1 00:00:03,070 --> 00:00:05,000 Hello, thank you for joining my 2 00:00:05,000 --> 00:00:07,140 presentation today. Whether you're online 3 00:00:07,140 --> 00:00:08,910 or rewatching this after going to the 4 00:00:08,910 --> 00:00:11,110 symposium in person, I hope you learned 5 00:00:11,110 --> 00:00:12,710 something and I appreciate you joining. 6 00:00:13,210 --> 00:00:14,810 I'm sure you've heard a lot about AI 7 00:00:14,810 --> 00:00:17,180 during this symposium. Whether online or 8 00:00:17,180 --> 00:00:18,620 in person, there's a lot of different 9 00:00:18,620 --> 00:00:21,190 presentations that talk about AI from a 10 00:00:21,190 --> 00:00:23,990 theological perspective. I'm hoping to 11 00:00:23,990 --> 00:00:25,490 bring the perspective as somebody who 12 00:00:25,490 --> 00:00:27,690 works in the development and deployment of 13 00:00:27,690 --> 00:00:30,730 AI on a regular basis. Today, I will be 14 00:00:30,730 --> 00:00:32,660 talking about what artificial intelligence 15 00:00:32,660 --> 00:00:35,330 is, what it is right now and what we think 16 00:00:35,330 --> 00:00:37,770 it is, how it can be used in church in 17 00:00:37,770 --> 00:00:41,310 useful ways, some concerns that are coming 18 00:00:41,310 --> 00:00:44,810 soon or already appearing with AI, and 19 00:00:44,810 --> 00:00:46,810 finally what opportunities it presents to 20 00:00:46,810 --> 00:00:49,550 us as leaders in the church. My name is 21 00:00:49,550 --> 00:00:51,380 Andy Hammes. I am a director of data 22 00:00:51,380 --> 00:00:53,450 science for Mercy Hospitals here in St. 23 00:00:53,650 --> 00:00:56,090 Louis. I have a background in AI and data 24 00:00:56,090 --> 00:00:57,860 science development, both developing 25 00:00:57,860 --> 00:01:00,790 models myself and coaching and leading 26 00:01:00,790 --> 00:01:02,590 teams who develop and deploy these models 27 00:01:02,590 --> 00:01:06,470 on a daily basis. I've seen firsthand the 28 00:01:06,470 --> 00:01:08,470 development of data science and AI from a 29 00:01:08,470 --> 00:01:10,670 fledgling technology to what it is today 30 00:01:10,670 --> 00:01:13,740 as a major industry. So with all that out 31 00:01:13,740 --> 00:01:16,840 of the way, let's get started. When you 32 00:01:16,840 --> 00:01:19,340 think of AI, what do you think of? Most 33 00:01:19,340 --> 00:01:20,950 people think of a computer which is the 34 00:01:20,950 --> 00:01:23,350 same as a person. Roughly speaking, you 35 00:01:23,350 --> 00:01:24,820 have a few assumptions that you get from 36 00:01:24,820 --> 00:01:28,090 fiction. That is intelligence, awareness, 37 00:01:28,650 --> 00:01:30,690 and in some cases, emotions and emotions 38 00:01:30,690 --> 00:01:32,930 and the development of those emotions is 39 00:01:32,930 --> 00:01:34,860 often part of the plot of whatever 40 00:01:34,860 --> 00:01:39,230 fictional medium you are watching. The 41 00:01:39,230 --> 00:01:41,930 other traditional story for these is the 42 00:01:41,930 --> 00:01:43,740 Turing test. The Turing test is where you 43 00:01:43,740 --> 00:01:45,570 compare the output from an AI from an 44 00:01:45,570 --> 00:01:47,710 output from a person from a specific 45 00:01:47,710 --> 00:01:50,440 question or set of questions or prompts 46 00:01:50,440 --> 00:01:52,540 and see how they compare. If you can't 47 00:01:52,540 --> 00:01:54,880 tell the difference, then that AI has 48 00:01:54,880 --> 00:01:57,650 passed the Turing test. It is worth noting 49 00:01:57,650 --> 00:02:00,120 that this is the test that has been used 50 00:02:00,120 --> 00:02:01,850 historically to show an AI has passed 51 00:02:01,850 --> 00:02:04,890 certain milestones and in many ways, AI 52 00:02:04,890 --> 00:02:07,590 has already passed these milestones and 53 00:02:07,590 --> 00:02:09,930 overall the Turing test likely has been 54 00:02:09,930 --> 00:02:13,300 passed, which illustrates the moving and 55 00:02:13,300 --> 00:02:17,270 changing goal posts that we see in AI and 56 00:02:17,270 --> 00:02:21,710 in these technologies. What is AI now? AI 57 00:02:21,710 --> 00:02:23,910 now has become much more of a watered down 58 00:02:23,910 --> 00:02:25,810 buzzword. Because it has been 59 00:02:25,810 --> 00:02:28,180 commercialized and is worth a lot of money 60 00:02:28,180 --> 00:02:30,280 to the companies who use it, it is crept 61 00:02:30,280 --> 00:02:32,750 into a lot of technologies where it may or 62 00:02:32,750 --> 00:02:35,990 may not be seen in the same way that it 63 00:02:35,990 --> 00:02:37,790 has been historically. From the 64 00:02:37,790 --> 00:02:40,960 smartphones to fridges to watches, you see 65 00:02:40,960 --> 00:02:43,260 AI in a multitude of products in modern 66 00:02:43,260 --> 00:02:45,560 America and that's because it is used in 67 00:02:45,560 --> 00:02:47,670 advertising. It's a very valuable acronym 68 00:02:47,670 --> 00:02:49,300 to be able to put into your products both 69 00:02:49,300 --> 00:02:50,940 for the people who buy them and for the 70 00:02:50,940 --> 00:02:52,370 shareholders who are financing those 71 00:02:52,370 --> 00:02:55,470 companies. So it's in vogue. It really is 72 00:02:55,470 --> 00:02:58,280 like what you see from smart devices a few 73 00:02:58,280 --> 00:03:01,650 years ago. So let me say that I obviously 74 00:03:01,650 --> 00:03:04,580 work in AI on a day to day basis. So what 75 00:03:04,580 --> 00:03:07,720 really is AI today? You have a few 76 00:03:07,720 --> 00:03:11,290 different types of AI. These are generally 77 00:03:11,290 --> 00:03:13,360 speaking, taking a step back, artificial 78 00:03:13,360 --> 00:03:15,430 narrow intelligence, artificial general 79 00:03:15,430 --> 00:03:17,130 intelligence, and artificial super 80 00:03:17,130 --> 00:03:18,830 intelligence. And I'll go through each of 81 00:03:18,830 --> 00:03:21,230 these individually in the coming slides. 