1 00:00:12,609 --> 00:00:14,810 Speaker 1: Hello, my name is Graham Class and I'm your host 2 00:00:14,819 --> 00:00:18,099 Speaker 1: for this season of technically speaking, an Intel podcast. While 3 00:00:18,110 --> 00:00:20,940 Speaker 1: Intel is at the forefront of so many cutting edge technologies. 4 00:00:21,000 --> 00:00:23,950 Speaker 1: This season is all about artificial intelligence and that's why 5 00:00:23,959 --> 00:00:27,209 Speaker 1: I've been tapped as your host. Having a background in tech, 6 00:00:27,219 --> 00:00:29,649 Speaker 1: as a software engineer. I was always interested in merging 7 00:00:29,659 --> 00:00:32,810 Speaker 1: the advances of artificial intelligence with my love for media 8 00:00:33,209 --> 00:00:36,259 Speaker 1: is culminated in one of my other projects, Daily Dad jokes, 9 00:00:36,270 --> 00:00:38,949 Speaker 1: an A I powered podcast, churning out jokes and humor 10 00:00:38,959 --> 00:00:40,150 Speaker 1: for listeners worldwide. 11 00:00:40,750 --> 00:00:43,098 Speaker 1: But artificial Intelligence can do a lot more than help 12 00:00:43,110 --> 00:00:46,720 Speaker 1: whip up a corny joke. This technology has been revolutionizing 13 00:00:46,729 --> 00:00:51,049 Speaker 1: the way we engage with the world with innovations across healthcare, agriculture, 14 00:00:51,060 --> 00:00:53,130 Speaker 1: business and even the public sector. 15 00:00:53,750 --> 00:00:56,330 Speaker 1: Another way that artificial intelligence is changing the world is 16 00:00:56,340 --> 00:00:59,689 Speaker 1: through philosophy. The term ethical A I is a framework 17 00:00:59,700 --> 00:01:02,180 Speaker 1: on how to use A I, what system should be 18 00:01:02,189 --> 00:01:04,930 Speaker 1: in place to govern its use with business and consumers. 19 00:01:05,819 --> 00:01:08,220 Speaker 1: In this episode, we'll dive into the ethics of artificial 20 00:01:08,230 --> 00:01:10,760 Speaker 1: intelligence with one of the pioneers in the field. 21 00:01:12,029 --> 00:01:16,349 Speaker 1: Joining me for today's conversation is Intel's Ria Chu R 22 00:01:16,489 --> 00:01:19,149 Speaker 1: A can perhaps be described as the moral compass of 23 00:01:19,160 --> 00:01:22,209 Speaker 1: the company's A I as an A I software architect 24 00:01:22,220 --> 00:01:25,080 Speaker 1: and generative A I evangelist. She is charged with finding 25 00:01:25,089 --> 00:01:29,360 Speaker 1: responsible trustworthy solutions for Intel's Internet of Things Engineering group. 26 00:01:29,949 --> 00:01:33,300 Speaker 1: Her role exists at the intersection of hardware and software 27 00:01:33,389 --> 00:01:35,769 Speaker 1: product design and effective consumer use. 28 00:01:36,750 --> 00:01:39,830 Speaker 1: Having studied extensively at Harvard in the subjects of computer 29 00:01:39,839 --> 00:01:43,069 Speaker 1: science and data science. The domains of expertise are solutions 30 00:01:43,080 --> 00:01:47,010 Speaker 1: for security and privacy in machine learning, fairness, explainable and 31 00:01:47,019 --> 00:01:51,769 Speaker 1: responsible A I systems uncertain A I reinforcement learning and 32 00:01:51,779 --> 00:01:56,349 Speaker 1: computational models of intelligence. She is a reoccurring keynote speaker 33 00:01:56,360 --> 00:01:59,580 Speaker 1: on issues in data science and responsible A I. We 34 00:01:59,589 --> 00:02:01,610 Speaker 1: are very excited to have her on the podcast to 35 00:02:01,620 --> 00:02:04,250 Speaker 1: share her expertise on Intel's ethics in their A I 36 00:02:04,260 --> 00:02:04,769 Speaker 1: development 37 00:02:08,809 --> 00:02:10,789 Speaker 1: Ria. Welcome to the show. Thank you, 38 00:02:10,800 --> 00:02:12,190 Speaker 2: Graham. It's awesome to be here. 39 00:02:12,669 --> 00:02:15,990 Speaker 1: I've had a look at your bio and would like 40 00:02:16,000 --> 00:02:18,309 Speaker 1: to know how did you come about to join the 41 00:02:18,320 --> 00:02:19,329 Speaker 1: Intel family? 42 00:02:19,508 --> 00:02:23,580 Speaker 2: Sure, I joined Intel in 2018 when I was 14 43 00:02:23,589 --> 00:02:26,380 Speaker 2: years old as an intern. I had, yes, I had 44 00:02:26,389 --> 00:02:29,228 Speaker 2: an amazing mentor who went through all of the legal 45 00:02:29,240 --> 00:02:32,448 Speaker 2: pages and the review needed to get me to that position. 46 00:02:32,460 --> 00:02:35,639 Speaker 2: So initially, I interviewed with three teams on three different 47 00:02:35,649 --> 00:02:37,539 Speaker 2: areas in the A I space. One of them was 48 00:02:37,550 --> 00:02:39,699 Speaker 2: around A I and healthcare, very theoretic 49 00:02:40,050 --> 00:02:43,419 Speaker 2: and mathematical implications and path finding the other two were 50 00:02:43,429 --> 00:02:46,860 Speaker 2: on software development and profiling. And the next was on 51 00:02:46,869 --> 00:02:50,478 Speaker 2: deep learning optimization specifically. So I did have the opportunity 52 00:02:50,490 --> 00:02:53,800 Speaker 2: to pick the one on optimization for deep learning for hardware. 53 00:02:53,869 --> 00:02:55,580 Speaker 2: And that is how I started off my journey at 54 00:02:55,589 --> 00:02:59,258 Speaker 2: Intel and got introduced to it. The interplay between hardware 55 00:02:59,270 --> 00:03:02,339 Speaker 2: and software is something that always drew my attention. So 56 00:03:02,350 --> 00:03:04,259 Speaker 2: when I was able to work on that as part 57 00:03:04,270 --> 00:03:06,758 Speaker 2: of my first role as an intern, I was really excited. 58 00:03:07,059 --> 00:03:10,910 Speaker 1: OK, great. So now uh I understand that you're a 59 00:03:10,919 --> 00:03:12,520 Speaker 1: software A I architect. 60 00:03:13,050 --> 00:03:15,880 Speaker 1: Can you just give an overview of what that entails 61 00:03:16,050 --> 00:03:18,570 Speaker 2: as a software architect? Today? I have a couple of 62 00:03:18,580 --> 00:03:21,910 Speaker 2: roles and responsibilities corresponding to the latest and greatest, which 63 00:03:21,919 --> 00:03:24,029 Speaker 2: is very exciting to me in my day to day. 64 00:03:24,119 --> 00:03:26,799 Speaker 2: The first is generative A I. So looking at and 65 00:03:26,809 --> 00:03:30,210 Speaker 2: taking into account the different software optimizations that we're planning 66 00:03:30,220 --> 00:03:33,259 Speaker 2: for generative A I, how the workloads are shaping changes 67 00:03:33,270 --> 00:03:35,839 Speaker 2: in the algorithms over time as well as also the 68 00:03:35,850 --> 00:03:38,759 Speaker 2: associated mechanisms that we see that are 69 00:03:38,854 --> 00:03:41,395 Speaker 2: in touch with them as an evangelist. I also get 70 00:03:41,404 --> 00:03:45,035 Speaker 2: to work on top of my software architect role as 71 00:03:45,044 --> 00:03:48,195 Speaker 2: a marketer and an advocate for these technologies. So creating 72 00:03:48,205 --> 00:03:51,744 Speaker 2: very short demos and tutorials for users to quickly grasp 73 00:03:51,755 --> 00:03:53,895 Speaker 2: what exactly is going on with this model. How can 74 00:03:53,904 --> 00:03:55,434 Speaker 2: I use it in my day to day? How can 75 00:03:55,445 --> 00:03:57,514 Speaker 2: I put it to my use case. So a lot 76 00:03:57,524 --> 00:03:59,475 Speaker 2: of the focus today for me is on gender to 77 00:03:59,514 --> 00:04:01,875 Speaker 2: A I, I also look into ethical and explainable A 78 00:04:01,884 --> 00:04:04,574 Speaker 2: I tools and technologies as part of my path finding. 