WEBVTT

1
00:00:04.080 --> 00:00:06.839
<v Speaker 1>Helping leaders motivate their people to a higher level of

2
00:00:06.879 --> 00:00:11.599
<v Speaker 1>performance through strong human relations, team building, and goalagiving. This

3
00:00:11.800 --> 00:00:16.280
<v Speaker 1>is the Seven Minute Leadership Podcast with your host Paul Fellovaliedo.

4
00:00:21.039 --> 00:00:24.760
<v Speaker 2>Hello everyone, and welcome to the Seven Minute Leadership Podcast.

5
00:00:24.839 --> 00:00:29.199
<v Speaker 2>It's episode two fifty four. Today we're diving into a

6
00:00:29.280 --> 00:00:32.439
<v Speaker 2>topic that every C suite leader needs to have on

7
00:00:32.479 --> 00:00:38.719
<v Speaker 2>their radar, and that's AI data driven leadership in cybersecurity.

8
00:00:39.000 --> 00:00:43.439
<v Speaker 2>In the digital age, the rapid advancements in artificial intelligence

9
00:00:43.439 --> 00:00:47.719
<v Speaker 2>and machine learning are no longer optional tools for organizations.

10
00:00:48.399 --> 00:00:53.200
<v Speaker 2>They are essential for optimizing operations, predicting market trends, and

11
00:00:53.280 --> 00:00:58.439
<v Speaker 2>delivering personalized customer experiences. But with this digital transformation comes

12
00:00:58.479 --> 00:01:04.599
<v Speaker 2>a major responsibility safeguarding data and ensuring ethical AI use.

13
00:01:05.239 --> 00:01:08.719
<v Speaker 2>To help us unpack this, I'm joined today by Josh Gelman,

14
00:01:08.920 --> 00:01:13.760
<v Speaker 2>a technology and cyber security expert from Gelman Integrative Consulting

15
00:01:13.879 --> 00:01:18.159
<v Speaker 2>in Pittsburgh. Josh has spent years helping businesses navigate the

16
00:01:18.159 --> 00:01:24.920
<v Speaker 2>complex world of cybersecurity, digital ethics, and AI driven decision making. Josh,

17
00:01:25.000 --> 00:01:26.480
<v Speaker 2>Welcome back to the show.

18
00:01:27.319 --> 00:01:28.159
<v Speaker 3>Thanks for having me.

19
00:01:28.280 --> 00:01:30.799
<v Speaker 2>Yeah, I think it's been over one hundred episodes. Ago

20
00:01:31.000 --> 00:01:34.760
<v Speaker 2>was the last time you were here. Yeah, so thanks

21
00:01:34.760 --> 00:01:35.519
<v Speaker 2>for making the time.

22
00:01:35.719 --> 00:01:37.799
<v Speaker 3>Yeah, it's been a long time yep.

23
00:01:38.000 --> 00:01:42.239
<v Speaker 2>So let's start with AI and data driven leadership. We

24
00:01:42.359 --> 00:01:46.239
<v Speaker 2>hear a lot about AI revolutionizing industries, but from a

25
00:01:46.319 --> 00:01:51.079
<v Speaker 2>leadership perspective, how should CEOs and executives be thinking about

26
00:01:51.200 --> 00:01:52.359
<v Speaker 2>AI implementation?

27
00:01:53.599 --> 00:01:56.799
<v Speaker 3>Well, I think with AI, you need to approach it

28
00:01:56.879 --> 00:02:01.159
<v Speaker 3>like any other new technology. So a lot of people

29
00:02:01.239 --> 00:02:04.640
<v Speaker 3>see something that's AI, this new shiny new toy that's AI,

30
00:02:05.000 --> 00:02:07.519
<v Speaker 3>and they download the app or they want to sign

31
00:02:07.640 --> 00:02:11.319
<v Speaker 3>up and immediately start using it without thinking about what

32
00:02:11.360 --> 00:02:15.199
<v Speaker 3>the implications of that might be, not necessarily just from

33
00:02:15.199 --> 00:02:20.360
<v Speaker 3>a security perspective, but from a data autonomy perspective, privacy perspective,

34
00:02:21.159 --> 00:02:24.960
<v Speaker 3>all kinds of risk that might be introduced by just

35
00:02:25.199 --> 00:02:27.680
<v Speaker 3>starting to use a new technology without really giving it

36
00:02:27.680 --> 00:02:31.319
<v Speaker 3>a lot of forethoughts. So I think having a strong

37
00:02:31.759 --> 00:02:35.479
<v Speaker 3>strategic plan for how you're going to use AI before

38
00:02:35.520 --> 00:02:38.479
<v Speaker 3>you just dive in head first and start using it

39
00:02:38.240 --> 00:02:43.639
<v Speaker 3>is really important and really giving it some thought instead

40
00:02:43.639 --> 00:02:46.439
<v Speaker 3>of just diving in headfirst and starting to use the

41
00:02:46.479 --> 00:02:48.560
<v Speaker 3>new and shiny, greatest new toy.

