WEBVTT 1 00:00:00.709 --> 00:00:03.819 Many or even most of you have already started thinking 2 00:00:03.819 --> 00:00:08.659 about how AI or artificial intelligence is impacting dietetics. 3 00:00:08.819 --> 00:00:11.449 There have been many changes in the past couple of years, 4 00:00:11.630 --> 00:00:16.309 but this is only a taste of the many changes that will be occurring in the near future. 5 00:00:16.700 --> 00:00:18.309 So before you start your assignment, 6 00:00:18.420 --> 00:00:21.620 this brief presentation will ensure that we're all starting 7 00:00:21.620 --> 00:00:25.059 with the same basic information and understanding about AI. 8 00:00:27.090 --> 00:00:29.799 So this short presentation will talk about the 9 00:00:29.799 --> 00:00:33.270 future roles of dietitians in an AI world. 10 00:00:33.479 --> 00:00:36.750 It'll include a quick update on the history of AI. 11 00:00:37.139 --> 00:00:39.159 We'll take a little glimpse into the future 12 00:00:39.159 --> 00:00:41.919 or the current AI applications and dietetics, 13 00:00:41.959 --> 00:00:44.549 and you can dream about what might be coming in the future, 14 00:00:44.759 --> 00:00:48.799 and then we'll spend a few moments talking about an overview of your assignment. 15 00:00:50.259 --> 00:00:52.759 And you've likely already been thinking about the 16 00:00:52.759 --> 00:00:55.509 future roles of dieticians in an AI world. 17 00:00:56.159 --> 00:00:57.049 In example, 18 00:00:57.209 --> 00:01:00.759 how can you use AI to streamline your own practice, 19 00:01:01.130 --> 00:01:02.599 or how can you 20 00:01:02.930 --> 00:01:07.330 help clients and guide them to select AI applications that have 21 00:01:07.569 --> 00:01:10.809 accurate information and complete information in their 22 00:01:10.809 --> 00:01:12.970 own nutrition care for the clients, 23 00:01:13.250 --> 00:01:16.529 and then how can you participate with AI developers 24 00:01:16.529 --> 00:01:20.120 to incorporate that accurate information and thought processes. 25 00:01:21.150 --> 00:01:23.169 So AI isn't new at all, 26 00:01:23.459 --> 00:01:27.819 but the application of AI is changing quickly and exponentially. 27 00:01:28.230 --> 00:01:31.669 AI was first emerged in the 1950s, 28 00:01:31.790 --> 00:01:35.180 and you've probably been using AI without really knowing what it was 29 00:01:35.430 --> 00:01:38.309 when you have automated phone prompts or if you 30 00:01:38.309 --> 00:01:41.059 have automated helpbots that help you find things. 31 00:01:41.690 --> 00:01:43.449 From 2012 to 2017, 32 00:01:43.529 --> 00:01:45.529 there was a great deal of interest in the medical 33 00:01:45.529 --> 00:01:49.199 field in terms of having computers help us interpret results. 34 00:01:49.529 --> 00:01:53.319 One of the classic example would be looking at X-ray reports. 35 00:01:53.489 --> 00:01:55.769 So they had a large number of X-ray reports 36 00:01:55.769 --> 00:01:59.239 that were labeled as either abnormal or normal, 37 00:01:59.690 --> 00:02:03.879 and then they Fed those into the computer and then ask the computer then 38 00:02:03.879 --> 00:02:08.130 to learn to give the report based on the computer reading of those images 39 00:02:08.410 --> 00:02:11.008 and then they used a different set of X-rays 40 00:02:11.250 --> 00:02:14.759 that were also labeled to see if the computer could get it right. 41 00:02:14.889 --> 00:02:19.199 Did they identify the right X-rays with abnormal versus normal? 42 00:02:19.869 --> 00:02:22.089 And from 2017 to present though, 43 00:02:22.220 --> 00:02:24.809 we have now what we call unsupervised learning, 44 00:02:24.979 --> 00:02:27.809 whereas we don't have to have a lot of labeled 45 00:02:27.809 --> 00:02:31.169 results for the computer to learn how to interpret things. 46 00:02:31.500 --> 00:02:34.210 And the recent release of the large language models, 47 00:02:34.279 --> 00:02:34.729 for example, 48 00:02:34.940 --> 00:02:36.029 chat GPT, 49 00:02:36.220 --> 00:02:38.169 have really been game changers, 50 00:02:38.539 --> 00:02:42.300 but we do know that these models are known to make mistakes. 