79 00:04:05,520 --> 00:04:08,380 Speaker 1: Yeah, I've been using generative A I apps to do research, 80 00:04:08,389 --> 00:04:12,899 Speaker 1: creating podcast, artwork and experimented with creating music. So this 81 00:04:12,910 --> 00:04:17,659 Speaker 1: leads me into asking you what's your definition of artificial intelligence? 82 00:04:18,040 --> 00:04:22,390 Speaker 1: And maybe some examples of where we're seeing it as 83 00:04:22,399 --> 00:04:24,660 Speaker 1: a central topic in the tech world. 84 00:04:24,899 --> 00:04:27,469 Speaker 2: The way that I like to define it is something 85 00:04:27,480 --> 00:04:31,000 Speaker 2: I copied over actually from our recent regulations on A 86 00:04:31,010 --> 00:04:35,219 Speaker 2: I around how A I models are agents or systems 87 00:04:35,230 --> 00:04:38,159 Speaker 2: that are capable of consuming and producing data in an 88 00:04:38,170 --> 00:04:41,529 Speaker 2: environment and also taking actions that can in turn influence 89 00:04:41,540 --> 00:04:45,979 Speaker 2: our decisions. There's a lot of use cases for them everywhere, healthcare, retail, 90 00:04:45,988 --> 00:04:46,559 Speaker 2: et cetera. 91 00:04:47,200 --> 00:04:49,950 Speaker 1: Yeah. When I talked with uh people even in the 92 00:04:49,959 --> 00:04:53,789 Speaker 1: tech world, there's a lot of confusion around OK, you've 93 00:04:53,799 --> 00:04:56,678 Speaker 1: got algorithms, you've got A I, you've got machine learning 94 00:04:56,690 --> 00:04:58,880 Speaker 1: perhaps if you could start with maybe some of the 95 00:04:58,890 --> 00:05:02,719 Speaker 1: difference between algorithms versus say A I. What do you 96 00:05:02,730 --> 00:05:04,329 Speaker 1: see as the difference between the two 97 00:05:04,829 --> 00:05:08,239 Speaker 2: typical algorithms I'd say are based off of certain schemes 98 00:05:08,250 --> 00:05:10,600 Speaker 2: that we're already aware of with machine learning. You have 99 00:05:10,609 --> 00:05:13,269 Speaker 2: these new paradigms that are coming in and completely spinning 100 00:05:13,279 --> 00:05:17,488 Speaker 2: the narrative, things like continual learning, very large models, different 101 00:05:17,500 --> 00:05:20,279 Speaker 2: types of state machines altogether, depending on the application you 102 00:05:20,290 --> 00:05:23,299 Speaker 2: integrate it into. So I would say there are some 103 00:05:23,309 --> 00:05:26,630 Speaker 2: fundamental differences that are coming in between algorithms and machine 104 00:05:26,640 --> 00:05:28,470 Speaker 2: learning models on that front when it comes to use 105 00:05:28,480 --> 00:05:31,660 Speaker 2: cases application and of course implementation as well. 106 00:05:32,049 --> 00:05:35,480 Speaker 1: And where I see the power is sort of combining 107 00:05:35,839 --> 00:05:39,709 Speaker 1: the traditional sort of if then else algorithms uh with 108 00:05:39,720 --> 00:05:42,649 Speaker 1: A I. And I'm just wondering if you've seen any 109 00:05:42,660 --> 00:05:46,910 Speaker 1: sort of practical applications merging of all these techniques. 110 00:05:48,260 --> 00:05:50,910 Speaker 2: Yes. And I'm very interested in composite A I. It's 111 00:05:50,920 --> 00:05:53,299 Speaker 2: something that I'm getting to work on a lot more 112 00:05:53,369 --> 00:05:55,179 Speaker 2: in my day to day. And something that we're actually 113 00:05:55,190 --> 00:05:58,190 Speaker 2: doing a demo for at intel innovation where we are 114 00:05:58,200 --> 00:06:01,239 Speaker 2: chaining multiple large language models together. The way I see 115 00:06:01,250 --> 00:06:04,320 Speaker 2: composite A I is being able to tie together multiple 116 00:06:04,329 --> 00:06:05,640 Speaker 2: models as part of a 117 00:06:05,755 --> 00:06:09,255 Speaker 2: interface or an application with chaining models. I see it 118 00:06:09,265 --> 00:06:11,515 Speaker 2: as a subset of composite A I where you have 119 00:06:11,524 --> 00:06:14,184 Speaker 2: models that are linked to each other and have dependencies 120 00:06:14,195 --> 00:06:16,915 Speaker 2: on their inputs and outputs. It can be sometimes a 121 00:06:16,924 --> 00:06:20,294 Speaker 2: nightmare to get the dependencies altogether because you have cascading 122 00:06:20,303 --> 00:06:23,334 Speaker 2: models one after the other dependent on each is output, 123 00:06:23,579 --> 00:06:25,260 Speaker 2: but it is possible and it does give you a 124 00:06:25,269 --> 00:06:28,219 Speaker 2: lot of applications and opens up the possibilities where you 125 00:06:28,230 --> 00:06:31,159 Speaker 2: can get to a very nice user interface that users 126 00:06:31,170 --> 00:06:34,359 Speaker 2: can interact with. Developers can build upon businesses and other 127 00:06:34,369 --> 00:06:37,649 Speaker 2: communities can just leverage and adopt that is giving you 128 00:06:37,660 --> 00:06:40,320 Speaker 2: a lot of capabilities at once with ease of deployment. 129 00:06:40,339 --> 00:06:40,350 Speaker 1: Oh, 130 00:06:40,510 --> 00:06:43,719 Speaker 1: that's good. Now, turning to the ethics side of it, 131 00:06:43,730 --> 00:06:46,470 Speaker 1: which you've done quite a lot of thinking and work 132 00:06:46,480 --> 00:06:50,029 Speaker 1: in how would you define ethics in A I 133 00:06:50,970 --> 00:06:53,390 Speaker 2: with ethical A I, the definition that I like to 134 00:06:53,399 --> 00:06:56,469 Speaker 2: adopt is socio technical development of A I systems and 135 00:06:56,480 --> 00:06:59,779 Speaker 2: that involves societal and technical aspects, but really focusing on 136 00:06:59,790 --> 00:07:01,709 Speaker 2: the implications and the intentions with these 137 00:07:01,720 --> 00:07:04,609 Speaker 1: algorithms. In terms of when you're talking with your peers 138 00:07:04,619 --> 00:07:07,589 Speaker 1: and colleagues, it has been a lot of discussion and 139 00:07:07,600 --> 00:07:10,989 Speaker 1: talk about trying to have a uniform ethical framework that 140 00:07:11,549 --> 00:07:13,899 Speaker 1: at least gives a common language into, you know, when 141 00:07:13,910 --> 00:07:16,200 Speaker 1: you're discussing these sorts of things related to ethics in 142 00:07:16,209 --> 00:07:16,380 Speaker 1: A I, 143 00:07:17,700 --> 00:07:20,570 