0:19:20.080 So think about you know, way back when when you 0:19:20.119 --> 0:19:23.440 know the cars were first invented, right, we didn't even 0:19:23.520 --> 0:19:26.040 have proper roads. We didn't have seed belts, we didn't 0:19:26.080 --> 0:19:29.080 have speed limits, right, and be in that phase where 0:19:29.119 --> 0:19:31.679 there are cars running on the road. They're taking us 0:19:31.680 --> 0:19:34.280 from point to point be faster, so we want to 0:19:34.400 --> 0:19:37.280 use it, but we don't have the seat belts put 0:19:37.320 --> 0:19:40.159 in place, we don't have the speed limits set in place, 0:19:40.200 --> 0:19:43.240 so you're going to see accidents. But we are humans. 0:19:43.359 --> 0:19:45.000 We're going to learn from it and we're going to 0:19:45.080 --> 0:19:47.000 come up with those speed limits. We're going to figure 0:19:47.000 --> 0:19:49.480 out what are those card rails, and it is going 0:19:49.640 --> 0:19:52.600 you know, we are going to you know, achieve a 0:19:52.680 --> 0:19:56.040 point where you know, we have those guard rails in 0:19:56.040 --> 0:19:59.120 place so that you can run with AI faster. It's 0:19:59.160 --> 0:20:02.720 just that this interim phase is when you know, we 0:20:02.840 --> 0:20:06.240 have to figure it out out in tandem while it's 0:20:06.520 --> 0:20:10.560 running in the real world, causing accidents. And in some 0:20:10.640 --> 0:20:13.919 cases that's that those accidents can be things where you 0:20:14.000 --> 0:20:16.560 have it maybe in a test environment and you think, oh, 0:20:16.640 --> 0:20:19.240 this isn't behaving the way I thought it was. But 0:20:19.480 --> 0:20:21.760 you know, thank goodness, it hasn't been deployed out in 0:20:21.800 --> 0:20:26.240 the real world for or within your company's UH processes, 0:20:26.240 --> 0:20:28.479 so you think, oh, well, it didn't wipe out all 0:20:28.480 --> 0:20:31.879 of our revenue because it's in a test environment. UH. 0:20:31.920 --> 0:20:34.440 And in other cases, I see I see some companies. 0:20:34.480 --> 0:20:36.600 I'm not gonna name names, Bina, I'm not gonna put 0:20:36.640 --> 0:20:39.760 anyone on blast here, but I have seen some companies 0:20:40.200 --> 0:20:43.439 that have taken that kind of idea and applied it 0:20:43.560 --> 0:20:48.080 in UH specific deployments of technology where there can have 0:20:48.160 --> 0:20:52.440 some some real world negative consequences to end users. UM. 0:20:52.760 --> 0:20:57.080 And that to me has always a concern I find, yeah, 0:20:57.200 --> 0:21:00.280 I find that I find it hits me wrong. Yes, 0:21:00.400 --> 0:21:04.040 And that's the reality of how we've evolved as a 0:21:04.160 --> 0:21:07.359 technology in their technology space. It's a bunch of you know, 0:21:07.440 --> 0:21:12.200 technologists coming together and building these cool, new shiny technologists. Look. 0:21:12.680 --> 0:21:14.879 You know, as I said, I am a technologist in 0:21:14.960 --> 0:21:18.159 my DNA my training, and it's very easy to just 0:21:18.200 --> 0:21:20.760 focus on all the good things that can do. But 0:21:20.920 --> 0:21:24.800 with AI, now that realization has hit, you need other 0:21:25.760 --> 0:21:27.720 you know, skill sets at the table, whether it is 0:21:27.720 --> 0:21:32.320 a social site, is philosophers, legal and compliance to help 0:21:32.440 --> 0:21:35.199 us figure out those seedbells and the you know, the 0:21:35.240 --> 0:21:39.439 speed the lanes, you know, because technologies by themselves cannot 0:21:39.520 --> 0:21:42.120 do it. So you'll see more of the discussions coming 0:21:42.160 --> 0:21:44.720 around ethics and which is resulting in new roles and 0:21:44.800 --> 0:21:50.200 new jobs, which becomes core and part of your engineering process. Right, 0:21:50.240 --> 0:21:53.600 So that scope of who is involved in designing and 0:21:53.640 --> 0:21:57.920 developing AI is definitely increasing. And the other big part, 0:21:58.080 --> 0:22:00.399 you know, and this has been a challenge