51 00:02:42.505 --> 00:02:43.485 So for example, 52 00:02:43.535 --> 00:02:48.235 if you use the AI companion for example in Zoom or Otter AI 53 00:02:48.485 --> 00:02:52.634 to listen to meeting minutes or meeting recordings 54 00:02:52.804 --> 00:02:54.514 and then create minutes from them, 55 00:02:54.725 --> 00:02:56.244 we know that the zoom will often 56 00:02:56.244 --> 00:02:59.884 misinterpret the information or will lack the context 57 00:02:59.884 --> 00:03:04.804 of what is being said as it tries to interpret the recording into minutes. 58 00:03:05.429 --> 00:03:08.360 We also know that hallucinations can occur. 59 00:03:08.649 --> 00:03:09.779 So for example, 60 00:03:09.889 --> 00:03:10.089 you know, 61 00:03:10.289 --> 00:03:11.199 since we know 62 00:03:11.449 --> 00:03:17.119 that AI is predicting what is likely to be the next word or the next answer, 63 00:03:17.320 --> 00:03:21.460 that they can hallucinate and identify that if these authors 64 00:03:21.460 --> 00:03:26.559 have previously recorded or published research on a similar topic, 65 00:03:26.729 --> 00:03:30.850 that they are also likely to have published something on the research topic that. 66 00:03:30.955 --> 00:03:32.214 You are interested in 67 00:03:32.425 --> 00:03:33.054 and so 68 00:03:33.335 --> 00:03:38.785 the chat GPT will create a whole citation and it has a citation, 69 00:03:38.865 --> 00:03:39.755 has a journal number, 70 00:03:40.024 --> 00:03:40.475 issue number, 71 00:03:40.505 --> 00:03:41.134 page number, 72 00:03:41.145 --> 00:03:41.934 and everything 73 00:03:42.264 --> 00:03:43.554 that doesn't exist 74 00:03:44.255 --> 00:03:48.425 because it is assuming that these authors who have published in this 75 00:03:48.425 --> 00:03:50.744 journal and published on very close 76 00:03:50.744 --> 00:03:54.014 similar topics would also have published something 77 00:03:54.494 --> 00:03:56.384 on the topic that you're interested in. 78 00:03:58.320 --> 00:04:00.059 There are a couple of articles that have 79 00:04:00.059 --> 00:04:03.910 summarized the examples of AI applications into dietetics. 80 00:04:03.949 --> 00:04:07.860 One of them is on this slide and the other one is in the acknowledgements at the end. 81 00:04:08.149 --> 00:04:09.949 But there are a number of examples. 82 00:04:10.149 --> 00:04:11.000 So for example, 83 00:04:11.110 --> 00:04:14.320 there are AI apps that will log foods that are 84 00:04:14.320 --> 00:04:17.190 eaten by looking at a single photo of the meals, 85 00:04:17.309 --> 00:04:17.760 and those, 86 00:04:17.790 --> 00:04:18.339 for example, 87 00:04:18.428 --> 00:04:23.390 are MyFitnessPal Lose It or Snap Calorie or RX Food. 88 00:04:23.820 --> 00:04:26.640 There are also chatbots that will allow the users, 89 00:04:27.209 --> 00:04:28.089 usually patients, 90 00:04:28.130 --> 00:04:31.220 to ask questions and get answers real time. 91 00:04:31.769 --> 00:04:33.890 Those apps will be things like Snap Calorie 92 00:04:33.890 --> 00:04:37.450 or there's ongoing client support examples in diabetes, 93 00:04:37.570 --> 00:04:38.239 for example, 94 00:04:38.250 --> 00:04:38.809 for example, 95 00:04:38.890 --> 00:04:40.179 Milton AI, 96 00:04:40.549 --> 00:04:44.049 which was introduced to dieticians but is primarily being 97 00:04:44.049 --> 00:04:46.890 used now by case managers for for diabetes. 98 00:04:47.730 --> 00:04:49.730 They also those chatbots can use 99 00:04:49.730 --> 00:04:53.320 medical information and make personalized recommendations, 100 00:04:53.649 --> 00:04:56.170 so it goes beyond just answering questions, 101 00:04:56.179 --> 00:04:57.279 but for example, 102 00:04:57.450 --> 00:04:58.519 in diabetes, 103 00:04:58.690 --> 00:04:59.320 they can, 104 00:04:59.410 --> 00:05:03.130 there is an app that will interpret the blood glucose levels and then make 105 00:05:03.130 --> 00:05:05.640 food choice recommendations to a client based 106 00:05:05.640 --> 00:05:07.690 on their current blood glucose levels. 