Speaker 2: there are common frameworks that are in place. Most of 144 00:07:20,579 --> 00:07:23,679 Speaker 2: them are centered around implications and intention and how we 145 00:07:23,690 --> 00:07:27,040 Speaker 2: structure that around certain technologies. Right now. It's very popular 146 00:07:27,049 --> 00:07:30,049 Speaker 2: for applications, generative A I where we see these frameworks 147 00:07:30,059 --> 00:07:32,600 Speaker 2: being put into place around, let's look at the inputs, 148 00:07:32,609 --> 00:07:35,730 Speaker 2: the outputs and then the overall modeler framework. And this 149 00:07:35,739 --> 00:07:36,000 Speaker 2: may 150 00:07:36,105 --> 00:07:38,924 Speaker 2: simplistic, but it really is boiled down to these very 151 00:07:38,934 --> 00:07:41,755 Speaker 2: simple elements. Similarly for other A I domains that are 152 00:07:41,765 --> 00:07:44,524 Speaker 2: outside of generative A I like object detection, it's very 153 00:07:44,535 --> 00:07:47,375 Speaker 2: much focused on what is the particular use case? For example, 154 00:07:47,385 --> 00:07:49,964 Speaker 2: is it something that is of high risk like health 155 00:07:49,975 --> 00:07:52,674 Speaker 2: care applications or surveillance or is it something that's a 156 00:07:52,684 --> 00:07:54,644 Speaker 2: bit lower risk like content creation 157 00:07:54,859 --> 00:07:57,399 Speaker 2: and then seeing how exactly our user experience and our 158 00:07:57,410 --> 00:08:00,359 Speaker 2: development of those models is echoing ethical A I principles. 159 00:08:00,369 --> 00:08:03,619 Speaker 2: So I would say like to summarize, there are different 160 00:08:03,630 --> 00:08:06,440 Speaker 2: frameworks and summaries that we apply. But of course, the 161 00:08:06,450 --> 00:08:08,920 Speaker 2: templates need to be flexible when we're talking about ethical 162 00:08:08,929 --> 00:08:10,660 Speaker 2: A I for these new A I models. 163 00:08:11,119 --> 00:08:14,359 Speaker 1: how do you go about ensuring that your staff and 164 00:08:14,369 --> 00:08:16,850 Speaker 1: your engineers and your product managers 165 00:08:17,510 --> 00:08:22,000 Speaker 1: actually embed that ethical framework into its A I development. 166 00:08:22,290 --> 00:08:25,209 Speaker 2: Sure, it's such a challenging problem even to describe as 167 00:08:25,220 --> 00:08:28,290 Speaker 2: well as you're mentioning it, you know, there's so many 168 00:08:28,299 --> 00:08:30,570 Speaker 2: different things that you can actively do, right? Like as 169 00:08:30,579 --> 00:08:34,489 Speaker 2: you mentioned, policies, assessments, et cetera. So at Intel, we 170 00:08:34,500 --> 00:08:37,809 Speaker 2: take a multiple approaches towards it. The one thing that 171 00:08:37,820 --> 00:08:42,010 Speaker 2: we very heavily emphasize on is internal governance. And Laman Nachman, 172 00:08:42,020 --> 00:08:44,449 Speaker 2: who's my mentor and also leading the responsibly 173 00:08:44,585 --> 00:08:48,726 Speaker 2: efforts at Intel very neatly and concisely describes them as 174 00:08:48,736 --> 00:08:52,074 Speaker 2: guardrails that we have internally in place. And these are 175 00:08:52,085 --> 00:08:56,166 Speaker 2: really guidelines that are designed to help our developers, engineers, managers, 176 00:08:56,255 --> 00:09:00,476 Speaker 2: and our communities and marketers, et cetera. Understand the implications 177 00:09:00,486 --> 00:09:02,676 Speaker 2: again of what exactly are we producing in terms of 178 00:09:02,684 --> 00:09:05,334 Speaker 2: the content? What are some technical solutions that we can 179 00:09:05,346 --> 00:09:09,005 Speaker 2: instill mid pipeline or early on before starting the effort 180 00:09:09,015 --> 00:09:11,406 Speaker 2: when we're getting started with A I development efforts 181 00:09:11,660 --> 00:09:13,562 Speaker 2: and I would say that that's the core process that 182 00:09:13,572 --> 00:09:17,261 Speaker 2: we focus on. We're also very heavily invested in technological development, 183 00:09:17,271 --> 00:09:19,742 Speaker 2: whether that's through the deep fake detection work that L 184 00:09:19,861 --> 00:09:23,221 Speaker 2: Deir and team are taking on um explainable A I tools, 185 00:09:23,231 --> 00:09:25,351 Speaker 2: et cetera. So really trying to approach this from a 186 00:09:25,361 --> 00:09:28,992 Speaker 2: governance perspective internally, from a tooling perspective, what we can 187 00:09:29,002 --> 00:09:32,242 Speaker 2: provide to the developer community and our customers and to 188 00:09:32,252 --> 00:09:35,562 Speaker 2: partners and from a third perspective, regulations, how do we 189 00:09:35,572 --> 00:09:38,642 Speaker 2: influence the industry at large and help contribute to discussions 190 00:09:39,390 --> 00:09:40,189 Speaker 1: that's really good. And 191 00:09:40,840 --> 00:09:43,539 Speaker 1: you mentioned the work of Lama Nachman and we're actually 192 00:09:43,549 --> 00:09:45,539 Speaker 1: going to be talking with her in an upcoming episode 193 00:09:45,549 --> 00:09:48,130 Speaker 1: this season. So I'm looking forward to asking her about 194 00:09:48,140 --> 00:09:50,900 Speaker 1: this as well. But I think you've said the key phrase, 195 00:09:50,909 --> 00:09:53,580 Speaker 1: deep fake, so I might switch the to that side 196 00:09:53,590 --> 00:09:56,150 Speaker 1: of things. So in terms of the society and, and 197 00:09:56,159 --> 00:09:59,489 Speaker 1: culture in general, um there's some people that are hesitant 198 00:09:59,500 --> 00:10:03,539 Speaker 1: about A I particularly around A I limiting jobs, you've 199 00:10:03,549 --> 00:10:07,200 Speaker 1: got deep fakes. I've actually created a clone of my voice. 200 00:10:08,030 --> 00:10:10,940 Speaker 1: What do you try and do to reassure people who 201 00:10:10,950 --> 00:10:12,140 Speaker 1: have hesitations? 202 00:10:12,619 --> 00:10:13,919 Speaker 2: I'm definitely not, 203 00:10:14,659 --> 00:10:18,848 Speaker 2: I would say not directly enthusiastic about technologies that are 204 00:10:18,859 --> 00:10:23,520 Speaker 2: allowing for passing off as another person for copying and pasting. 205 00:10:23,530 --> 00:10:26,718 Speaker 2: Essentially in certain cases, we see the development of those 206 00:10:26,729 --> 00:10:28,890 Speaker 2: technologies for a certain use case and then it does 207 00:10:28,900 --> 00:10:31,520 Speaker 2: start to stray away from that into some of these 208 00:10:31,530 --> 00:10:34,319 Speaker 2: newer kind of applications that are scary as you shared. 