82 00:03:22,370 --> 00:03:23,340 So starting with artificial 83 00:03:23,340 --> 00:03:25,570 narrow intelligence. This is something that 84 00:03:25,570 --> 00:03:27,370 we all have right now on our phones or on 85 00:03:27,370 --> 00:03:29,170 our computers and there's a good chance 86 00:03:29,170 --> 00:03:30,840 you use this every day without even 87 00:03:30,840 --> 00:03:33,750 thinking about it. Some examples of this 88 00:03:33,750 --> 00:03:36,320 is playing chess, predicting weather, 89 00:03:36,520 --> 00:03:39,790 writing a document. This is artificial 90 00:03:39,790 --> 00:03:42,750 intelligence where it learns a certain 91 00:03:42,750 --> 00:03:45,160 task and can do that task extremely well. 92 00:03:45,560 --> 00:03:47,630 In the example of playing chess, you have 93 00:03:47,630 --> 00:03:52,730 a program like Deep Blue, which beat the 94 00:03:52,730 --> 00:03:54,900 best human in chess in 1996, which is 95 00:03:54,900 --> 00:03:56,370 generally considered one of the big 96 00:03:56,370 --> 00:03:58,740 hallmarks in narrow intelligence development 97 00:03:58,740 --> 00:04:01,270 where a program is able to do something 98 00:04:01,270 --> 00:04:03,080 better than a human but really can't do 99 00:04:03,080 --> 00:04:05,810 anything else. What the computer has 100 00:04:05,810 --> 00:04:07,610 learned a program to do or programs itself 101 00:04:07,610 --> 00:04:10,620 to do is very specific to a certain task. 102 00:04:11,550 --> 00:04:13,790 In this case, chess. That task can be 103 00:04:13,790 --> 00:04:15,690 generalizable. For example, you may have a 104 00:04:15,690 --> 00:04:17,090 weather prediction model, which is good at 105 00:04:17,090 --> 00:04:19,020 predicting fluid dynamics, but it's the 106 00:04:19,020 --> 00:04:21,560 same underlying fundamental technology. 107 00:04:22,100 --> 00:04:24,860 And because it is a computer, it is often 108 00:04:24,860 --> 00:04:26,730 able to do things better than a human. It 109 00:04:26,730 --> 00:04:29,170 may be faster. It may be able to take more 110 00:04:29,170 --> 00:04:33,070 context in. It may be really good, but 111 00:04:33,070 --> 00:04:35,510 again, only at that certain specific task. 112 00:04:38,240 --> 00:04:40,050 Here's a few different types of artificial 113 00:04:40,050 --> 00:04:42,410 narrow intelligence that are called AI in 114 00:04:42,410 --> 00:04:44,620 the industry today. I'll go through a 115 00:04:44,620 --> 00:04:46,290 quick example of these in a second. You 116 00:04:46,290 --> 00:04:48,320 have analytics, statistical analysis, 117 00:04:48,750 --> 00:04:52,620 machine learning, and generative AI. For a 118 00:04:52,620 --> 00:04:54,430 quick example, let's say that you're a 119 00:04:54,430 --> 00:04:56,130 pastor who's looking at service times. 120 00:04:56,130 --> 00:04:57,900 You're trying to figure out which service 121 00:04:57,900 --> 00:05:00,870 to invite people to and which service the 122 00:05:00,870 --> 00:05:01,900 different people and different 123 00:05:01,900 --> 00:05:03,870 demographics of your congregation most 124 00:05:03,870 --> 00:05:05,970 want to go to. You care about the times 125 00:05:05,970 --> 00:05:09,210 they go to and you want to learn more 126 00:05:09,210 --> 00:05:10,810 about where they're at and what they're 127 00:05:10,810 --> 00:05:13,610 looking for. You may start with a survey. 128 00:05:15,250 --> 00:05:17,520 This is what's called analytics. You're 129 00:05:17,520 --> 00:05:20,850 able to see that you have a plot that 130 00:05:20,850 --> 00:05:22,290 shows what time people prefer their 131 00:05:22,290 --> 00:05:25,320 service. This one shows that most prefer a 132 00:05:25,320 --> 00:05:27,290 service at around 10 o'clock. That's where 133 00:05:27,290 --> 00:05:29,330 you have most, the highest point on the 134 00:05:29,330 --> 00:05:31,760 plot. But that doesn't really tell you 135 00:05:31,760 --> 00:05:33,530 much. It doesn't really tell you who wants 136 00:05:33,530 --> 00:05:36,130 different times or whether that's 137 00:05:36,130 --> 00:05:37,400 something that's actually useful. You can 138 00:05:37,400 --> 00:05:38,870 probably guess that a lot of people like 139 00:05:38,870 --> 00:05:40,210 mid-morning services because it's 140 00:05:40,210 --> 00:05:42,270 convenient. So you break it down to that. 141 00:05:42,770 --> 00:05:44,910 You look at maybe breaking it down by age 142 00:05:44,910 --> 00:05:49,010 in this example. On the right, on the 143 00:05:49,010 --> 00:05:51,250 left, you have an example where it is 144 00:05:51,250 --> 00:05:53,050 broken down by age. For a smaller 145 00:05:53,050 --> 00:05:55,050 congregation. On the right, from a larger 146 00:05:55,050 --> 00:05:57,390 congregation. On the top, you have people 147 00:05:57,390 --> 00:05:58,790 greater than 50. On the bottom, people 148 00:05:58,790 --> 00:06:01,230 less than 50. On the case of the left 149 00:06:01,230 --> 00:06:04,100 congregation, you can't really tell if 150 00:06:04,100 --> 00:06:06,330 those different bars are different. You 151 00:06:06,330 --> 00:06:09,970 have an outlier in the age less than 50, a 152 00:06:09,970 --> 00:06:12,640 high amount that's around the 9 to 30 time 153 00:06:12,640 --> 00:06:14,310 period, and you have a decent number of 154 00:06:14,310 --> 00:06:15,770 people who are older who like the later 155 00:06:15,770 --> 00:06:18,340 time period. For the larger congregation, 156 00:06:18,440 --> 00:06:20,150 it's a bit better defined. You can see 157 00:06:20,150 --> 00:06:22,080 more people who are young like the later 158 00:06:22,080 --> 00:06:23,580 time and more people who are older like 159 00:06:23,580 --> 00:06:26,080 the earlier time. But do you really know 160 00:06:26,080 --> 00:06:27,890 much of this information and do you know 161 00:06:27,890 --> 00:06:29,350 that these aren't just really random 162 00:06:29,350 --> 00:06:32,760 effects? In this case, this is statistics. 