since I 0:22:00.440 --> 0:22:02.639 started in tech. You know, there's a lack of diversity 0:22:02.680 --> 0:22:06.800 in tech. It's a reality, right, But unfortunately, because AI 0:22:06.880 --> 0:22:09.760 is so closely tied to human intelligence, if you don't 0:22:09.800 --> 0:22:13.679 have enough diversity from you know, not only from a gender, 0:22:13.800 --> 0:22:17.440 race at necessity perspective, but even a diversity of thought, right, 0:22:17.880 --> 0:22:21.119 you're the AI solution you built is not going to 0:22:21.200 --> 0:22:24.480 be as robust as it could be if you had 0:22:24.520 --> 0:22:27.560 a diverse team at the table. Right, you've probably heard 0:22:27.600 --> 0:22:31.480 of that classic example of you know, the robotic vacuums, right, 0:22:31.560 --> 0:22:33.720 how it was designed and now it was built out. 0:22:34.200 --> 0:22:37.200 And then in the Eastern cultures it's normal to sleep 0:22:37.280 --> 0:22:41.480 on the floor and it sucked up human somebody who's 0:22:41.480 --> 0:22:44.040 sleeping their hair because it was never trade on it. 0:22:44.040 --> 0:22:46.879 It didn't come you know, it didn't come to the discussion, 0:22:47.200 --> 0:22:49.880 and it was being designed because nobody was there from 0:22:49.920 --> 0:22:53.879 that culture. Right. So I think, you know, the realization 0:22:53.920 --> 0:22:56.080 that you need more diversity at the table, you need 0:22:56.119 --> 0:22:59.080 more controls in place. It's all coming to the forefront. 0:22:59.160 --> 0:23:03.000 I definitely see companies addressing it. But the the DNA 0:23:03.119 --> 0:23:05.200 will now has been oh, look at all the cool 0:23:05.240 --> 0:23:08.200 things this technology can do, let's go put it out right. 0:23:08.480 --> 0:23:11.520 But I do think, you know, companies are getting mindful 0:23:11.560 --> 0:23:17.400 about it and hopefully we'll reduce the number of unintended consequences. Yeah. 0:23:17.480 --> 0:23:22.280 I see the same thing reflected in the open source community, 0:23:22.440 --> 0:23:27.160 where you have an open source approach to developing software, 0:23:27.800 --> 0:23:32.760 and because it's open and and anyone interested and capable 0:23:33.320 --> 0:23:40.000 can contribute ideas get tested, very quickly. New new perspectives 0:23:40.000 --> 0:23:43.600 get incorporated very quickly. Things that are working stick around, 0:23:43.640 --> 0:23:47.520 things that don't work get improved. And the way I've 0:23:47.520 --> 0:23:50.840 described it to other people is, if you have a 0:23:50.880 --> 0:23:54.440 closed off garden that you're working on, you're only as 0:23:54.480 --> 0:23:57.920 good as the smart people who happen to work for you. 0:23:58.400 --> 0:24:00.439 And if you go with this other approach where you 0:24:00.680 --> 0:24:04.439 purposefully open it up, which is like the biggest version 0:24:04.480 --> 0:24:06.959 of let's let's try and get as much diversity of 0:24:07.000 --> 0:24:10.600 thought in here as possible. Uh, you don't have that 0:24:10.680 --> 0:24:13.520 limitation because you've You've just said, well, now the world 0:24:13.760 --> 0:24:16.199 is I mean it's not the whole world, but but 0:24:16.280 --> 0:24:19.800 effectively the world. The world can contribute if if they, 0:24:19.880 --> 0:24:24.320 if they wish, and uh agree, I think having that 0:24:24.400 --> 0:24:29.560 diversity is absolutely key to creating solutions that work for 0:24:29.640 --> 0:24:34.320 as many people and as many potential uses of that 0:24:34.359 --> 0:24:39.640 technology as possible. I being a a a white man 0:24:39.840 --> 0:24:44.600 in the United States, I am I am essentially the 0:24:44.760 --> 0:24:48.120 catered to audience for a lot of tech, and so 0:24:48.800 --> 0:24:52.440 I've seen how things that were made to work really 0:24:52.440 --> 0:24:55.480 well for me do not work for some other people. 0:24:55.520 --> 0:24:58.639 And that's such a tiny little microcosm when we're looking 0:24:58.680 --> 0:25:01.320