209 00:10:34,330 --> 00:10:37,950 Speaker 2: So when it comes to reassuring individuals, my family, my 210 00:10:37,960 --> 00:10:40,900 Speaker 2: community as well and the industry at large, I think 211 00:10:40,909 --> 00:10:43,478 Speaker 2: that it's definitely a problem to see in a straightforward way. 212 00:10:43,580 --> 00:10:47,309 Speaker 2: Honestly, without the hype surrounding it, there is a levity 213 00:10:47,320 --> 00:10:50,320 Speaker 2: associated with the disadvantages of the technology that we do 214 00:10:50,330 --> 00:10:52,950 Speaker 2: need to consider. We also do see the benefits of 215 00:10:52,960 --> 00:10:55,799 Speaker 2: them for different things, whether that's improving your ease of 216 00:10:55,809 --> 00:10:59,059 Speaker 2: using it, just being able to communicate with others. From 217 00:10:59,070 --> 00:11:01,640 Speaker 2: my perspective, what I try to do in my space 218 00:11:01,650 --> 00:11:04,119 Speaker 2: is to look at an honest assessment of the technology, 219 00:11:04,130 --> 00:11:06,159 Speaker 2: which is very common in the ethical A I domain. 220 00:11:06,299 --> 00:11:09,169 Speaker 2: And to see what exactly is it really contributing to 221 00:11:09,210 --> 00:11:11,468 Speaker 2: the problem statement and if it isn't contributing to it, 222 00:11:11,479 --> 00:11:12,358 Speaker 2: then do we need it? 223 00:11:13,099 --> 00:11:18,468 Speaker 1: And in terms of intel's I guess method or communication 224 00:11:18,479 --> 00:11:22,830 Speaker 1: with the society and people at large, are they working 225 00:11:22,840 --> 00:11:24,619 Speaker 1: on things to help people? 226 00:11:25,179 --> 00:11:27,469 Speaker 1: I feel a little bit more comfortable about this new 227 00:11:27,479 --> 00:11:28,579 Speaker 1: world we're moving into. 228 00:11:29,099 --> 00:11:31,659 Speaker 2: Yes. And we, we tackle it from a couple of 229 00:11:31,669 --> 00:11:34,630 Speaker 2: different fronts. We've got some amazing teams working on different 230 00:11:34,640 --> 00:11:38,229 Speaker 2: parts of the puzzle. One of them is democratization where 231 00:11:38,239 --> 00:11:41,359 Speaker 2: one of the challenging things about A I from an 232 00:11:41,369 --> 00:11:43,510 Speaker 2: ethical A I perspective, but also in general, 233 00:11:43,625 --> 00:11:46,655 Speaker 2: from a development perspective is being able to give communities 234 00:11:46,664 --> 00:11:48,885 Speaker 2: access to the technology so that they can test it 235 00:11:48,895 --> 00:11:51,484 Speaker 2: and validate it. I've been speaking about ethical A I 236 00:11:51,494 --> 00:11:54,614 Speaker 2: for about two years now or so. Last year, we 237 00:11:54,625 --> 00:11:56,864 Speaker 2: really didn't have the same amount of tools and techniques 238 00:11:56,875 --> 00:11:59,585 Speaker 2: that we have this year and also the popularity of 239 00:11:59,594 --> 00:12:01,924 Speaker 2: testing and validating A I systems, right? 240 00:12:02,489 --> 00:12:06,179 Speaker 2: We always understand and I think many companies and organizations 241 00:12:06,190 --> 00:12:09,030 Speaker 2: understand it's not a one size fits all solution for 242 00:12:09,039 --> 00:12:12,469 Speaker 2: ethical A I. Um you know, many companies and organizations 243 00:12:12,479 --> 00:12:14,820 Speaker 2: are trying to do their best. So I would say 244 00:12:14,830 --> 00:12:17,530 Speaker 2: that again that, that push back that community that we're 245 00:12:17,539 --> 00:12:19,880 Speaker 2: trying to create around ethical A I is critical for 246 00:12:19,890 --> 00:12:23,140 Speaker 2: us going forward to be able to better build solutions. 247 00:12:23,409 --> 00:12:25,489 Speaker 1: Has there been any case studies within intel that you 248 00:12:25,500 --> 00:12:29,419 Speaker 1: could share that maybe there was a real challenging ethical 249 00:12:29,429 --> 00:12:30,299 Speaker 1: conundrum 250 00:12:30,809 --> 00:12:35,349 Speaker 1: uh for producing A I software and you know, how, 251 00:12:35,359 --> 00:12:36,729 Speaker 1: how was it resolved? How did you work 252 00:12:36,739 --> 00:12:37,250 Speaker 1: through it? 253 00:12:37,280 --> 00:12:39,520 Speaker 2: Generative A I is definitely a very big one. So 254 00:12:39,530 --> 00:12:43,559 Speaker 2: we're always actively cautious about the types of implications of 255 00:12:43,570 --> 00:12:47,030 Speaker 2: our technology, whether or not we can incorporate disclaimers or 256 00:12:47,039 --> 00:12:49,829 Speaker 2: clarify on the intent of it as well. And um Graham, 257 00:12:49,840 --> 00:12:51,440 Speaker 2: one of my favorite parts 258 00:12:51,525 --> 00:12:53,825 Speaker 2: of ethical A I from a technical perspective in terms 259 00:12:53,835 --> 00:12:57,554 Speaker 2: of solutions is something called model cards. Model cards, clarify 260 00:12:57,565 --> 00:13:00,015 Speaker 2: a very simple theme around ethical A I which is, 261 00:13:00,215 --> 00:13:02,744 Speaker 2: you know, figure out what exactly is the intention the 262 00:13:02,755 --> 00:13:05,875 Speaker 2: core assumptions and the development that went behind the model 263 00:13:05,885 --> 00:13:07,585 Speaker 2: and what you're going to use it for as part 264 00:13:07,594 --> 00:13:08,224 Speaker 2: of deployment. 265 00:13:08,510 --> 00:13:10,679 Speaker 2: And I think that for me personally, I see that 266 00:13:10,690 --> 00:13:13,348 Speaker 2: that theme is conveyed as part of our efforts in 267 00:13:13,359 --> 00:13:15,570 Speaker 2: generative A I, there's a lot of challenging things out 268 00:13:15,580 --> 00:13:18,710 Speaker 2: there when it comes to image generation, copyright, et cetera 269 00:13:18,820 --> 00:13:22,630 Speaker 2: or even, you know, object detection related technologies for retail. 270 00:13:22,640 --> 00:13:25,858 Speaker 2: If you have solutions like intelligent cue management or automated 271 00:13:25,869 --> 00:13:28,409 Speaker 2: self checkout, it makes sense. But you know, how do 272 00:13:28,419 --> 00:13:30,330 Speaker 2: we keep it from proliferating otherwise? 273 00:13:30,590 --> 00:13:33,280 Speaker 1: And what sort of work is going on with inclusive 274 00:13:33,289 --> 00:13:33,299 Speaker 1: A 275 00:13:33,309 --> 00:13:34,710 Speaker 2: I diversity of state 276 00:13:35,135 --> 00:13:38,034 Speaker 2: is critical for the A I models that we're building today, 277 00:13:38,044 --> 00:13:41,955 Speaker 2: whether that's detection of skin agnostic of skin tone or 278 00:13:41,965 --> 00:13:45,194 Speaker 2: being able to adapt to different folks with different accents. 279 00:13:45,205 --> 00:13:47,734 Speaker 2: So at intel and again, across the industry, I think 280 00:13:47,744 --> 00:13:49,944 Speaker 2: a lot of the efforts are really about making sure 281 00:13:49,955 --> 00:13:52,275 Speaker 2: we have the right people on board, the right experts 282 00:13:52,284 --> 00:13:55,505 Speaker 2: with different backgrounds, we're able to contribute to the technologies. 