42
00:02:49.960 --> 00:02:52.719
<v Speaker 2>Yeah that's a good point. So what do you think

43
00:02:52.759 --> 00:02:55.719
<v Speaker 2>are some of the common mistakes that you see leaders

44
00:02:55.800 --> 00:02:58.439
<v Speaker 2>make when integrating AI.

45
00:02:59.759 --> 00:03:05.759
<v Speaker 3>Well, I think again, just going directly into starting to

46
00:03:05.840 --> 00:03:09.400
<v Speaker 3>use a technology without thinking about how it's going to

47
00:03:09.599 --> 00:03:13.879
<v Speaker 3>help them meet business objectives. They just see that it's

48
00:03:13.879 --> 00:03:16.360
<v Speaker 3>a shiny new toy, immediately want to start playing with it,

49
00:03:16.400 --> 00:03:19.560
<v Speaker 3>and they're not really considering how this can be used

50
00:03:19.599 --> 00:03:23.080
<v Speaker 3>to increase revenue, how it can be used to drive

51
00:03:23.199 --> 00:03:27.080
<v Speaker 3>change within the organization. They just start using it without

52
00:03:27.120 --> 00:03:29.000
<v Speaker 3>giving a whole lot of thought. So I think having

53
00:03:29.039 --> 00:03:31.520
<v Speaker 3>a better strategic plan on how you're you going to

54
00:03:31.680 --> 00:03:35.759
<v Speaker 3>use AI is important. I think another thing that leaders

55
00:03:36.319 --> 00:03:39.840
<v Speaker 3>where people in leadership positions don't really consider much with

56
00:03:39.960 --> 00:03:45.919
<v Speaker 3>AI is the information they're feeding it. So, for example,

57
00:03:46.479 --> 00:03:50.240
<v Speaker 3>if you're sending it or if you're prompting it with

58
00:03:50.800 --> 00:03:55.800
<v Speaker 3>your private corporate information like, for example, things that contain

59
00:03:55.879 --> 00:03:58.800
<v Speaker 3>personally identifiable information, or if you work in healthcare, like

60
00:03:58.800 --> 00:04:03.159
<v Speaker 3>protected health information or trade secrets or financial data, all

61
00:04:03.199 --> 00:04:06.759
<v Speaker 3>of that data that you're submitting to the AI model

62
00:04:07.039 --> 00:04:10.599
<v Speaker 3>could potentially be used to train that model and could

63
00:04:10.680 --> 00:04:13.960
<v Speaker 3>end up in the wrong hands. And that's especially true

64
00:04:14.879 --> 00:04:19.600
<v Speaker 3>if you're using something that maybe isn't as reputable as

65
00:04:19.639 --> 00:04:22.800
<v Speaker 3>some of the more prominent AI tools that are out there.

66
00:04:24.360 --> 00:04:27.959
<v Speaker 3>People get concerned about AI tools like Deepseek, which is

67
00:04:28.040 --> 00:04:31.759
<v Speaker 3>based out of China, and people just feeded all kinds

68
00:04:31.759 --> 00:04:34.959
<v Speaker 3>of information and you're not really sure where that information's going,

69
00:04:35.240 --> 00:04:37.040
<v Speaker 3>who's getting their hands on it, and what they can

70
00:04:37.079 --> 00:04:40.240
<v Speaker 3>do with it. So I think it all goes back

71
00:04:40.279 --> 00:04:43.519
<v Speaker 3>to having that strategic plan about how you're going to

72
00:04:43.639 --> 00:04:46.680
<v Speaker 3>use AI. And I think a lot of organizations also

73
00:04:46.720 --> 00:04:50.279
<v Speaker 3>would benefit with by developing their own internal AI tools

74
00:04:50.319 --> 00:04:53.680
<v Speaker 3>as opposed to using publicly available AI. That way, they

75
00:04:53.680 --> 00:04:55.720
<v Speaker 3>have more control over where the data is going and

76
00:04:55.720 --> 00:04:56.560
<v Speaker 3>how it's being used.