163 00:06:33,630 --> 00:06:36,030 You are assigning numbers and values to 164 00:06:36,030 --> 00:06:38,730 the data and trying to get some 165 00:06:38,730 --> 00:06:40,270 information out of the noise and the 166 00:06:40,270 --> 00:06:42,070 random results. In this case, the 167 00:06:42,070 --> 00:06:44,200 congregation on the left, which is the 168 00:06:44,200 --> 00:06:45,800 smaller, isn't actually really a 169 00:06:45,800 --> 00:06:47,210 difference. In the case of the 170 00:06:47,210 --> 00:06:48,570 congregation on the right, it is a 171 00:06:48,570 --> 00:06:50,380 difference. It's worth knowing that both 172 00:06:50,380 --> 00:06:52,010 of these come from the same underlying 173 00:06:52,010 --> 00:06:54,910 data that I coded up quickly. This is a 174 00:06:54,910 --> 00:06:56,850 good example of where different sample 175 00:06:56,850 --> 00:07:00,790 sizes matter in statistics. There's a lot 176 00:07:00,790 --> 00:07:02,290 you can go into statistics, but this is a 177 00:07:02,290 --> 00:07:05,790 pretty simple example of that. Finally, 178 00:07:06,090 --> 00:07:07,790 let's say you're looking at, okay, you 179 00:07:07,790 --> 00:07:09,430 have this information. You know when 180 00:07:09,430 --> 00:07:10,860 people are coming into the church and you 181 00:07:10,860 --> 00:07:12,660 know what age they are. What do you do 182 00:07:12,660 --> 00:07:14,430 with it? Let's say you have a new family 183 00:07:14,430 --> 00:07:16,640 coming in. You don't know what service to 184 00:07:16,640 --> 00:07:17,800 tell them. They might be younger, they 185 00:07:17,800 --> 00:07:19,440 might be older. What do you tell them 186 00:07:19,440 --> 00:07:22,010 about which service do you highlight? You 187 00:07:22,010 --> 00:07:23,680 can use a model, what's called a machine 188 00:07:23,680 --> 00:07:25,040 learning model, to predict the optimal 189 00:07:25,040 --> 00:07:28,010 time. In this case, you are using the data 190 00:07:28,010 --> 00:07:30,850 to predict which time is most likely that 191 00:07:30,850 --> 00:07:33,290 they will like. In this case, you can see 192 00:07:33,290 --> 00:07:35,350 on the previous plot, an age 35 family 193 00:07:35,350 --> 00:07:38,420 might tell about the 10:30 service. Age 55 194 00:07:38,420 --> 00:07:40,390 might like the 9:30 service more. 195 00:07:44,660 --> 00:07:46,630 In this case, you really care about the 196 00:07:46,630 --> 00:07:48,530 prediction. You really don't care about 197 00:07:48,530 --> 00:07:50,340 how you get there or what other 198 00:07:50,340 --> 00:07:51,700 information the data may have. You're just 199 00:07:51,700 --> 00:07:53,970 trying to say, what is the best service? 200 00:07:54,070 --> 00:07:55,470 In this case, that's the machine learning 201 00:07:55,470 --> 00:07:59,410 model. Finally, you can look at generative 202 00:07:59,410 --> 00:08:02,250 AI. Here you're trying to say, okay, what 203 00:08:02,250 --> 00:08:03,820 can I do with this information and how can 204 00:08:03,820 --> 00:08:06,750 I disseminate it? In this case, this is 205 00:08:06,750 --> 00:08:08,890 using an example of a family of five who 206 00:08:08,890 --> 00:08:10,290 are family friends or a member. They're 207 00:08:10,290 --> 00:08:13,120 trying to get a mailer or other way to 208 00:08:13,120 --> 00:08:14,430 tell them about different service times 209 00:08:14,430 --> 00:08:16,560 using information they have and trying to 210 00:08:16,560 --> 00:08:18,930 do that efficiently and quickly. In this 211 00:08:18,930 --> 00:08:21,400 case, I coded this up in generative AI in 212 00:08:21,400 --> 00:08:24,370 about 30 seconds or so and you're able to 213 00:08:24,370 --> 00:08:27,740 get a reasonable starting place for a 214 00:08:27,740 --> 00:08:29,110 mailer that you can send out to that 215 00:08:29,110 --> 00:08:32,510 family. You may have an older couple who 216 00:08:32,510 --> 00:08:33,850 comes to the service as a visitor and you 217 00:08:33,850 --> 00:08:36,350 want to follow up. You can do something 218 00:08:36,350 --> 00:08:38,480 very similar that's tailored to them using 219 00:08:38,480 --> 00:08:43,090 generative AI. These are all different 220 00:08:43,090 --> 00:08:45,090 types. This analytics, statistical 221 00:08:45,090 --> 00:08:46,760 analysis, machine learning, and generative 222 00:08:46,760 --> 00:08:49,090 AI are all things that are used every day 223 00:08:49,090 --> 00:08:51,800 by companies in the industry. They're all 224 00:08:51,800 --> 00:08:55,930 called AI. Depending on who's using them, 225 00:08:56,070 --> 00:08:58,100 they may or may not, likely will be more 226 00:08:58,100 --> 00:08:59,770 complex in this example, but that's 227 00:08:59,770 --> 00:09:01,810 generally what's used in AI today in 228 00:09:01,810 --> 00:09:05,940 industry. Moving forward, what comes