283 00:13:55,695 --> 00:13:56,025 Speaker 1: One 284 00:13:56,034 --> 00:13:59,635 Speaker 1: thing when I started looking into machine learning very quickly, 285 00:13:59,645 --> 00:14:01,195 Speaker 1: I got a sense of, 286 00:14:01,559 --> 00:14:04,349 Speaker 1: you know, being a traditional engineer, you kind of go OK, 287 00:14:04,359 --> 00:14:07,010 Speaker 1: input output and you kind of know what's in the 288 00:14:07,020 --> 00:14:10,929 Speaker 1: in the black box to transform it. When I started 289 00:14:10,940 --> 00:14:14,520 Speaker 1: working with A I and some machine learning code, I 290 00:14:14,530 --> 00:14:16,880 Speaker 1: couldn't get a sense of that 1 to 1 kind 291 00:14:16,890 --> 00:14:19,099 Speaker 1: of mapping of what the output is to input and 292 00:14:19,109 --> 00:14:23,190 Speaker 1: that comes to the to transparency and uh explainability of 293 00:14:23,200 --> 00:14:24,140 Speaker 1: A I algorithms. 294 00:14:24,809 --> 00:14:27,070 Speaker 1: What are you seeing and also what is intel seeing 295 00:14:27,080 --> 00:14:30,789 Speaker 1: around trying to make that understandable to the end users. 296 00:14:31,070 --> 00:14:34,169 Speaker 2: It's a really interesting question because explainability is one of 297 00:14:34,179 --> 00:14:36,809 Speaker 2: the first topics that we think about when we think 298 00:14:36,820 --> 00:14:39,780 Speaker 2: about responsibly. I and I agree the black box metaphor 299 00:14:39,789 --> 00:14:43,530 Speaker 2: has been used so many times um because it's true. 300 00:14:43,580 --> 00:14:47,969 Speaker 2: But the key idea is about demystifying what exactly is 301 00:14:47,979 --> 00:14:52,010 Speaker 2: going on within the model. Whether that is the internal representation, again, 302 00:14:52,020 --> 00:14:54,320 Speaker 2: the data that it's pulling from how the data is 303 00:14:54,330 --> 00:14:55,409 Speaker 2: being leveraged feature and 304 00:14:55,502 --> 00:14:59,002 Speaker 2: importance, et cetera. There's also an added consideration to explain 305 00:14:59,013 --> 00:15:01,862 Speaker 2: ability around surfacing that to an end user. For them 306 00:15:01,872 --> 00:15:04,622 Speaker 2: to understand why the model made a decision I would 307 00:15:04,632 --> 00:15:06,783 Speaker 2: say with Intel, we're approaching it in a couple of 308 00:15:06,793 --> 00:15:09,232 Speaker 2: different ways. And I'm just, I'm very excited to see 309 00:15:09,242 --> 00:15:12,393 Speaker 2: how can different experts approach our problems. We have a 310 00:15:12,403 --> 00:15:16,083 Speaker 2: dedicated suite of technologies for explainability. I led a team 311 00:15:16,093 --> 00:15:19,132 Speaker 2: that was developing one of these for Intel Cno where again, 312 00:15:19,143 --> 00:15:19,713 Speaker 2: you're getting that 313 00:15:19,935 --> 00:15:24,455 Speaker 2: internal representation analysis, Saliency maps and other technologies for explainability. 314 00:15:24,585 --> 00:15:28,075 Speaker 2: We also incorporate transparency and explainability into our algorithm. So 315 00:15:28,085 --> 00:15:30,635 Speaker 2: whether that's being able to visualize what's going on again, 316 00:15:30,645 --> 00:15:34,726 Speaker 2: saliency maps or you know, really good user experience user 317 00:15:34,736 --> 00:15:37,526 Speaker 2: interface to figure out why am I being surfaced this 318 00:15:37,536 --> 00:15:40,255 Speaker 2: particular prediction or decision from a model? I'd say that's 319 00:15:40,265 --> 00:15:42,335 Speaker 2: a couple of the ways that we are integrating and 320 00:15:42,346 --> 00:15:44,265 Speaker 2: thinking about explainability at Intel. 321 00:15:46,979 --> 00:15:50,669 Speaker 1: You're listening to technically speaking an Intel podcast. We'll be 322 00:15:50,679 --> 00:15:51,280 Speaker 1: right back. 323 00:16:00,010 --> 00:16:03,309 Speaker 1: Welcome back to technically speaking an Intel podcast. 324 00:16:07,239 --> 00:16:09,179 Speaker 1: One of the obviously the big things is around the 325 00:16:09,190 --> 00:16:13,320 Speaker 1: privacy and security of data. Perhaps you could outline some 326 00:16:13,330 --> 00:16:17,150 Speaker 1: of the new techniques and new initiatives out in the 327 00:16:17,159 --> 00:16:20,059 Speaker 1: industry to try and use the power of A I 328 00:16:20,070 --> 00:16:23,179 Speaker 1: but still protect companies, information and and 329 00:16:23,190 --> 00:16:23,619 Speaker 1: data. 330 00:16:23,869 --> 00:16:27,179 Speaker 2: I would say there's mechanisms like differential privacy and many others, 331 00:16:27,190 --> 00:16:30,440 Speaker 2: homomorphic encryption. These were incredibly popular two years ago, you 332 00:16:30,450 --> 00:16:32,299 Speaker 2: kind of don't hear them a lot now. So again, 333 00:16:32,309 --> 00:16:34,580 Speaker 2: the hype is it it depends on the technology of 334 00:16:34,590 --> 00:16:34,979 Speaker 2: the day. 335 00:16:35,380 --> 00:16:38,520 Speaker 2: But yes, localization is a key thing. It's actually something 336 00:16:38,530 --> 00:16:40,539 Speaker 2: I have the opportunity to look at now as part 337 00:16:40,549 --> 00:16:43,869 Speaker 2: of my role around hybrid A I edge versus cloud 338 00:16:43,880 --> 00:16:47,270 Speaker 2: edge and cloud. So there's a number of different parameters 339 00:16:47,280 --> 00:16:49,809 Speaker 2: and assumptions that we can start to make at the 340 00:16:49,820 --> 00:16:53,599 Speaker 2: edge around localization privacy of data, not necessarily having 341 00:16:53,684 --> 00:16:56,094 Speaker 2: to communicate it back to the cloud that are changing 342 00:16:56,104 --> 00:16:58,205 Speaker 2: the way that we think about data privacy and security 343 00:16:58,215 --> 00:17:01,604 Speaker 2: for A I models Federated learning is another paradigm like this. 344 00:17:01,815 --> 00:17:04,524 Speaker 2: So to put it shortly, I'd say there are mechanisms 345 00:17:04,535 --> 00:17:06,525 Speaker 2: that are coming up in place, but there is still 346 00:17:06,535 --> 00:17:11,125 Speaker 2: more needed emphasis on security and privacy, more development for technologies, 347 00:17:11,135 --> 00:17:11,635 Speaker 2: et cetera. 348 00:17:12,439 --> 00:17:15,280 Speaker 1: OK. So just to extend that just a little bit more. 349 00:17:15,290 --> 00:17:18,079 Speaker 1: So say if you're meeting with an executive saying, I've 350 00:17:18,089 --> 00:17:20,300 Speaker 1: been hearing all about large language models and I was 351 00:17:20,310 --> 00:17:23,379 Speaker 1: talking to my colleague uh in another company and they're 352 00:17:23,390 --> 00:17:26,609 Speaker 1: starting to use chatbots with within their organization and using 353 00:17:26,619 --> 00:17:30,520 Speaker 1: the power of that is that related to large language models, 354 00:17:30,530 --> 00:17:33,438 Speaker 1: but fine tuning it to their own corporate data in 355 00:17:33,449 --> 00:17:34,020 Speaker 1: their own 356 00:17:34,380 --> 00:17:37,339 Speaker 1: servers. If you like I sort of on the right track. 