77
00:04:56.800 --> 00:05:01.199
<v Speaker 2>Yeah, and you touched on a couple important points. Then

78
00:05:01.199 --> 00:05:03.920
<v Speaker 2>I'm going to circle back on here in a few minutes.

79
00:05:03.920 --> 00:05:09.399
<v Speaker 2>But for leaders looking to maximize AI's potential, what are

80
00:05:09.439 --> 00:05:14.079
<v Speaker 2>the key steps in building a true data driven organization

81
00:05:14.800 --> 00:05:17.160
<v Speaker 2>that actually makes smarter decisions.

82
00:05:18.199 --> 00:05:20.800
<v Speaker 3>I think the key is the people that you have

83
00:05:20.959 --> 00:05:23.439
<v Speaker 3>involved in the organization. You need to have buy in

84
00:05:23.519 --> 00:05:27.519
<v Speaker 3>from all levels of management, from your senior leaders to

85
00:05:27.920 --> 00:05:31.240
<v Speaker 3>the people that are on the you know, on the

86
00:05:31.279 --> 00:05:33.480
<v Speaker 3>factory floor, or the people that are just involved in

87
00:05:33.519 --> 00:05:36.120
<v Speaker 3>the day to day operations of the organization. And then

88
00:05:36.120 --> 00:05:38.959
<v Speaker 3>I think it helps to form It doesn't have to

89
00:05:38.959 --> 00:05:41.399
<v Speaker 3>be a formal committee, but some sort of committee or

90
00:05:41.480 --> 00:05:44.680
<v Speaker 3>working group or cask force where everybody can kind of

91
00:05:44.680 --> 00:05:47.959
<v Speaker 3>get around and discuss the latest and greatest in AI

92
00:05:48.079 --> 00:05:51.079
<v Speaker 3>trends and how those AI tools and trends can be

93
00:05:51.199 --> 00:05:55.639
<v Speaker 3>used to benefit the organization in the best way possible

94
00:05:55.680 --> 00:05:57.879
<v Speaker 3>as opposed to just having you know, we talk about

95
00:05:57.920 --> 00:06:01.160
<v Speaker 3>like shadow technology, where people within the organizations start using

96
00:06:01.199 --> 00:06:05.040
<v Speaker 3>technology that isn't sanctioned by their leadership. Instead of having

97
00:06:05.120 --> 00:06:08.519
<v Speaker 3>AI become a form of shadow technology, get a step

98
00:06:08.560 --> 00:06:12.000
<v Speaker 3>ahead of it by involving people that you work with,

99
00:06:12.079 --> 00:06:14.920
<v Speaker 3>developing these committees, these working groups and talking about the

100
00:06:15.000 --> 00:06:17.120
<v Speaker 3>latest tools that are out there, how they might benefit

101
00:06:17.160 --> 00:06:20.199
<v Speaker 3>you in your organization, and what the risks are. And

102
00:06:20.240 --> 00:06:22.839
<v Speaker 3>it really needs to be part of your overall strategic

103
00:06:22.839 --> 00:06:25.680
<v Speaker 3>plan but also your risk management plan and your organization.

104
00:06:26.560 --> 00:06:30.879
<v Speaker 2>Do we need to create the position shadow technologies are?

105
00:06:32.360 --> 00:06:34.439
<v Speaker 3>I don't think you need to create the position, but

106
00:06:34.519 --> 00:06:36.480
<v Speaker 3>you certainly need to be cognizant of it. I mean,

107
00:06:36.519 --> 00:06:39.920
<v Speaker 3>you need to be aware that people are probably using

108
00:06:39.959 --> 00:06:43.480
<v Speaker 3>technologies that you haven't sanctioned yet. And we saw that

109
00:06:43.519 --> 00:06:45.959
<v Speaker 3>a lot during COVID you know, we saw, for example,

110
00:06:46.040 --> 00:06:48.800
<v Speaker 3>I always give the example with contact tracers and a

111
00:06:48.839 --> 00:06:53.160
<v Speaker 3>lot of municipalities and states contracted with contact tracers and

112
00:06:53.240 --> 00:06:56.240
<v Speaker 3>they didn't really have the systems in place to store

113
00:06:56.360 --> 00:07:00.560
<v Speaker 3>the data that they were collecting from patients. They started

114
00:07:00.560 --> 00:07:04.360
<v Speaker 3>to use public Google Drive and didn't have that secure,

115
00:07:04.439 --> 00:07:07.120
<v Speaker 3>so then anybody could access that information. And I think