107 00:05:07.809 --> 00:05:11.489 So these are just a few examples of AI in dietetics. 108 00:05:12.820 --> 00:05:17.359 It's easy to get excited about the possibility of using these AI applications, 109 00:05:17.489 --> 00:05:19.929 but we do need to have a bit of understanding about 110 00:05:19.929 --> 00:05:23.299 how the chat GPT and the large language models work. 111 00:05:23.609 --> 00:05:27.929 The AI models learn based on the data that's provided to it, 112 00:05:28.410 --> 00:05:32.529 and biases may be built into the what they call training data. 113 00:05:32.970 --> 00:05:32.980 Um, 114 00:05:33.640 --> 00:05:33.799 they, 115 00:05:33.839 --> 00:05:34.399 for example, 116 00:05:34.440 --> 00:05:39.709 can be primarily all of one gender or they can lack racial diversity 117 00:05:39.920 --> 00:05:43.399 or they can only represent one culture and so these 118 00:05:43.399 --> 00:05:47.709 biases can be problematic when you're asking them to provide recommendations 119 00:05:47.920 --> 00:05:51.519 to individuals that were different than what was included in the training data. 120 00:05:52.709 --> 00:05:57.790 Some of the earlier chatbots were trained using all of the internet data, 121 00:05:57.839 --> 00:06:01.799 and we know that not all internet data is completely accurate. 122 00:06:02.190 --> 00:06:03.149 So for example, 123 00:06:03.190 --> 00:06:07.600 the early free chat GPT that you'll be using in your assignment was 124 00:06:07.600 --> 00:06:11.820 trained on data that was available on the internet up to 2000. 125 00:06:11.910 --> 00:06:16.600 So you know that there is a problem with currency because the data ended in 2000, 126 00:06:16.709 --> 00:06:19.510 but there's also the understanding that everything 127 00:06:19.510 --> 00:06:22.190 on the internet isn't always completely accurate. 128 00:06:22.649 --> 00:06:25.630 And these models are continually learning and improving, 129 00:06:25.880 --> 00:06:27.829 so the results will change over time. 130 00:06:28.040 --> 00:06:29.160 When I ran a question, 131 00:06:29.170 --> 00:06:30.950 the same question in July 132 00:06:31.200 --> 00:06:33.029 and then ran it again in December, 133 00:06:33.230 --> 00:06:37.320 the chatbot had learned more between July and December, 134 00:06:37.359 --> 00:06:39.000 and so the response will differ. 135 00:06:39.799 --> 00:06:41.940 And and to go along with this, 136 00:06:42.230 --> 00:06:46.500 remember that the understanding is that chat GPT is about predicting 137 00:06:46.790 --> 00:06:48.529 what is likely to come next, 138 00:06:48.630 --> 00:06:51.739 what word or concept is likely to be in the next sentence 139 00:06:51.950 --> 00:06:55.149 based on all the knowledge that it has learned 140 00:06:55.149 --> 00:06:57.540 from the training database that it was provided. 141 00:06:57.750 --> 00:07:01.700 So it's predicting what people are likely to say about a particular issue. 142 00:07:03.260 --> 00:07:04.329 When using AI, 143 00:07:04.380 --> 00:07:06.260 there are two things that are really critical, 144 00:07:06.299 --> 00:07:07.850 and that's transparency 145 00:07:08.059 --> 00:07:09.410 and accountability. 146 00:07:09.700 --> 00:07:14.200 So let's talk for a moment about best practice and adding acknowledgements. 147 00:07:14.339 --> 00:07:15.829 Whenever you use AI, 148 00:07:15.880 --> 00:07:18.940 if it's for creating the initial draft gathering. 149 00:07:19.325 --> 00:07:20.785 in whatever way you've used it, 150 00:07:20.795 --> 00:07:22.734 you need to acknowledge that use 151 00:07:22.945 --> 00:07:23.855 of AI, 152 00:07:24.144 --> 00:07:27.575 so acknowledge how it was used and also what version. 153 00:07:27.864 --> 00:07:28.545 So for example, 154 00:07:28.625 --> 00:07:32.144 you might say something like chat GPT 3 was 155 00:07:32.144 --> 00:07:35.035 used to create a draft of the trailered outline. 156 00:07:35.720 --> 00:07:38.119 And you also need to acknowledge 157 00:07:38.380 --> 00:07:40.940 that the content was then reviewed and 158 00:07:40.940 --> 00:07:44.660 approved by yourself with your professional credentials. 