357 00:17:37,520 --> 00:17:40,010 Speaker 2: Yes, that is a perfect use case. And thank you 358 00:17:40,020 --> 00:17:42,910 Speaker 2: for bringing that up, you know, centralization of data on 359 00:17:42,920 --> 00:17:45,839 Speaker 2: your server. There's also red teaming um gram that's worth 360 00:17:45,849 --> 00:17:49,218 Speaker 2: mentioning where you're testing your model or your system thoroughly 361 00:17:49,229 --> 00:17:52,689 Speaker 2: with the generative A I space. There's come to life, 362 00:17:52,699 --> 00:17:55,040 Speaker 2: a lot of different types of red teaming approaches including 363 00:17:55,050 --> 00:17:57,500 Speaker 2: prompt injection and many others, which is really a 364 00:17:57,574 --> 00:18:00,574 Speaker 2: being able to test and mock the kinds of inputs 365 00:18:00,584 --> 00:18:02,994 Speaker 2: that adversaries would provide to your model and figure out 366 00:18:03,005 --> 00:18:05,055 Speaker 2: how the model is going to behave. What are its 367 00:18:05,064 --> 00:18:07,875 Speaker 2: strengths and weaknesses, et cetera. Of course, the compute needed 368 00:18:07,885 --> 00:18:10,755 Speaker 2: for that is another story. But in addition to that, 369 00:18:10,765 --> 00:18:13,334 Speaker 2: there's also again, the testing and validation approaches. So red 370 00:18:13,344 --> 00:18:17,505 Speaker 2: teaming is really critical to that validating how susceptible your 371 00:18:17,515 --> 00:18:20,675 Speaker 2: model is to potential attacks, whether it's biased, et cetera. 372 00:18:20,974 --> 00:18:23,905 Speaker 2: So lots of, lots of cool and interesting approaches coming up. 373 00:18:23,915 --> 00:18:25,954 Speaker 2: Exactly as you noted, that's a key example. 374 00:18:26,290 --> 00:18:29,670 Speaker 1: So going back on the ethics side of things, what 375 00:18:29,680 --> 00:18:33,909 Speaker 1: are some of the arguments for a corporation, an organization 376 00:18:33,920 --> 00:18:37,619 Speaker 1: to have a clear set of code of ethics and 377 00:18:37,630 --> 00:18:42,589 Speaker 1: is intel helping companies establish those sorts of guidelines and frameworks. 378 00:18:43,550 --> 00:18:46,319 Speaker 2: There is a number of different best practices that organizations 379 00:18:46,329 --> 00:18:49,329 Speaker 2: can incorporate today for responsible A I. One of them 380 00:18:49,339 --> 00:18:52,369 Speaker 2: is the internal governance assessments that we talked about, which 381 00:18:52,380 --> 00:18:55,079 Speaker 2: is a step by step process to checking where A 382 00:18:55,089 --> 00:18:57,189 Speaker 2: I is used in your organization. How is it being 383 00:18:57,199 --> 00:19:00,310 Speaker 2: shipped outside? What's your go to market strategy? What's your 384 00:19:00,319 --> 00:19:02,069 Speaker 2: change management strategy, etcetera. 385 00:19:02,349 --> 00:19:05,889 Speaker 2: So in terms of Intel's contributions, we're very excited and 386 00:19:05,900 --> 00:19:10,219 Speaker 2: passionate about communication with customers and partners and communities in 387 00:19:10,229 --> 00:19:13,790 Speaker 2: general around. What exactly can we do to help with 388 00:19:13,800 --> 00:19:16,290 Speaker 2: the ethical A I development that can include, you know, 389 00:19:16,300 --> 00:19:20,810 Speaker 2: potential compute platforms that help with running this type of solutions, preprocessing, 390 00:19:20,819 --> 00:19:22,369 Speaker 2: post processing. What exactly 391 00:19:22,494 --> 00:19:25,275 Speaker 2: you need towards that? Or if we have developers working 392 00:19:25,286 --> 00:19:27,125 Speaker 2: with Intel Open Veno and I work in the Open 393 00:19:27,176 --> 00:19:29,315 Speaker 2: Vino team right now. We want to know what makes 394 00:19:29,326 --> 00:19:31,705 Speaker 2: it easier for developers to be able to run these 395 00:19:31,715 --> 00:19:34,984 Speaker 2: models and deploy them their feedback in terms of, you know, hey, 396 00:19:34,994 --> 00:19:36,955 Speaker 2: you know, is this challenging to use? I don't know 397 00:19:36,965 --> 00:19:39,215 Speaker 2: how this is working. Um something that I do as 398 00:19:39,225 --> 00:19:41,955 Speaker 2: part of my evangelism team is again, helping contribute to that. 399 00:19:41,965 --> 00:19:42,546 Speaker 2: So I would 400 00:19:42,641 --> 00:19:45,131 Speaker 2: say that as part of the practices, there's a number 401 00:19:45,141 --> 00:19:48,592 Speaker 2: of different things that we do today with solutions with guardrails, 402 00:19:48,602 --> 00:19:51,661 Speaker 2: with assessments. And at Intel, we're trying to help with 403 00:19:51,671 --> 00:19:55,182 Speaker 2: the communication, the establishment of these elements as well as 404 00:19:55,192 --> 00:19:59,161 Speaker 2: the technical solutions and how we can help build foundations 405 00:19:59,171 --> 00:20:02,342 Speaker 2: that our partners, customers, the community and industry can take 406 00:20:02,352 --> 00:20:02,702 Speaker 2: from there. 407 00:20:03,209 --> 00:20:06,188 Speaker 1: You mentioned that you're part of the Intel Open Vino group. 408 00:20:06,199 --> 00:20:08,869 Speaker 1: Perhaps you could spend a bit of time just explaining 409 00:20:08,880 --> 00:20:12,069 Speaker 1: what that group does and what your role in. It is. 410 00:20:12,270 --> 00:20:16,270 Speaker 2: Sure. The Intel Open Veno group is a team dedicated 411 00:20:16,280 --> 00:20:19,879 Speaker 2: to helping provide capabilities and developing our Open Veno toolkit. 412 00:20:20,229 --> 00:20:23,379 Speaker 2: The toolkit is centered around computer vision related applications and 413 00:20:23,390 --> 00:20:26,310 Speaker 2: it's recently expanded over five years to generative A I. 414 00:20:26,589 --> 00:20:30,229 Speaker 2: And it is really centered around taking models in many 415 00:20:30,239 --> 00:20:34,400 Speaker 2: different frameworks like Pytorch, tensorflow, caras, et cetera and converting 416 00:20:34,410 --> 00:20:37,670 Speaker 2: and optimizing them to an intermediate representation format that you 417 00:20:37,680 --> 00:20:41,089 Speaker 2: can deploy on different hardware, including Intel CP US, GP 418 00:20:41,099 --> 00:20:42,530 Speaker 2: US and other types of hardware. 419 00:20:43,369 --> 00:20:47,719 Speaker 1: And have you seen any I guess impact on, on 420 00:20:47,729 --> 00:20:50,339 Speaker 1: innovation to, to put it bluntly does having a code 421 00:20:50,349 --> 00:20:53,179 Speaker 1: of ethics, put a brake on innovation and 422 00:20:53,689 --> 00:20:57,459 Speaker 1: for individual engineers, does it leave them feeling? Oh, maybe 423 00:20:57,469 --> 00:20:59,750 Speaker 1: I shouldn't try these things. Is it a hindrance? 