116
00:07:07.199 --> 00:07:10.000
<v Speaker 3>you run into the same risk with AI if you're

117
00:07:10.040 --> 00:07:13.000
<v Speaker 3>not careful with how you're using it. You could have

118
00:07:13.040 --> 00:07:16.800
<v Speaker 3>employees that are essentially turning into insider threats. Right, They're

119
00:07:16.920 --> 00:07:21.800
<v Speaker 3>ex filtrating data from your organization into an AI model

120
00:07:22.240 --> 00:07:28.600
<v Speaker 3>that is potentially used by competing companies or maybe even

121
00:07:28.639 --> 00:07:33.279
<v Speaker 3>foreign governments and could be used in nefarious ways. So

122
00:07:33.319 --> 00:07:34.920
<v Speaker 3>you just really have to give it a lot of thought,

123
00:07:35.000 --> 00:07:37.879
<v Speaker 3>and it has to have to involve a lot of strategy,

124
00:07:37.920 --> 00:07:40.800
<v Speaker 3>a lot of forethought into actually using AI.

125
00:07:41.120 --> 00:07:47.600
<v Speaker 2>Yeah, so explain the importance of clean data, predictive analytics

126
00:07:47.639 --> 00:07:50.759
<v Speaker 2>in AI, transparency in decision making.

127
00:07:51.519 --> 00:07:54.079
<v Speaker 3>So with clean data, I mean garbage in, garbage out

128
00:07:54.120 --> 00:07:56.920
<v Speaker 3>is the old saying, right, So if you're not providing

129
00:07:56.959 --> 00:08:00.199
<v Speaker 3>clean data to a model, then you can't expect the

130
00:08:00.240 --> 00:08:03.399
<v Speaker 3>results that you get to be very accurate or very precise.

131
00:08:04.079 --> 00:08:07.279
<v Speaker 3>So you need to make sure that whatever data you're

132
00:08:07.319 --> 00:08:12.040
<v Speaker 3>giving the AI model is accurate. It's it's not tampered with.

133
00:08:12.079 --> 00:08:15.560
<v Speaker 3>You know, its integrity is intact. Uh and it's uh

134
00:08:15.639 --> 00:08:17.560
<v Speaker 3>and and that goes with your prompts too. I mean,

135
00:08:17.600 --> 00:08:21.199
<v Speaker 3>if you anyone who's played with chech ept, if if

136
00:08:21.240 --> 00:08:23.639
<v Speaker 3>you learn about prompt engineering and how to prompt AI,

137
00:08:23.959 --> 00:08:25.680
<v Speaker 3>there's good ways to do it and bad ways to

138
00:08:25.720 --> 00:08:28.959
<v Speaker 3>do it. You can give an AI model, a large

139
00:08:29.040 --> 00:08:31.000
<v Speaker 3>language model, a bad prompt and it's going to give

140
00:08:31.000 --> 00:08:32.720
<v Speaker 3>you bad output. But you could give it a good

141
00:08:32.720 --> 00:08:34.440
<v Speaker 3>prompt and it's going to give you good output. So

142
00:08:34.840 --> 00:08:37.120
<v Speaker 3>garbage in, garbage out is what I really think. That

143
00:08:37.200 --> 00:08:38.039
<v Speaker 3>all goes back.

144
00:08:37.879 --> 00:08:43.600
<v Speaker 2>To, Yeah, good h good point and special thanks to

145
00:08:43.679 --> 00:08:46.399
<v Speaker 2>the guy in the room that for Christmas gave me

146
00:08:46.799 --> 00:08:51.519
<v Speaker 2>a masterclass in prompt engineering. It's it's helped me a lot.

147
00:08:51.960 --> 00:08:54.279
<v Speaker 3>Yeah, and prompt engineering is really one of those things

148
00:08:54.320 --> 00:08:58.159
<v Speaker 3>that you know, it's always evolving, especially with new models

149
00:08:58.200 --> 00:09:02.360
<v Speaker 3>that are out there, because you know, every model is

150
00:09:02.399 --> 00:09:05.080
<v Speaker 3>a little bit different. And you know, even when you

151
00:09:05.080 --> 00:09:09.600
<v Speaker 3>look at like consider a copilot versus open AI. Co

152
00:09:09.840 --> 00:09:12.639
<v Speaker 3>pilot uses open AI models on the back end, but

153
00:09:12.720 --> 00:09:15.559
<v Speaker 3>they have essentially an interface or a wrapper in front

154
00:09:15.600 --> 00:09:18.159
<v Speaker 3>of the model that intercepts the prompt you give it