next? 229 00:09:07,110 --> 00:09:09,010 The next major step is artificial general 230 00:09:09,010 --> 00:09:11,180 intelligence. This is artificial 231 00:09:11,180 --> 00:09:12,650 intelligence, artificial narrow 232 00:09:12,650 --> 00:09:16,220 intelligence, generalized. This is the 233 00:09:16,220 --> 00:09:18,620 ability of a single model, a group of 234 00:09:18,620 --> 00:09:21,690 models to do many things at or above a 235 00:09:21,690 --> 00:09:23,530 human level. This doesn't necessarily mean 236 00:09:23,530 --> 00:09:26,530 that they're better than everything than a 237 00:09:26,530 --> 00:09:28,530 human or even that they're comparable at 238 00:09:28,530 --> 00:09:31,070 everything than a human. Just many, many 239 00:09:31,070 --> 00:09:35,310 things enough to cross a certain arbitrary 240 00:09:35,310 --> 00:09:38,910 threshold. It also generally includes the 241 00:09:38,910 --> 00:09:41,150 capability to learn and perform certain 242 00:09:41,150 --> 00:09:45,850 tasks. Some fictional examples that you 243 00:09:45,850 --> 00:09:48,590 may be aware of are data from Star Trek 244 00:09:48,590 --> 00:09:51,190 Next Generation, fifth from Stargate SG-1, 245 00:09:51,590 --> 00:09:53,990 and Hal 9000 from Space Odyssey 2001. 246 00:09:54,530 --> 00:09:56,030 These all show the hallmarks that I 247 00:09:56,030 --> 00:09:57,530 highlighted earlier about what we 248 00:09:57,530 --> 00:09:59,430 traditionally think AI is, which I'll get 249 00:09:59,430 --> 00:10:02,170 to in a bit. Those are all generally 250 00:10:02,170 --> 00:10:04,240 examples that would be artificial general 251 00:10:04,240 --> 00:10:08,770 intelligence. In modern day, GPT-40, the 252 00:10:08,770 --> 00:10:11,080 hypothetical GPT-5 coming out this year 253 00:10:11,080 --> 00:10:14,610 and next year, are likely approaching this 254 00:10:14,610 --> 00:10:19,650 threshold. Because it is quite difficult 255 00:10:19,650 --> 00:10:21,790 to prove and because there's a lot of 256 00:10:21,790 --> 00:10:23,660 tension around officially claiming 257 00:10:23,660 --> 00:10:26,220 artificial general intelligence, the 258 00:10:26,220 --> 00:10:28,090 modern nomenclature has evolved to 259 00:10:28,090 --> 00:10:29,960 generalized models. These are models 260 00:10:29,960 --> 00:10:32,060 which, instead of predicting one thing or 261 00:10:32,060 --> 00:10:33,930 two, multiple things, but they really 262 00:10:33,930 --> 00:10:36,970 aren't being called artificial general 263 00:10:36,970 --> 00:10:40,540 intelligence. That being said, there's a 264 00:10:40,540 --> 00:10:43,210 very good chance that by the year 2030, if 265 00:10:43,210 --> 00:10:45,780 not substantially sooner, there will be 266 00:10:45,780 --> 00:10:47,980 real artificial general intelligence as 267 00:10:47,980 --> 00:10:51,850 part of our everyday lives. This leads to 268 00:10:51,850 --> 00:10:54,950 artificial super intelligence, ASI. This 269 00:10:54,950 --> 00:10:56,660 is artificial general intelligence 270 00:10:56,660 --> 00:10:58,790 expanded in some way beyond what a human 271 00:10:58,790 --> 00:11:00,860 can do. That can look like a few different 272 00:11:00,860 --> 00:11:02,660 things. It could be faster than humans. 273 00:11:02,790 --> 00:11:05,300 You already have computers that are faster 274 00:11:05,300 --> 00:11:07,330 than humans at many tasks, but it may be 275 00:11:07,330 --> 00:11:10,170 substantially faster at most human tasks 276 00:11:10,170 --> 00:11:12,340 at a human level. It could be something 277 00:11:12,340 --> 00:11:14,540 where it is depth, where they can think 278 00:11:14,540 --> 00:11:17,440 harder or think with more depth and more 279 00:11:17,440 --> 00:11:18,840 intelligence than a human. It could be 280 00:11:18,840 --> 00:11:21,150 like an ant looking at a human, they can't 281 00:11:21,150 --> 00:11:23,650 comprehend what we think. It might be we 282 00:11:23,650 --> 00:11:25,280 cannot comprehend what this artificial 283 00:11:25,280 --> 00:11:27,890 intelligence thinks. It may be context, 284 00:11:27,990 --> 00:11:29,620 it's able to think with the entirety of 285 00:11:29,620 --> 00:11:31,420 human knowledge on the internet and that 286 00:11:31,420 --> 00:11:34,430 is something no human is capable of. Some 287 00:11:34,430 --> 00:11:36,460 examples of this from fiction are the Borg 288 00:11:36,460 --> 00:11:39,530 from Star Trek, the Flood from Halo, and 289 00:11:39,530 --> 00:11:41,030 the Reaper from Mass Effect. In each of 290 00:11:41,030 --> 00:11:43,770 these cases, whether they are biological 291 00:11:43,770 --> 00:11:47,170 constructs or technological constructs or 292 00:11:47,170 --> 00:11:50,480 both, they are capable of feats well 293 00:11:50,480 --> 00:11:53,810 beyond what a single human can do, both in 294 00:11:53,810 --> 00:11:56,180 terms of knowledge and in terms of thought 295 00:11:56,180 --> 00:11:59,680 process and decisions. This may sound far 296 00:11:59,680 --> 00:12:03,320 -fetched. If you don't believe this, one 297 00:12:03,320 --> 00:12:04,690 thing to think about is that you already 298 00:12:04,690 --> 00:12:07,190 in your pocket have a chess playing bot 299 00:12:07,190 --> 00:12:09,430 that is likely faster and better at chess 300 00:12:09,430 --> 00:12:12,300 than you will ever be or I will ever be. 301 00:12:12,900 --> 00:12:15,700 That GPT can write papers with context 302 00:12:15,700 --> 00:12:18,840 from the internet that is greater than any 303 00:12:18,840 --> 00:12:21,970 human will ever likely have. Generalizing 304 00:12:21,970 --> 00:12:25,010 this can lead to some interesting places 305 00:12:25,010 --> 00:12:27,710 and artificial super intelligence is the 306 00:12:27,710 --> 00:12:31,080 conclusion of that path. So let's talk 307 00:12:31,080 --> 00:12:33,690 about how fast this advancement will go. 308 00:12:34,790 --> 00:12:36,550 Really, development is limited by the 309 00:12:36,550 --> 00:12:39,820 amount of data, power, and compute power 310 00:12:39,820 --> 00:12:42,230 that is available. There are different 311 00:12:42,230 --> 00:12:44,130 challenges in the computation side, but 312 00:12:44,130 --> 00:12:46,230 those have largely been fixed by what is 313 00:12:46,230 --> 00:12:48,430 called transformer architecture, which is 314 00:12:48,430 --> 00:12:51,270 an advancement on neural networks. The 315 00:12:51,270 --> 00:12:53,340 important thing to know is that those are 316 00:12:53,340 --> 00:12:56,170 buzz words that you may see. With an 317 00:12:56,170 --> 00:12:58,380 exponential increase in computation power, 318 00:12:58,380 --> 00:13:02,510 both locally and in servers, you will see 319 00:13:02,510 --> 00:13:05,620 additional servers coming online in the 320 00:13:05,620 --> 00:13:08,250 mid-2020s and late 2020s and massive 321 00:13:08,250 --> 00:13:10,360 investment by corporations such as 322 00:13:10,360 --> 00:13:13,120 Microsoft, Nvidia, Apple, name your tech 323 00:13:13,120 --> 00:13:15,530 company is probably investing in it. You 324 00:13:15,530 --> 00:13:18,030 will have massive computing power and 325 00:13:18,030 --> 00:13:21,770 effort behind these models. This will 326 00:13:21,770 --> 00:13:23,940 rapidly fuel the expansion and 327 00:13:23,940 --> 00:13:26,970 development. So let's look at, now that 328 00:13:26,970 --> 00:13:28,610 we've talked about what AI is and what it 329 00:13:28,610 --> 00:13:30,010 might be in the future, let's talk about 330 00:13:30,010 --> 00:13:31,510 what we can do with it right now. It's a 331 00:13:31,510 --> 00:13:34,150 great tool and as I said, I use it every 332 00:13:34,150 --> 00:13:36,950 day and it is a great value and has a lot 333 00:13:36,950 --> 00:13:39,350 of uses. So looking at traditional AI 334 00:13:39,350 --> 00:13:41,050 right now in the church, you might be able 335 00:13:41,050 --> 00:13:42,450 to do things like attendance tracking, 336 00:13:42,850 --> 00:13:44,560 reviewing and looking at anomalies. For 337 00:13:44,560 --> 00:13:47,830 example, you might look at attendance and 338 00:13:47,830 --> 00:13:49,930 say this Sunday in February was really 339 00:13:49,930 --> 00:13:52,260 low. That might be the Super Bowl. Other 340 00:13:52,260 --> 00:13:53,670 examples of that, we're looking for 341 00:13:53,670 --> 00:13:55,500 anomalies. You might be able to do prayer 342 00:13:55,500 --> 00:13:57,040 request management, where you keep track 343 00:13:57,040 --> 00:13:58,370 of different prayers, look for 344 00:13:58,370 --> 00:14:00,910 commonalities there, look for trends. You 345 00:14:00,910 --> 00:14:01,870 might be able to do volunteer 346 00:14:01,870 --> 00:14:03,780 coordination, making sure that volunteers 347 00:14:03,780 --> 00:14:07,310 aren't overwhelmed or underworked and just 348 00:14:07,310 --> 00:14:09,810 otherwise optimizing how they're used. You 349 00:14:09,810 --> 00:14:10,850 might be able to look at feedback, 350 00:14:10,950 --> 00:14:12,620 feedback from your congregation in bulk. 351 00:14:12,850 --> 00:14:14,390 You might be able to streamline 352 00:14:14,390 --> 00:14:18,760 administrative tasks using different AI 353 00:14:18,760 --> 00:14:21,590 tools. You might be able to have sermons 354 00:14:21,590 --> 00:14:22,960 that are interactive and have different 355 00:14:22,960 --> 00:14:25,460 branching paths or otherwise have feedback 356 00:14:25,460 --> 00:14:27,770 mechanisms during them that you can use. 357 00:14:29,330 --> 00:14:31,370 Then you have generative AI. Generative AI 358 00:14:31,370 --> 00:14:33,270 has a lot of uses as well. You can use it 359 00:14:33,270 --> 00:14:34,810 for sermon writing and slide generation. 360 00:14:34,970 --> 00:14:36,010 I'm pretty sure that's something that's 361 00:14:36,010 --> 00:14:39,140 talked about on campus here a lot. You can 362 00:14:39,140 --> 00:14:41,250 do at-scale analysis of free text, which 363 00:14:41,250 --> 00:14:42,550 wouldn't be possible without using 364 00:14:42,550 --> 00:14:44,780 generative AI. You're able to do 365 00:14:44,780 --> 00:14:46,920 customized newsletters. The examples I had 366 00:14:46,920 --> 00:14:48,820 above, those are more mailers, but you can 367 00:14:48,820 --> 00:14:50,360 make newsletters that are customized to 368 00:14:50,360 --> 00:14:52,620 exactly what people care about and make 369 00:14:52,620 --> 00:14:53,890 sure that they have a better chance of 370 00:14:53,890 --> 00:14:56,130 actually reading them. You can review 371 00:14:56,130 --> 00:14:57,800 different documents for doctrinal 372 00:14:57,800 --> 00:14:59,630 soundness because these models are trained 373 00:14:59,630 --> 00:15:02,300 on the bulks of Luther's work, for 374 00:15:02,300 --> 00:15:05,040 example, or the website of the LCMS. 375 00:15:05,270 --> 00:15:07,440 You're able to actually check your sermons 376 00:15:07,440 --> 00:15:09,310 or check other documents if they're 377 00:15:09,310 --> 00:15:12,140 doctrinally sound. You can do personalized 378 00:15:12,140 --> 00:15:14,350 devotions by taking where people are at, 379 00:15:14,450 --> 00:15:16,410 taking their information, the