WEBVTT 1 00:00:00.450 --> 00:00:02.010 Artificial intelligence 2 00:00:02.010 --> 00:00:04.020 and the algorithms that power it 3 00:00:04.020 --> 00:00:07.673 can be a tremendous benefit to businesses and to society. 4 00:00:07.673 --> 00:00:09.539 AI can enhance productivity, 5 00:00:09.539 --> 00:00:11.972 sharpen analytical capabilities, 6 00:00:11.972 --> 00:00:14.880 and vastly improve the consumer experience. 7 00:00:14.880 --> 00:00:17.430 But it can also cause harm, damaging the planet, 8 00:00:17.430 --> 00:00:19.560 and marginalizing smaller businesses 9 00:00:19.560 --> 00:00:21.510 and underrepresented communities. 10 00:00:21.510 --> 00:00:24.450 Implementing responsible AI is the process 11 00:00:24.450 --> 00:00:28.080 of understanding how AI can lead to negative outcomes, 12 00:00:28.080 --> 00:00:31.050 and then putting guardrails in place to mitigate harm 13 00:00:31.050 --> 00:00:33.840 while still delivering the business impact we're after. 14 00:00:33.840 --> 00:00:36.930 Let's consider a fictitious company, Olive Market, 15 00:00:36.930 --> 00:00:38.370 that wants to take full advantage 16 00:00:38.370 --> 00:00:41.643 of the power of AI and do so in a responsible way. 17 00:00:42.600 --> 00:00:45.720 Olive Market wants what every retail company wants: 18 00:00:45.720 --> 00:00:48.090 to deliver a consistent shopping experience 19 00:00:48.090 --> 00:00:50.010 by maintaining a well-stocked store 20 00:00:50.010 --> 00:00:52.050 amid increasing market volatility. 21 00:00:52.050 --> 00:00:54.810 Responsible AI starts with an understanding 22 00:00:54.810 --> 00:00:56.550 that long-term success depends 23 00:00:56.550 --> 00:00:58.770 on building customer trust and loyalty. 24 00:00:58.770 --> 00:01:01.440 For Olive Market, that means a corporate commitment 25 00:01:01.440 --> 00:01:05.010 to respect the environment, source products with integrity, 26 00:01:05.010 --> 00:01:07.050 and act in ways that support the health 27 00:01:07.050 --> 00:01:09.300 and well-being of its surrounding communities. 28 00:01:09.300 --> 00:01:11.850 As it begins to build its AI infrastructure, 29 00:01:11.850 --> 00:01:13.980 Olive Market would reinforce its commitment 30 00:01:13.980 --> 00:01:17.820 to responsible AI by using external reviewers and tools 31 00:01:17.820 --> 00:01:19.980 to test and assess algorithms 32 00:01:19.980 --> 00:01:21.930 and implement bias-mitigation measures 33 00:01:21.930 --> 00:01:24.060 when using any kind of training data. 34 00:01:24.060 --> 00:01:26.580 One example of the need for bias mitigation 35 00:01:26.580 --> 00:01:28.380 is the use of computer vision. 36 00:01:28.380 --> 00:01:30.990 Computer vision algorithms must be trained 37 00:01:30.990 --> 00:01:32.730 by feeding them labeled images 38 00:01:32.730 --> 00:01:34.620 of customers walking into stores. 39 00:01:34.620 --> 00:01:37.830 But if those images aren't representative of the population, 40 00:01:37.830 --> 00:01:40.050 including people with different skin tones, 41 00:01:40.050 --> 00:01:43.380 Olive Market's computer vision system may undercount people 42 00:01:43.380 --> 00:01:45.840 of different racial and cultural backgrounds. 43 00:01:45.840 --> 00:01:48.210 Thus, these customers' unique dietary needs 44 00:01:48.210 --> 00:01:50.460 and preferences might not be reflected 45 00:01:50.460 --> 00:01:52.560 in Olive Market's purchasing decisions. 46 00:01:52.560 --> 00:01:54.780 To mitigate this bias, store owners 47 00:01:54.780 --> 00:01:57.630 could require their AI solution vendor to test 48 00:01:57.630 --> 00:02:00.780 for various skin tones and achieve comparable performance 49 00:02:00.780 --> 00:02:03.810 before using the computer vision system in their stores. 