155
00:09:18.320 --> 00:09:20.720
<v Speaker 3>and inserts some of it their own logic before it

156
00:09:20.720 --> 00:09:23.759
<v Speaker 3>actually gets to the model, whereas with open AI it

157
00:09:23.799 --> 00:09:28.080
<v Speaker 3>doesn't add that additional seasoning to it. So the way

158
00:09:28.080 --> 00:09:31.600
<v Speaker 3>that you prompt the open AAI model might be completely

159
00:09:31.639 --> 00:09:34.879
<v Speaker 3>different than the way you prompt the copilot model. They're

160
00:09:34.879 --> 00:09:37.159
<v Speaker 3>the same models on the back end, but the results

161
00:09:37.159 --> 00:09:39.639
<v Speaker 3>you're getting are different. So prompt engineering is really important.

162
00:09:39.919 --> 00:09:43.480
<v Speaker 2>Yeah, it was really neat to go through that class,

163
00:09:43.519 --> 00:09:47.399
<v Speaker 2>and I'd highly encourage everybody that's serious about wanting to

164
00:09:47.480 --> 00:09:51.080
<v Speaker 2>learn AI to take some form of a prompt engineering

165
00:09:51.120 --> 00:09:54.559
<v Speaker 2>course online. But now we've got to talk about the

166
00:09:54.679 --> 00:09:58.360
<v Speaker 2>elephant in the room, Gelman in that cyber security in

167
00:09:58.480 --> 00:10:03.399
<v Speaker 2>digital ethics. So I guess with every advancement in AI

168
00:10:04.000 --> 00:10:08.720
<v Speaker 2>that we're seeing, all these new vulnerabilities are starting to emerge.

169
00:10:08.759 --> 00:10:12.480
<v Speaker 2>So what are some of the biggest cybersecurity threats facing

170
00:10:12.679 --> 00:10:13.720
<v Speaker 2>businesses today?

171
00:10:14.600 --> 00:10:18.360
<v Speaker 3>So when you think of emerging threats and the current

172
00:10:18.399 --> 00:10:26.440
<v Speaker 3>threat landscape in let's just say the United States, because

173
00:10:26.799 --> 00:10:34.000
<v Speaker 3>different countries certainly have different vulnerabilities and different threats that

174
00:10:34.080 --> 00:10:38.080
<v Speaker 3>might impact them differently. But when you're thinking about the

175
00:10:38.399 --> 00:10:40.360
<v Speaker 3>emerging threats in the United States, you have to think

176
00:10:40.360 --> 00:10:43.279
<v Speaker 3>of not just how AI can be used for good,

177
00:10:43.320 --> 00:10:46.320
<v Speaker 3>but how AI can be used for bad. So there

178
00:10:46.360 --> 00:10:51.320
<v Speaker 3>are AI models out there that are specifically designed to

179
00:10:51.519 --> 00:10:57.159
<v Speaker 3>generate malware, and attackers can use those models to manipulate

180
00:10:57.159 --> 00:11:00.720
<v Speaker 3>people through social engineering, phishing campaigns, things like that in

181
00:11:00.799 --> 00:11:03.960
<v Speaker 3>ways that you would never think are possible, or ways

182
00:11:03.960 --> 00:11:07.279
<v Speaker 3>that never were possible before. Like, for example, you can

183
00:11:07.759 --> 00:11:14.600
<v Speaker 3>essentially create an AI avatar of your CEO and have

184
00:11:14.720 --> 00:11:19.120
<v Speaker 3>that CEO avatar produce a message, a video message that

185
00:11:19.200 --> 00:11:23.080
<v Speaker 3>says they want you to send a payment in a

186
00:11:23.120 --> 00:11:26.279
<v Speaker 3>certain amount to this account that they give you, and

187
00:11:26.360 --> 00:11:29.600
<v Speaker 3>it looks like it's coming from that particular CEO. To

188
00:11:29.679 --> 00:11:33.279
<v Speaker 3>the untrained eye, it looks real, has human like features.