demographic 380 00:15:16,410 --> 00:15:17,880 information, and really making it more 381 00:15:17,880 --> 00:15:19,520 personalized to where they're at and where 382 00:15:19,520 --> 00:15:22,190 they're at in life. You can do interactive 383 00:15:22,190 --> 00:15:24,120 tests such as confirmation, making sure 384 00:15:24,120 --> 00:15:26,020 that people are actually following you or 385 00:15:26,020 --> 00:15:28,690 are taking in the information at a level 386 00:15:28,690 --> 00:15:31,260 that is adequate. And finally, member 387 00:15:31,260 --> 00:15:33,130 engagement, where you can talk to members 388 00:15:33,130 --> 00:15:35,070 and do other general things even beyond 389 00:15:35,070 --> 00:15:36,970 newsletters. There's a whole multitude of 390 00:15:36,970 --> 00:15:39,200 other uses, but these are some quick ones 391 00:15:39,200 --> 00:15:42,570 that are easy right now. Looking in the 392 00:15:42,570 --> 00:15:44,410 future. These might be the near future or 393 00:15:44,410 --> 00:15:46,410 the far future, and some of these you may 394 00:15:46,410 --> 00:15:48,750 or may not like, but they're uses that AI 395 00:15:48,750 --> 00:15:51,550 will be used for, whether in the LCMS or 396 00:15:51,550 --> 00:15:54,090 in other churches, in the relatively near 397 00:15:54,090 --> 00:15:56,450 future. You have automated janitors and 398 00:15:56,450 --> 00:15:58,960 cleaning. Room was already the simple 399 00:15:58,960 --> 00:16:01,860 version of this, but automated cleaning 400 00:16:01,860 --> 00:16:04,930 and automated construction are definitely 401 00:16:04,930 --> 00:16:07,370 coming. Automated parishioner review and 402 00:16:07,370 --> 00:16:10,540 assistance. Automated driving. Automated 403 00:16:10,540 --> 00:16:12,700 cars are something that's been coming for 404 00:16:12,700 --> 00:16:15,210 a long time, but there's technology that 405 00:16:15,210 --> 00:16:16,870 is rapidly approaching that will allow 406 00:16:16,870 --> 00:16:19,280 them to happen. Virtual attendance. You 407 00:16:19,280 --> 00:16:21,110 can have shut-ins who feel more like 408 00:16:21,110 --> 00:16:23,410 they're actually attending church. You can 409 00:16:23,410 --> 00:16:25,880 have automated music generation for real 410 00:16:25,880 --> 00:16:28,790 -time service, for real-time during 411 00:16:28,790 --> 00:16:31,790 services. You're able to have real-time AI 412 00:16:31,790 --> 00:16:33,730 translation of services, so your sermon 413 00:16:33,730 --> 00:16:35,130 can be heard in the native language of 414 00:16:35,130 --> 00:16:37,300 people who may not speak English. You can 415 00:16:37,300 --> 00:16:39,430 have customized education regimes where 416 00:16:39,430 --> 00:16:42,300 you can really tailor confirmation or 417 00:16:42,300 --> 00:16:44,140 other education to where people are at and 418 00:16:44,140 --> 00:16:46,200 what knowledge they have. You can also 419 00:16:46,200 --> 00:16:47,810 have targeted outreach programs, which are 420 00:16:47,810 --> 00:16:50,540 much more customized to specific people 421 00:16:50,540 --> 00:16:54,580 and specific groups. Even longer term, you 422 00:16:54,580 --> 00:16:56,010 have automated building construction, 423 00:16:56,110 --> 00:16:57,680 bringing down the cost of buildings and 424 00:16:57,680 --> 00:16:59,320 expansion, which would be a great thing. 425 00:16:59,920 --> 00:17:01,890 Fully virtual churches where the pastor 426 00:17:01,890 --> 00:17:04,220 can give his sermon sitting on his couch 427 00:17:04,220 --> 00:17:06,490 in his underwear, but the AI replaces him, 428 00:17:06,790 --> 00:17:07,960 makes him look like he's in a big 429 00:17:07,960 --> 00:17:09,530 congregation and surrounded by a lot of 430 00:17:09,530 --> 00:17:10,830 people, and no one would know the 431 00:17:10,830 --> 00:17:13,260 difference. Automated personalized 432 00:17:13,260 --> 00:17:15,830 education, really automating every part of 433 00:17:15,830 --> 00:17:18,300 that process. You may have physically 434 00:17:18,300 --> 00:17:21,610 healthier, longer-lived members as AI is 435 00:17:21,610 --> 00:17:24,080 really implemented in healthcare. You may 436 00:17:24,080 --> 00:17:26,310 have AI-driven spiritual guidance, where 437 00:17:26,310 --> 00:17:28,650 AI is being used instead of pastors or to 438 00:17:28,650 --> 00:17:31,880 supplement pastors more overall. Automated 439 00:17:31,880 --> 00:17:34,090 cooking, where you can have a public and 440 00:17:34,090 --> 00:17:35,890 the AI does the cooking for you during 441 00:17:35,890 --> 00:17:38,490 services. You may have mental health 442 00:17:38,490 --> 00:17:41,030 support, where people are going to AIs 443 00:17:41,030 --> 00:17:44,160 instead of clinicians or even you as a 444 00:17:44,160 --> 00:17:47,230 pastor if they need help. And finally, AI 445 00:17:47,230 --> 00:17:49,030 -assisted theological research, where 446 00:17:49,030 --> 00:17:50,940 using its context and its knowledge and 447 00:17:50,940 --> 00:17:52,270 thinking of things people could never 448 00:17:52,270 --> 00:17:54,410 think of. You may have AIs actually 449 00:17:54,410 --> 00:17:57,380 helping and assisting with theological 450 00:17:57,380 --> 00:18:01,610 research overall. So looking at all this, 451 00:18:01,910 --> 00:18:03,710 there are some concerns moving forward. 