159 00:07:44.820 --> 00:07:45.529 So for example, 160 00:07:45.700 --> 00:07:48.140 you'd say content of this handout was reviewed for 161 00:07:48.140 --> 00:07:51.890 accuracy and approved by a registered dietitian slash nutritionist, 162 00:07:52.260 --> 00:07:55.730 your signature and name and your CDR registration number, 163 00:07:55.739 --> 00:07:59.290 or you could say acknowledge the types of modifications you made. 164 00:07:59.339 --> 00:07:59.980 You might say. 165 00:08:00.230 --> 00:08:01.619 Content was reviewed, 166 00:08:02.029 --> 00:08:02.910 content modified, 167 00:08:03.070 --> 00:08:04.500 new content added, 168 00:08:04.630 --> 00:08:08.190 and the final document was approved by again your name, 169 00:08:08.309 --> 00:08:09.239 signature block, 170 00:08:09.329 --> 00:08:11.179 and your CDR reference number 171 00:08:11.510 --> 00:08:16.339 because if you are providing any document that was created in any way with AI, 172 00:08:16.510 --> 00:08:18.079 you need to acknowledge that 173 00:08:18.309 --> 00:08:21.429 and you also need to acknowledge that you have approved 174 00:08:21.429 --> 00:08:24.230 it and are accountable for the content that's included. 175 00:08:26.630 --> 00:08:30.140 So now let's talk about the objectives of this short AI assignment. 176 00:08:30.390 --> 00:08:34.630 You're going to use a free chat GPT to create an outline handout 177 00:08:34.869 --> 00:08:35.849 for your client. 178 00:08:36.150 --> 00:08:40.630 You're going to determine the type of information to use in chat GPT prompts. 179 00:08:40.789 --> 00:08:42.429 You're going to compare the results from 180 00:08:42.429 --> 00:08:45.260 two different levels of specificity in prompts, 181 00:08:45.549 --> 00:08:48.520 then evaluate and modify the AI product 182 00:08:48.830 --> 00:08:51.330 and add acknowledgements that AI was a used, 183 00:08:51.590 --> 00:08:52.140 approved, 184 00:08:52.190 --> 00:08:54.750 and or modified by a qualified professional. 185 00:08:56.429 --> 00:08:57.489 So for your assignment, 186 00:08:57.530 --> 00:09:00.719 the first thing you need to do is to download the handout, 187 00:09:00.929 --> 00:09:05.080 and we're going to be using a free chat GPT for this assignment. 188 00:09:06.450 --> 00:09:10.429 You'll use the simple prompt that is included in your handout 189 00:09:10.710 --> 00:09:13.260 and upload that into the AI. 190 00:09:13.869 --> 00:09:15.659 When you get the resulting handout, 191 00:09:15.789 --> 00:09:20.830 then copy and paste that handout into the place on the template. 192 00:09:22.250 --> 00:09:25.690 Then you're going to use a more complete and specific prompt 193 00:09:25.690 --> 00:09:29.440 to create that same one page handout but with more information. 194 00:09:29.729 --> 00:09:30.239 Again, 195 00:09:30.369 --> 00:09:33.250 that complex prompt is included in your handout, 196 00:09:33.330 --> 00:09:37.690 so you copy and paste that into the GPT instructing it 197 00:09:37.690 --> 00:09:41.330 to create a one page handout with this additional information. 198 00:09:41.950 --> 00:09:43.229 After you've got the handout, 199 00:09:43.309 --> 00:09:47.900 you can also use prompts to ask chat GPT to modify that handout. 200 00:09:48.070 --> 00:09:48.820 You might say, 201 00:09:48.989 --> 00:09:53.510 Give me more information about this or include more current citations or 202 00:09:53.510 --> 00:09:57.369 whatever modifications you want the chat GPT to make to it. 203 00:09:57.710 --> 00:09:59.919 When you have got it the way you think you want it, 204 00:10:00.190 --> 00:10:02.869 then copy and paste that answer and 205 00:10:03.159 --> 00:10:06.460 And dra put it into the template saying this is 206 00:10:06.460 --> 00:10:09.330 the response I got when I used a more sophisticated 207 00:10:09.539 --> 00:10:13.539 and more complete prompt to create a one-page handout. 