189
00:11:34.679 --> 00:11:36.879
<v Speaker 3>And even if you're not using video audio, I mean,

190
00:11:36.879 --> 00:11:39.360
<v Speaker 3>that's been a big one too, where you will clone

191
00:11:39.360 --> 00:11:42.279
<v Speaker 3>of somebody's voice and use AI to make it sound

192
00:11:42.320 --> 00:11:44.480
<v Speaker 3>like a message came from them. Then you could start

193
00:11:44.519 --> 00:11:48.399
<v Speaker 3>a teams meeting, say with the AI clone voice model

194
00:11:48.440 --> 00:11:51.159
<v Speaker 3>that your camera doesn't work and it sounds just like

195
00:11:51.200 --> 00:11:54.600
<v Speaker 3>the message is coming from that person. There are specific

196
00:11:54.759 --> 00:11:59.720
<v Speaker 3>AI models even that are just for malware worm GPT

197
00:11:59.879 --> 00:12:03.200
<v Speaker 3>was one of the original ones. There's AI models that

198
00:12:03.200 --> 00:12:07.879
<v Speaker 3>are trained on the dark Web. So using AI for evil,

199
00:12:07.919 --> 00:12:09.799
<v Speaker 3>I think is something that has to be considered as

200
00:12:09.840 --> 00:12:11.960
<v Speaker 3>one of the top emerging threats right now.

201
00:12:12.039 --> 00:12:14.559
<v Speaker 2>Yeah, no, because nobody in the world would ever use

202
00:12:14.600 --> 00:12:16.320
<v Speaker 2>AI for evil, right yeah.

203
00:12:16.399 --> 00:12:19.639
<v Speaker 3>I mean there's a model called dark Bird and it's

204
00:12:19.840 --> 00:12:23.320
<v Speaker 3>just trained completely on beta from the dark Web, and

205
00:12:23.960 --> 00:12:27.159
<v Speaker 3>you can use that model to do a lot of

206
00:12:27.320 --> 00:12:29.320
<v Speaker 3>a lot of bad things. Of course, we don't really

207
00:12:29.320 --> 00:12:30.600
<v Speaker 3>I guess want to go down that road because we

208
00:12:30.600 --> 00:12:32.240
<v Speaker 3>don't want to turn this into a training course for

209
00:12:32.360 --> 00:12:37.039
<v Speaker 3>for attackers. But you know, AI, you could do some

210
00:12:37.039 --> 00:12:40.159
<v Speaker 3>scary things with it, and I even think of like

211
00:12:40.240 --> 00:12:43.440
<v Speaker 3>if you have ever heard of the grandparent scam. No,

212
00:12:43.799 --> 00:12:47.799
<v Speaker 3>so that's like when somebody gets a text message from

213
00:12:48.240 --> 00:12:52.480
<v Speaker 3>they claim it's the attacker, will claim it's somebody's grandchild

214
00:12:52.919 --> 00:12:55.559
<v Speaker 3>or maybe it's a loved one, and they say that

215
00:12:55.639 --> 00:12:58.320
<v Speaker 3>you know, they got in trouble and they're gonna they're

216
00:12:58.320 --> 00:13:00.320
<v Speaker 3>gonna need money in order to get that trouble, and

217
00:13:00.320 --> 00:13:02.279
<v Speaker 3>you've got to send this money to this account or

218
00:13:02.279 --> 00:13:06.080
<v Speaker 3>you got to send bitcoin to this particular account, and

219
00:13:06.120 --> 00:13:08.360
<v Speaker 3>now you know, people kind of caught onto that. But

220
00:13:08.440 --> 00:13:12.519
<v Speaker 3>now with AI, you could go on to somebody's YouTube page,

221
00:13:12.759 --> 00:13:16.639
<v Speaker 3>a Facebook page where they have videos, clone their voice,

222
00:13:17.000 --> 00:13:18.879
<v Speaker 3>So clone the voice of the person that you want

223
00:13:18.919 --> 00:13:23.720
<v Speaker 3>to emulate, and actually send a recording or a voicemail

224
00:13:24.159 --> 00:13:26.879
<v Speaker 3>to somebody and make it sound like you know that

225
00:13:26.919 --> 00:13:29.960
<v Speaker 3>person is actually in trouble and need help, and it's

226
00:13:30.080 --> 00:13:32.360
<v Speaker 3>it's indistinguishable from the actual person.

227
00:13:32.639 --> 00:13:35.679
<v Speaker 2>Yeah. So so that brings me to AI ethics, right,

228
00:13:35.720 --> 00:13:40.360
<v Speaker 2>because we've seen major concerns around bias and AI, like

229
00:13:40.559 --> 00:13:45.320
<v Speaker 2>we've talked about with data privacy and even regulatory crackdown.

230
00:13:45.399 --> 00:13:50.519
<v Speaker 2>So how can CEOs navigate this landscape responsibly?