50 00:02:03.810 --> 00:02:05.880 That way, Olive Market has confidence 51 00:02:05.880 --> 00:02:07.680 the system will work for their customers. 52 00:02:07.680 --> 00:02:10.650 Hallucinations can occur in personalized texts 53 00:02:10.650 --> 00:02:13.500 that are sent out to customers with non-existent items 54 00:02:13.500 --> 00:02:15.630 and incorrect store locations. 55 00:02:15.630 --> 00:02:18.780 Olive Market can adopt a responsible AI approach 56 00:02:18.780 --> 00:02:21.210 by layering on information retrieval 57 00:02:21.210 --> 00:02:23.760 and checking to reduce hallucination occurrences 58 00:02:23.760 --> 00:02:26.490 in the texts sent out to customers. 59 00:02:26.490 --> 00:02:28.440 Negative outcomes can also be a result 60 00:02:28.440 --> 00:02:30.060 of the restocking process. 61 00:02:30.060 --> 00:02:32.250 On its own, the scheduling algorithms 62 00:02:32.250 --> 00:02:35.460 might select produce sources based on the lowest price. 63 00:02:35.460 --> 00:02:37.620 If so, it might source produce 64 00:02:37.620 --> 00:02:40.800 from a large corporate farm far from the store location, 65 00:02:40.800 --> 00:02:42.960 meaning higher transportation costs 66 00:02:42.960 --> 00:02:44.610 and a higher carbon footprint. 67 00:02:44.610 --> 00:02:46.830 To prevent this, the scheduling algorithm 68 00:02:46.830 --> 00:02:49.470 could be trained on data specifically curated 69 00:02:49.470 --> 00:02:51.600 for each store region so it could factor 70 00:02:51.600 --> 00:02:54.030 in sustainable solutions available locally. 71 00:02:54.030 --> 00:02:56.490 Sourcing in a responsible way might even lead 72 00:02:56.490 --> 00:02:59.580 to investments in local farms and local communities. 73 00:02:59.580 --> 00:03:03.090 In fact, the company's IT staff could input local sourcing 74 00:03:03.090 --> 00:03:06.780 as an explicit goal of Olive Market's scheduling algorithms. 75 00:03:06.780 --> 00:03:09.390 And it could make sure that its AI implementation 76 00:03:09.390 --> 00:03:12.000 includes explainability tools that open 77 00:03:12.000 --> 00:03:14.280 the black box because the better its executives 78 00:03:14.280 --> 00:03:17.070 and managers understood how these advanced learning models 79 00:03:17.070 --> 00:03:19.140 make recommendations, the more they would be able 80 00:03:19.140 --> 00:03:21.090 to adjust the algorithms to meet both 81 00:03:21.090 --> 00:03:24.300 Olive Market's commercial goals and its social commitments. 82 00:03:24.300 --> 00:03:26.010 To be successful in the long term, 83 00:03:26.010 --> 00:03:27.390 Olive Market would make sure that 84 00:03:27.390 --> 00:03:30.180 its AI system had human beings in the loop. 85 00:03:30.180 --> 00:03:33.150 We call this our Human plus AI Strategy. 86 00:03:33.150 --> 00:03:35.670 Only humans can give AI recommendations 87 00:03:35.670 --> 00:03:38.400 the kind of sniff test, or reality check, 88 00:03:38.400 --> 00:03:40.200 that prevents companies from blindly 89 00:03:40.200 --> 00:03:42.480 following machine-based recommendations. 90 00:03:42.480 --> 00:03:45.180 Here at BCG, we are devoted to helping companies 91 00:03:45.180 --> 00:03:47.880 like Olive Market build AI the right way. 92 00:03:47.880 --> 00:03:51.120 That's why we have a team of responsible AI experts 93 00:03:51.120 --> 00:03:54.000 that's committed to all our work with machine learning. 94 00:03:54.000 --> 00:03:55.350 We're here to help you transform 95 00:03:55.350 --> 00:03:57.840 your organization and build your bottom line, 96 00:03:57.840 --> 00:03:59.910 and do it in a way that supports you, 97 00:03:59.910 --> 00:04:02.350 your customers, and the community's 98 00:04:02.350 --> 00:04:06.500 shared values-now and in the future.