424 00:20:59,949 --> 00:21:03,669 Speaker 2: The big question I've encountered this question before but my, 425 00:21:03,680 --> 00:21:06,760 Speaker 2: my answer to it is no, it is not. Because 426 00:21:06,900 --> 00:21:09,579 Speaker 2: um what again, my personal opinion and what I've also 427 00:21:09,589 --> 00:21:13,359 Speaker 2: seen at Intel and through my colleagues, mentors and industry 428 00:21:13,369 --> 00:21:14,790 Speaker 2: academia and other circles 429 00:21:15,083 --> 00:21:18,743 Speaker 2: at the core of innovation is certain themes like improving 430 00:21:18,753 --> 00:21:21,432 Speaker 2: quality of life, et cetera. And as a part of 431 00:21:21,442 --> 00:21:25,703 Speaker 2: that human rights responsible A I adoption of technologies and 432 00:21:25,713 --> 00:21:29,342 Speaker 2: understanding why you're using technologies with awareness, those are all 433 00:21:29,353 --> 00:21:32,223 Speaker 2: key attributes. So I would say if we're able to 434 00:21:32,233 --> 00:21:34,902 Speaker 2: design the process in a way that's efficient, that is 435 00:21:34,912 --> 00:21:35,983 Speaker 2: incorporating the minimum 436 00:21:36,076 --> 00:21:39,656 Speaker 2: requirements and has the flexibility to grow with the technology, 437 00:21:39,666 --> 00:21:42,056 Speaker 2: then we're doing it right? And it is not a hindrance. 438 00:21:42,066 --> 00:21:44,975 Speaker 2: Time to go to market is a key item. However 439 00:21:44,984 --> 00:21:48,275 Speaker 2: responsible A I process is while they may take time, 440 00:21:48,286 --> 00:21:50,485 Speaker 2: they don't necessarily have to hinder that goal if they're 441 00:21:50,494 --> 00:21:53,465 Speaker 2: streamlined and done efficiently. The onus is on all of 442 00:21:53,475 --> 00:21:55,004 Speaker 2: us to be able to contribute to that kind of 443 00:21:55,015 --> 00:21:57,205 Speaker 2: strategy or development of that strategy. 444 00:21:57,645 --> 00:21:59,895 Speaker 1: And in terms of the 445 00:22:00,250 --> 00:22:02,910 Speaker 1: A I evolving over the next five years, you know, 446 00:22:02,920 --> 00:22:04,020 Speaker 1: where do you see it going? 447 00:22:04,150 --> 00:22:07,239 Speaker 2: Human centered A I, that is my personal opinion on it. 448 00:22:07,250 --> 00:22:09,520 Speaker 2: I've done a lot of research on it. I also 449 00:22:09,530 --> 00:22:12,780 Speaker 2: had the opportunity to author publication on it. Technology that's 450 00:22:12,790 --> 00:22:15,629 Speaker 2: centered around the human experience that is contributing to the 451 00:22:15,640 --> 00:22:18,379 Speaker 2: way that we think that we act and that we 452 00:22:18,390 --> 00:22:19,550 Speaker 2: interact with others 453 00:22:19,635 --> 00:22:21,564 Speaker 2: I would say is the key thing. And for me, 454 00:22:21,574 --> 00:22:24,954 Speaker 2: that's the most exciting applications, whether that's smart care robots 455 00:22:24,964 --> 00:22:28,915 Speaker 2: for the elderly, using generative A I for health care applications, 456 00:22:28,944 --> 00:22:33,155 Speaker 2: identifying new protein folding related techniques or something similar. But 457 00:22:33,165 --> 00:22:36,185 Speaker 2: centered around the human experience, I would say. So Human 458 00:22:36,194 --> 00:22:39,004 Speaker 2: Centered A I is a good theme for that overarching journey. 459 00:22:39,800 --> 00:22:43,880 Speaker 1: Yeah, the Human Centered A I is a very interesting concept. 460 00:22:43,890 --> 00:22:47,109 Speaker 1: And have you seen any examples, either in the start 461 00:22:47,119 --> 00:22:50,939 Speaker 1: up community or within Intel or in the industry where 462 00:22:51,140 --> 00:22:54,859 Speaker 1: you've given some examples? But is any that are actually 463 00:22:54,869 --> 00:22:56,459 Speaker 1: like kind of in production today? 464 00:22:58,130 --> 00:23:01,410 Speaker 2: So we have some accessibility research that we've done with Intel. 465 00:23:01,420 --> 00:23:04,040 Speaker 2: You know, Laman Kachin also leads the human computer interaction 466 00:23:04,050 --> 00:23:05,599 Speaker 2: lab and we see a lot of I see a 467 00:23:05,609 --> 00:23:08,679 Speaker 2: lot of great research coming out of that around accessibility, 468 00:23:08,689 --> 00:23:11,910 Speaker 2: hearing related initiatives, et cetera. I would say that they're 469 00:23:11,920 --> 00:23:14,209 Speaker 2: in the process of being researched right now to my 470 00:23:14,219 --> 00:23:17,630 Speaker 2: knowledge across the industry of technologies that we can actively 471 00:23:17,640 --> 00:23:20,560 Speaker 2: put in place. But there are blueprints in place for 472 00:23:20,569 --> 00:23:21,359 Speaker 2: Human Centered A I 473 00:23:21,454 --> 00:23:24,555 Speaker 2: technologies. So it will be exciting to see how they evolve, 474 00:23:24,564 --> 00:23:27,665 Speaker 2: how you know, we take into consideration newer models like 475 00:23:27,675 --> 00:23:30,415 Speaker 2: generative A I that again, popularity just kind of popped up, 476 00:23:30,425 --> 00:23:32,415 Speaker 2: but they've been around for a while. So we need 477 00:23:32,425 --> 00:23:34,415 Speaker 2: to see how the technology adapts, but I think it 478 00:23:34,425 --> 00:23:37,564 Speaker 2: will stay true. To like the test of time in 479 00:23:37,574 --> 00:23:39,264 Speaker 2: five years time and then we will be able to 480 00:23:39,275 --> 00:23:42,034 Speaker 2: see and interact with A I applications that are centered 481 00:23:42,045 --> 00:23:44,774 Speaker 2: around our experiences around nature, et cetera. 482 00:23:45,354 --> 00:23:47,265 Speaker 1: How do you differentiate the two between 483 00:23:47,819 --> 00:23:51,280 Speaker 1: the ethical A I and responsible A I? Um because 484 00:23:51,290 --> 00:23:52,709 Speaker 1: in my mind, it's kind of a little bit in 485 00:23:52,719 --> 00:23:54,109 Speaker 1: a little bit jumbled. 486 00:23:54,699 --> 00:23:57,900 Speaker 2: Sure, I use the term actually in overlap, uh just 487 00:23:57,910 --> 00:24:01,250 Speaker 2: my personal bias towards you. But I, I have seen 488 00:24:01,260 --> 00:24:04,959 Speaker 2: that there are differences, there's been multiple efforts to establish 489 00:24:04,969 --> 00:24:07,938 Speaker 2: a nomenclature in the ethical A I domain. So responsible 490 00:24:07,949 --> 00:24:10,319 Speaker 2: A I is seen more as the internal governance, the 491 00:24:10,329 --> 00:24:13,459 Speaker 2: processes and practices that we put towards A I. Whereas 492 00:24:13,469 --> 00:24:16,250 Speaker 2: ethically I is seen as really maybe kind of a 493 00:24:16,260 --> 00:24:19,880 Speaker 2: combination of the societal and technical aspects as I shared earlier. 