231
00:13:52.440 --> 00:13:56.639
<v Speaker 3>It's it's definitely a challenge. I think training is an

232
00:13:56.639 --> 00:13:59.679
<v Speaker 3>important element to it. I was just watching a LinkedIn

233
00:13:59.759 --> 00:14:03.799
<v Speaker 3>learn course actually the other day on ethics and AI,

234
00:14:04.399 --> 00:14:07.519
<v Speaker 3>and it was really providing a lot of useful insights

235
00:14:07.559 --> 00:14:09.720
<v Speaker 3>on how to use AI ethically and some of the

236
00:14:09.720 --> 00:14:12.240
<v Speaker 3>ethical challenges that you may not even consider with AI,

237
00:14:12.639 --> 00:14:15.440
<v Speaker 3>like how the model was trained and the environmental impacts

238
00:14:15.480 --> 00:14:18.159
<v Speaker 3>behind it, and the fact that you know you might

239
00:14:18.200 --> 00:14:21.320
<v Speaker 3>be getting or the AI model might have been using

240
00:14:21.320 --> 00:14:24.519
<v Speaker 3>copyrighted data, so you might be the output might be

241
00:14:24.559 --> 00:14:27.600
<v Speaker 3>copyrighted and you're not aware of it. So training and

242
00:14:27.639 --> 00:14:32.039
<v Speaker 3>awareness is certainly, I think the top thing that leaders

243
00:14:32.039 --> 00:14:38.679
<v Speaker 3>can do to kind of address that concern. But beyond that,

244
00:14:38.799 --> 00:14:40.919
<v Speaker 3>I mean, it's just really staying aware of all of

245
00:14:40.919 --> 00:14:42.399
<v Speaker 3>the emerging threats that are out there.

246
00:14:43.519 --> 00:14:48.840
<v Speaker 2>Here's what I think scares people or turns them off

247
00:14:48.879 --> 00:14:53.480
<v Speaker 2>to AI, especially in the business world, because I saw

248
00:14:53.519 --> 00:14:58.039
<v Speaker 2>this article just this morning that said this, and it

249
00:14:58.080 --> 00:15:00.360
<v Speaker 2>was Elon Musk, and it was in one of the

250
00:15:00.480 --> 00:15:04.679
<v Speaker 2>AI news articles that I subscribe to. Or they're talking

251
00:15:04.679 --> 00:15:08.440
<v Speaker 2>about his new what is it, GROC three? Yeah, okay,

252
00:15:09.320 --> 00:15:12.519
<v Speaker 2>and he was quoted saying GROCK three is more capable

253
00:15:12.960 --> 00:15:17.039
<v Speaker 2>than GROC two and is a maximally truth seeking AI

254
00:15:18.000 --> 00:15:21.159
<v Speaker 2>as it's been trained with ten times more computing power

255
00:15:21.919 --> 00:15:26.440
<v Speaker 2>using a data set that includes filings from court cases.

256
00:15:27.480 --> 00:15:30.399
<v Speaker 2>So people that aren't following the AI surge may not

257
00:15:30.519 --> 00:15:35.399
<v Speaker 2>be aware of what data is actually being put into

258
00:15:35.679 --> 00:15:38.000
<v Speaker 2>in the AI models are being trained with. And I

259
00:15:38.080 --> 00:15:41.639
<v Speaker 2>think that when regular people that aren't really plugged into

260
00:15:41.639 --> 00:15:45.399
<v Speaker 2>AI read an article that says, oh my god, this AI,

261
00:15:45.440 --> 00:15:49.519
<v Speaker 2>it's been trained with actual court cases, they may not

262
00:15:49.679 --> 00:15:54.240
<v Speaker 2>be aware of what information is being truly dumped into

263
00:15:54.279 --> 00:15:57.279
<v Speaker 2>these large language models. So can you talk a little

264
00:15:57.320 --> 00:15:59.600
<v Speaker 2>bit about that kind of stuff.

265
00:16:00.159 --> 00:16:03.039
<v Speaker 3>Yeah, So there's two things that really come to mind

266
00:16:03.039 --> 00:16:06.679
<v Speaker 3>with that. First is bias. AI models all have some

267
00:16:06.679 --> 00:16:09.120
<v Speaker 3>sort of algorithmic bias based on the data that they've

268
00:16:09.159 --> 00:16:13.039
<v Speaker 3>been trained with. So if you train a model on

269
00:16:13.399 --> 00:16:18.840
<v Speaker 3>specific court cases that are about a topic that is

270
00:16:19.120 --> 00:16:21.840
<v Speaker 3>something you're particularly interested in, something that's in your favor,