208 00:10:14.479 --> 00:10:17.400 Then you'll compare the two results between the first time 209 00:10:17.400 --> 00:10:20.190 when you used a very simple prompt and the second time 210 00:10:20.400 --> 00:10:22.359 to see what kind of differences there are and 211 00:10:22.359 --> 00:10:25.500 observe and record your observations for your instructor. 212 00:10:26.530 --> 00:10:29.880 Then conduct a full evaluation of that second handout 213 00:10:30.169 --> 00:10:31.710 because that's the one we're going to pretend 214 00:10:31.710 --> 00:10:33.520 that you're going to use for your patient. 215 00:10:33.830 --> 00:10:35.950 Now see the instructions in the 216 00:10:36.969 --> 00:10:41.900 template for assignment that explain what we mean by a full evaluation. 217 00:10:42.049 --> 00:10:45.289 It means you have to double check each and every citation 218 00:10:45.289 --> 00:10:48.440 to make sure it's a real citation and that it's accurate. 219 00:10:48.609 --> 00:10:49.679 It has the right. 220 00:10:50.965 --> 00:10:51.525 You number, 221 00:10:51.534 --> 00:10:52.385 the right journal, 222 00:10:52.474 --> 00:10:53.424 the right date, 223 00:10:53.635 --> 00:10:55.684 that all that stuff is accurate, 224 00:10:55.914 --> 00:10:57.755 as well as looking at whether the 225 00:10:57.755 --> 00:11:01.065 content is consistent with current evidence-based guidelines, 226 00:11:01.155 --> 00:11:03.875 you'll need to refer back to your perspective or back to the 227 00:11:03.875 --> 00:11:08.184 resources that are included on your resource page because you are responsible 228 00:11:08.434 --> 00:11:13.385 to make sure that everything on that page is 100% accurate as a professional 229 00:11:13.724 --> 00:11:15.474 because you're going to give it to your client. 230 00:11:16.070 --> 00:11:17.700 When you've got it the way you want it, 231 00:11:17.869 --> 00:11:20.619 then also add the acknowledgements and the approval, 232 00:11:20.820 --> 00:11:23.179 and then copy that final handout that you would 233 00:11:23.179 --> 00:11:25.789 be giving to the patient into the template. 234 00:11:27.059 --> 00:11:29.700 Then you're going to summarize what types of changes you 235 00:11:29.700 --> 00:11:33.200 ended up having to make to the chat GPT product, 236 00:11:33.539 --> 00:11:33.690 not, 237 00:11:33.820 --> 00:11:34.450 not the, 238 00:11:34.619 --> 00:11:35.340 not the prompts, 239 00:11:35.380 --> 00:11:37.590 but after you looked at that final handout, 240 00:11:37.739 --> 00:11:39.260 what changes did you make in it? 241 00:11:40.909 --> 00:11:45.010 You may also want to look at the NCCPR policy 242 00:11:45.179 --> 00:11:46.250 on AI, 243 00:11:46.500 --> 00:11:47.780 and most organizations, 244 00:11:48.059 --> 00:11:50.169 including your educational institution, 245 00:11:50.500 --> 00:11:54.719 likely have policies and procedures for how or if and when 246 00:11:54.859 --> 00:11:56.700 you can use an AI product. 247 00:11:59.239 --> 00:12:02.950 These are the acknowledgements of the resources that I used for this presentation. 248 00:12:03.179 --> 00:12:05.840 I did not use AI to produce the presentation. 249 00:12:05.919 --> 00:12:06.270 However, 250 00:12:06.330 --> 00:12:06.950 I used 251 00:12:07.190 --> 00:12:11.080 the navigating future practice technology that's available from the House 252 00:12:11.080 --> 00:12:14.830 of Delegates that addresses AI as well as other technology. 253 00:12:15.080 --> 00:12:17.880 I used information from the Harvard executive course 254 00:12:17.880 --> 00:12:21.000 called AI in Healthcare from strategies to implementation. 255 00:12:21.299 --> 00:12:26.799 And these two research articles were reviews of the nutritional management and 256 00:12:27.359 --> 00:12:29.880 effectiveness of chatbots and lifestyle behaviors 257 00:12:29.880 --> 00:12:31.799 as resources for this presentation. 258 00:12:33.710 --> 00:12:36.809 If you have questions about the content of this presentation, 259 00:12:37.169 --> 00:12:41.030 you can reach me at the email address on the slide, 260 00:12:41.250 --> 00:12:45.890 or you can use the contact us button on the NCpro.com website. 261 00:12:46.090 --> 00:12:46.530 Thank you.