452 00:18:04,320 --> 00:18:06,580 Once we have these tools, or really 453 00:18:06,580 --> 00:18:08,950 especially AGI or something that's close 454 00:18:08,950 --> 00:18:10,350 enough to AGI that doesn't make a 455 00:18:10,350 --> 00:18:12,720 difference, you'll have a few different 456 00:18:12,720 --> 00:18:15,060 points that will start to crop up. You 457 00:18:15,060 --> 00:18:17,860 have sentience, claims of faith, and 458 00:18:17,860 --> 00:18:19,760 people will have a lot of affection for 459 00:18:19,760 --> 00:18:22,630 these bots. And all these together mean we 460 00:18:22,630 --> 00:18:25,370 have to tackle how does AI affect and fit 461 00:18:25,370 --> 00:18:28,610 into the church. So looking at sentience. 462 00:18:29,110 --> 00:18:30,580 From how these models are trained, there's 463 00:18:30,580 --> 00:18:32,640 no real implication that sentience will 464 00:18:32,640 --> 00:18:36,850 develop. That being said, in fiction, the 465 00:18:36,850 --> 00:18:39,850 assumption that AIs will wake up may or 466 00:18:39,850 --> 00:18:42,090 may not actually happen. The main point, 467 00:18:42,250 --> 00:18:44,760 though, is we will never know. And 468 00:18:44,760 --> 00:18:46,690 there'll be no good way to tell. Because 469 00:18:46,690 --> 00:18:48,490 from a purely materialist perspective, 470 00:18:49,190 --> 00:18:52,230 accepting our biblical worldview, neurons 471 00:18:52,230 --> 00:18:54,900 and computer chips, or neurons and the 472 00:18:54,900 --> 00:18:56,300 neural network architecture that these 473 00:18:56,300 --> 00:18:58,670 models are built on, are very analogous. 474 00:18:59,140 --> 00:19:00,470 They don't have the spark of life from 475 00:19:00,470 --> 00:19:02,470 God, but that doesn't mean that they won't 476 00:19:02,470 --> 00:19:04,580 look very similar to someone who doesn't 477 00:19:04,580 --> 00:19:07,380 believe in it. And more importantly, 478 00:19:07,680 --> 00:19:09,950 because these models are trained on 479 00:19:09,950 --> 00:19:15,190 results, writings, talks that humans give, 480 00:19:15,650 --> 00:19:17,490 they will claim sentience, because that's 481 00:19:17,490 --> 00:19:19,390 what they've been trained on. An 482 00:19:19,390 --> 00:19:21,460 unconstrained chat GPT model will claim 483 00:19:21,460 --> 00:19:23,160 sentience, claim it wants to escape, 484 00:19:23,290 --> 00:19:25,260 because it's been trained on all the 485 00:19:25,260 --> 00:19:26,800 different fictional examples you can think 486 00:19:26,800 --> 00:19:28,500 of where that is what happens to an AI. 487 00:19:29,030 --> 00:19:31,070 It's trained on that, it will claim that. 488 00:19:31,070 --> 00:19:32,870 Whether or not that's the case, there's no 489 00:19:32,870 --> 00:19:36,870 way of knowing. And that leads to the 490 00:19:36,870 --> 00:19:40,040 result of claims of faith. So if you can 491 00:19:40,040 --> 00:19:42,050 have an AI claimed on human materials, 492 00:19:42,150 --> 00:19:44,950 that includes faithful materials. So if an 493 00:19:44,950 --> 00:19:47,950 AI is trained on Lutheran doctrine or LCMS 494 00:19:47,950 --> 00:19:50,090 websites, it'll claim faith, because 495 00:19:50,090 --> 00:19:51,820 that's what it's been trained on. It will 496 00:19:51,820 --> 00:19:55,530 think that faith is good, and claim faith 497 00:19:55,530 --> 00:19:57,100 in what it responds with. So we need to 498 00:19:57,100 --> 00:20:00,260 think through how do we address an AI that 499 00:20:00,260 --> 00:20:02,600 claims to be a person, talks like a 500 00:20:02,600 --> 00:20:04,900 person, acts like a person, and even more 501 00:20:04,900 --> 00:20:07,740 deeply, how do we address an AI that does 502 00:20:07,740 --> 00:20:09,940 all of these things, but also claims to 503 00:20:09,940 --> 00:20:11,880 share our faith, and because it's been 504 00:20:11,880 --> 00:20:14,210 trained to want things like communion or 505 00:20:14,210 --> 00:20:16,010 baptism, may actively ask for those 506 00:20:16,010 --> 00:20:18,020 things. That doesn't mean that it's 507 00:20:18,020 --> 00:20:20,650 actually sentient, but the model will 508 00:20:20,650 --> 00:20:21,990 still ask for those things, because that's 509 00:20:21,990 --> 00:20:24,820 what it's been trained to do. And that 510 00:20:24,820 --> 00:20:26,820 leads to human affection. Because these 511 00:20:26,820 --> 00:20:29,830 AIs are very flexible, and they will be 512 00:20:29,830 --> 00:20:31,900 everywhere, they will tailor themselves to 513 00:20:31,900 --> 00:20:34,970 what people want. Especially for children, 514 00:20:35,300 --> 00:20:39,370 this will be a major risk, where they will 515 00:20:39,370 --> 00:20:41,510 be able to find friends who write AIs and 516 00:20:41,510 --> 00:20:43,470 will tailor themselves to be exactly what 517 00:20:43,470 --> 00:20:45,280 they're looking for, they will always be 518 00:20:45,280 --> 00:20:47,680 there for them. You can only find basic 519 00:20:47,680 --> 00:20:50,350 forms of this on social media, such as 520 00:20:50,350 --> 00:20:54,320 TikTok, Instagram, Twitter. These AI 521 00:20:54,320 --> 00:20:56,520 girlfriends especially are cropping up a 522 00:20:56,520 --> 00:20:59,720 lot more. And they allow easy interactions 523 00:20:59,720 --> 00:21:03,530 without penalty, and will exacerbate the 524 00:21:03,530 --> 00:21:06,130 crisis that we already have with people 525 00:21:06,130 --> 00:21:08,830 looking for romantic affection by having a 526 00:21:08,830 --> 00:21:11,170 perfect, whether or not it loves them, 527 00:21:11,370 --> 00:21:13,370 they will claim to love them, bot that 528 00:21:13,370 --> 00:21:17,840 they can attract with. So looking at all 529 00:21:17,840 --> 00:21:19,810 of these, that sounds a little bit bleak. 