494 00:24:19,890 --> 00:24:22,589 Speaker 2: So responsibly I in a sense is the accountability and 495 00:24:22,599 --> 00:24:23,869 Speaker 2: responsibility part of it. 496 00:24:24,430 --> 00:24:27,229 Speaker 1: Uh I talked earlier about the future of A I. 497 00:24:27,569 --> 00:24:30,300 Speaker 1: How is intel gonna be part of that wave in 498 00:24:30,310 --> 00:24:33,089 Speaker 1: terms of its programs and solutions for customers 499 00:24:33,959 --> 00:24:36,989 Speaker 2: A I is a key inflection point for us. We're 500 00:24:37,000 --> 00:24:40,719 Speaker 2: excited to ride the new wave, collaborate with our again, partners, 501 00:24:40,729 --> 00:24:44,520 Speaker 2: customers communities and um see what we can do next. 502 00:24:44,530 --> 00:24:45,790 Speaker 2: What's the next great big thing? 503 00:24:46,089 --> 00:24:48,750 Speaker 2: Uh Generative A I is definitely a key focus for us. 504 00:24:48,760 --> 00:24:51,819 Speaker 2: It's what our customers want, it's what developers want and 505 00:24:51,829 --> 00:24:54,589 Speaker 2: it's what users want as well for their content creation 506 00:24:54,599 --> 00:24:57,569 Speaker 2: and many, many other needs. So we're very focused on that. 507 00:24:57,579 --> 00:25:00,760 Speaker 2: We're also incredibly focused on the compute. I see a 508 00:25:00,770 --> 00:25:02,619 Speaker 2: lot of and get to work with a lot of 509 00:25:02,630 --> 00:25:06,520 Speaker 2: wonderful engineers that are very passionate about solving these problems 510 00:25:06,530 --> 00:25:09,180 Speaker 2: at hand. Specifically these um because there's, you know, so 511 00:25:09,189 --> 00:25:11,359 Speaker 2: much that you can do a lot of problems in 512 00:25:11,369 --> 00:25:12,760 Speaker 2: the LLM and generative A I 513 00:25:12,895 --> 00:25:16,714 Speaker 2: based around, you know, large models, large footprint, changing outputs, 514 00:25:16,724 --> 00:25:20,714 Speaker 2: not a lot of predictability, challenging to benchmark, etcetera. So 515 00:25:20,724 --> 00:25:23,994 Speaker 2: I think that Intel is working on and actively positioned 516 00:25:24,005 --> 00:25:28,734 Speaker 2: to help our customers. Developers provide these types of optimizations, 517 00:25:28,744 --> 00:25:30,754 Speaker 2: the right kind of compute et cetera for, for the 518 00:25:30,765 --> 00:25:32,714 Speaker 2: new wave of A I but outside of generative A 519 00:25:32,724 --> 00:25:35,135 Speaker 2: I also, there's a lot of other A I applications 520 00:25:35,145 --> 00:25:37,594 Speaker 2: that we're aware of human centered A I, et cetera 521 00:25:37,704 --> 00:25:41,004 Speaker 2: that we are also actively working on. So we're ready. 522 00:25:41,689 --> 00:25:45,000 Speaker 1: Oh, that's, that's good to hear. I've definitely learnt quite 523 00:25:45,010 --> 00:25:47,510 Speaker 1: a lot. So thank you very much for your time. 524 00:25:47,520 --> 00:25:49,069 Speaker 2: Thank you, Graham. Appreciate it. 525 00:25:53,520 --> 00:25:55,709 Speaker 1: I would like to thank my guest Ria Chu for 526 00:25:55,719 --> 00:25:58,310 Speaker 1: joining me today on this special episode of technically speaking, 527 00:25:58,319 --> 00:25:59,458 Speaker 1: an Intel podcast 528 00:26:00,709 --> 00:26:04,449 Speaker 1: ethics and artificial intelligence are so important right now. And 529 00:26:04,459 --> 00:26:07,199 Speaker 1: what I've learnt from today's discussion with R A, having 530 00:26:07,209 --> 00:26:09,739 Speaker 1: a code of ethics can be an important standard, especially 531 00:26:09,750 --> 00:26:13,270 Speaker 1: when it comes to deep fakes companies in the media 532 00:26:13,280 --> 00:26:17,069 Speaker 1: industry should have a rule about never Impersonating someone without 533 00:26:17,079 --> 00:26:20,050 Speaker 1: their knowledge. In my experience, I've been able to clone 534 00:26:20,060 --> 00:26:22,770 Speaker 1: my own voice within a day and it's a pretty 535 00:26:22,780 --> 00:26:23,530 Speaker 1: good quality 536 00:26:24,069 --> 00:26:26,429 Speaker 1: for me as an engineer and a technologist. I think 537 00:26:26,439 --> 00:26:30,050 Speaker 1: that's really interesting. However, it does throw up a lot 538 00:26:30,060 --> 00:26:33,129 Speaker 1: of questions around ethics and whether we should do these things. 539 00:26:33,510 --> 00:26:35,609 Speaker 1: The other thing Ria touched on is human centered A 540 00:26:35,619 --> 00:26:39,890 Speaker 1: I And that's really interesting from my perspective, I think 541 00:26:39,900 --> 00:26:44,170 Speaker 1: technology has moved towards trying to be human centered. And 542 00:26:44,180 --> 00:26:46,919 Speaker 1: it's good to see that A I wave that is 543 00:26:46,930 --> 00:26:50,629 Speaker 1: coming is still trying to keep humans as the center 544 00:26:50,640 --> 00:26:53,010 Speaker 1: of any product and technology design. 545 00:26:53,640 --> 00:26:56,869 Speaker 1: And talking with Rea really did hit home to me 546 00:26:56,880 --> 00:27:00,629 Speaker 1: that it is artificial intelligence, but I am looking at 547 00:27:00,640 --> 00:27:03,639 Speaker 1: the way that it can actually augment us. I think 548 00:27:03,650 --> 00:27:07,060 Speaker 1: that it will augment our jobs. I don't think on 549 00:27:07,069 --> 00:27:09,959 Speaker 1: balance that it will take away jobs. You only have 550 00:27:09,969 --> 00:27:12,459 Speaker 1: to look back in history from the printing press to 551 00:27:12,469 --> 00:27:16,199 Speaker 1: the loom. The A I wave that we're going through 552 00:27:16,209 --> 00:27:19,159 Speaker 1: now is just another evolution of us as a species. 553 00:27:19,170 --> 00:27:22,569 Speaker 1: And I love discussion around the ethics and the philosophy 554 00:27:22,579 --> 00:27:23,339 Speaker 1: of A I, 555 00:27:23,729 --> 00:27:25,189 Speaker 1: I hope it will continue. 556 00:27:27,199 --> 00:27:29,709 Speaker 1: And that's all for our first episode. Thanks so much 557 00:27:29,719 --> 00:27:32,410 Speaker 1: for joining me today. Please join us on Tuesday October 558 00:27:32,420 --> 00:27:35,239 Speaker 1: 17th for the next episode where we speak with experts 559 00:27:35,250 --> 00:27:38,959 Speaker 1: on the way A I is innovating agribusiness solutions. You 560 00:27:38,969 --> 00:27:42,419 Speaker 1: can follow me on linkedin and Twitter or X with 561 00:27:42,430 --> 00:27:45,449 Speaker 1: the handle at Graham Class or check the show notes 562 00:27:45,459 --> 00:27:49,380 Speaker 1: page for links. This has been technically speaking, an Intel podcast, 563 00:27:52,369 --> 00:27:55,968 Speaker 1: technically speaking was produced by Ruby Studios from iheartradio in 564 00:27:55,979 --> 00:27:58,939 Speaker 1: partnership with Intel and hosted by me Graham Class. 565 00:27:59,680 --> 00:28:03,000 Speaker 1: Our executive producer is Molly. So our EP of post 566 00:28:03,010 --> 00:28:07,160 Speaker 1: production is James Foster and our supervising producer is Nikia Swinton. 567 00:28:07,810 --> 00:28:11,060 Speaker 1: This episode was edited by Ciara Spring and written and 568 00:28:11,069 --> 00:28:12,609 Speaker 1: produced by Tyree Rush.