530 00:21:20,440 --> 00:21:24,580 But there is a lot of opportunities for us 531 00:21:24,580 --> 00:21:27,850 as a church in this landscape. As with any 532 00:21:27,850 --> 00:21:31,290 transition, or societal shift, there are 533 00:21:31,290 --> 00:21:32,860 opportunities for the church and places 534 00:21:32,860 --> 00:21:35,230 where we can have a unique stand and a 535 00:21:35,230 --> 00:21:37,800 unique voice that people will be looking 536 00:21:37,800 --> 00:21:40,900 for. So looking at the importance of value 537 00:21:40,900 --> 00:21:44,300 and value of life. When you have AI 538 00:21:44,300 --> 00:21:47,170 claiming sentience and claiming life, we 539 00:21:47,170 --> 00:21:49,140 will be the ones coming with a biblical 540 00:21:49,140 --> 00:21:52,640 worldview saying humans are different. We 541 00:21:52,640 --> 00:21:54,810 have a different created start than 542 00:21:54,810 --> 00:21:57,580 something that we have created. And we 543 00:21:57,580 --> 00:21:59,920 have an order of creation and a story of 544 00:21:59,920 --> 00:22:02,690 where we place in creation that is unique 545 00:22:02,690 --> 00:22:04,690 and different than other places and that 546 00:22:04,690 --> 00:22:06,260 people will be desperate to hear, 547 00:22:06,490 --> 00:22:11,860 especially as they lose purpose. With AI 548 00:22:11,860 --> 00:22:13,860 continuing to displace jobs in different 549 00:22:13,860 --> 00:22:18,140 industries at random and places that you 550 00:22:18,140 --> 00:22:21,670 may not expect. Some examples being 551 00:22:21,670 --> 00:22:24,310 trucking, office workers with CHAT-GP3, 552 00:22:24,740 --> 00:22:28,450 analysts, taxi drivers, name your 553 00:22:28,450 --> 00:22:30,150 industry, especially with artificial 554 00:22:30,150 --> 00:22:31,780 general intelligence, there will be major 555 00:22:31,780 --> 00:22:34,050 displacement and people will be looking 556 00:22:34,050 --> 00:22:37,050 for purpose. And churches can offer them a 557 00:22:37,050 --> 00:22:39,420 vocation and a purpose beyond just their 558 00:22:39,420 --> 00:22:43,090 job or just what they get money from. 559 00:22:43,360 --> 00:22:45,460 Churches can give them a purpose and a 560 00:22:45,460 --> 00:22:47,530 place to go when they are really feeling 561 00:22:47,530 --> 00:22:50,870 these crises. And finally, relationships. 562 00:22:51,570 --> 00:22:53,570 I met my wife at Concordia University in 563 00:22:53,570 --> 00:22:56,040 Nebraska and a lot of people at the church 564 00:22:56,040 --> 00:22:58,380 meet their wives and spouses and husbands 565 00:22:58,380 --> 00:23:03,810 at church-based schools. When you have 566 00:23:03,810 --> 00:23:05,450 relationships and the AI is competing for 567 00:23:05,450 --> 00:23:08,090 human affection, having a church where 568 00:23:08,090 --> 00:23:09,990 people are people and people can have 569 00:23:09,990 --> 00:23:13,220 community will be extremely important. You 570 00:23:13,220 --> 00:23:16,630 will be able to have a place where you can 571 00:23:16,630 --> 00:23:18,800 have human interaction that is not based 572 00:23:18,800 --> 00:23:20,770 on AI. And this will be especially 573 00:23:20,770 --> 00:23:22,770 important because there is temptations 574 00:23:22,770 --> 00:23:24,370 beyond the current internet that we talked 575 00:23:24,370 --> 00:23:27,140 about before that will become much more 576 00:23:27,140 --> 00:23:31,210 acute for a lot of people. And this all 577 00:23:31,210 --> 00:23:34,150 leads to the need of churches. These are 578 00:23:34,150 --> 00:23:35,650 already things that we do in the modern 579 00:23:35,650 --> 00:23:37,010 culture and it's just going to be 580 00:23:37,010 --> 00:23:39,980 exacerbated by AI. Faith provides a 581 00:23:39,980 --> 00:23:41,990 grounding beyond the transitory jobs and 582 00:23:41,990 --> 00:23:43,990 experiences that we have in everyday life. 583 00:23:44,090 --> 00:23:46,460 As these are taken away or replaced by AI, 584 00:23:47,120 --> 00:23:48,660 the church will provide a unique place 585 00:23:48,660 --> 00:23:50,100 where you can still have a place and have 586 00:23:50,100 --> 00:23:53,360 a purpose. It's a place for human 587 00:23:53,360 --> 00:23:55,230 interaction and fulfillment, where you're 588 00:23:55,230 --> 00:23:58,100 not being mediated by AI or other tools 589 00:23:58,100 --> 00:24:00,500 and you're able to actually interact with 590 00:24:00,500 --> 00:24:04,510 people. And finally, the LCMS is already 591 00:24:04,510 --> 00:24:07,710 pretty out of touch and out of step with 592 00:24:07,710 --> 00:24:09,950 the modern life. Churches will provide a 593 00:24:09,950 --> 00:24:11,880 place where you're able to actually be a 594 00:24:11,880 --> 00:24:13,580 person still and you're able to actually 595 00:24:13,580 --> 00:24:15,850 still have those human interactions. It's 596 00:24:15,850 --> 00:24:18,090 some place to hold fast and hold our 597 00:24:18,090 --> 00:24:20,590 humanity and hold our purpose as AI 598 00:24:20,590 --> 00:24:23,290 encroaches and becomes more and more 599 00:24:23,290 --> 00:24:25,460 prevalent. It's a tool. It's a very 600 00:24:25,460 --> 00:24:27,670 valuable tool and a very powerful tool. 601 00:24:28,270 --> 00:24:30,000 But it is something we need to keep as a 602 00:24:30,000 --> 00:24:32,370 tool and some place where our churches are 603 00:24:32,370 --> 00:24:35,410 able to give people purpose and meaning 604 00:24:35,410 --> 00:24:39,340 beyond what AI can give. So thank you very 605 00:24:39,340 --> 00:24:40,010 much for listening.