271
00:16:21.879 --> 00:16:25.559
<v Speaker 3>something that is aligned with your political beliefs, then the

272
00:16:25.600 --> 00:16:27.960
<v Speaker 3>results that you're going to get are also going to

273
00:16:27.960 --> 00:16:31.440
<v Speaker 3>be aligned with those political beliefs. So that goes back

274
00:16:31.440 --> 00:16:35.600
<v Speaker 3>to garbage in, garbage out. So biases are are a

275
00:16:35.639 --> 00:16:39.759
<v Speaker 3>big thing that can happen with with generative AI. The

276
00:16:39.799 --> 00:16:41.159
<v Speaker 3>other thing is hallucinations.

277
00:16:41.879 --> 00:16:42.240
<v Speaker 2>AI.

278
00:16:43.080 --> 00:16:45.200
<v Speaker 3>AI can make shut up, I can say shit right,

279
00:16:46.519 --> 00:16:50.519
<v Speaker 3>A AI can make shit up. That's that's hallucinations is

280
00:16:50.519 --> 00:16:52.440
<v Speaker 3>what they is. That what that's referred to in the

281
00:16:52.480 --> 00:16:56.279
<v Speaker 3>AI world. And it may it may sound correct when

282
00:16:56.360 --> 00:16:58.320
<v Speaker 3>whenever you get it out and get the output from

283
00:16:58.360 --> 00:17:00.679
<v Speaker 3>the model, it may sound like, oh, this is legitimate,

284
00:17:00.679 --> 00:17:04.839
<v Speaker 3>this is real, and it's not. And that's the other

285
00:17:04.880 --> 00:17:07.000
<v Speaker 3>thing you have to be careful for or careful and

286
00:17:07.160 --> 00:17:10.519
<v Speaker 3>whenever you're using especially some of these newer AI models,

287
00:17:10.720 --> 00:17:14.200
<v Speaker 3>when they don't have the training data to really answer

288
00:17:14.279 --> 00:17:17.279
<v Speaker 3>the prompt that you're providing it, they will make it

289
00:17:17.359 --> 00:17:19.920
<v Speaker 3>up and it will sound accurate. And we saw that

290
00:17:19.960 --> 00:17:22.240
<v Speaker 3>a lot with chat GPTU in its infancy.

291
00:17:22.720 --> 00:17:26.119
<v Speaker 2>Yeah, good points, Josh, Well, I got to tell you

292
00:17:26.160 --> 00:17:30.680
<v Speaker 2>this has been a great conversation before we wrap up.

293
00:17:30.720 --> 00:17:32.920
<v Speaker 2>Do you have one more little piece of advice you'd

294
00:17:32.920 --> 00:17:37.559
<v Speaker 2>give to any CEO or executive out there who's looking

295
00:17:37.599 --> 00:17:39.839
<v Speaker 2>to plug into AI for their business.

296
00:17:40.720 --> 00:17:46.240
<v Speaker 3>Yeah, I would say maybe number one, always keep training,

297
00:17:46.400 --> 00:17:50.200
<v Speaker 3>keep learning, stay on top of things, because the stuff

298
00:17:50.240 --> 00:17:52.039
<v Speaker 3>that's out there today is not the same stuff that's

299
00:17:52.039 --> 00:17:54.480
<v Speaker 3>going to be out there tomorrow. I would also say,

300
00:17:54.599 --> 00:17:57.839
<v Speaker 3>be transparent about your use of AI. If you're using

301
00:17:57.880 --> 00:18:01.559
<v Speaker 3>AI for something, you should make it clear to the

302
00:18:01.559 --> 00:18:04.359
<v Speaker 3>people that are the consumers of whatever you're generating that

303
00:18:04.400 --> 00:18:08.039
<v Speaker 3>you did use AI because there could be other implications

304
00:18:08.119 --> 00:18:10.799
<v Speaker 3>that come out down the road, especially with copyrights and

305
00:18:10.839 --> 00:18:14.039
<v Speaker 3>privacy and that type of thing. And then always make

306
00:18:14.079 --> 00:18:18.839
<v Speaker 3>sure that you're incorporating AI into your overall risk management strategy.

307
00:18:18.880 --> 00:18:21.920
<v Speaker 3>Of your organization. If you're not, you're leaving yourself open

308
00:18:21.920 --> 00:18:27.559
<v Speaker 3>and vulnerabilities to lawsuits to possible attacks. So it's really

309
00:18:27.559 --> 00:18:31.400
<v Speaker 3>important to consider AI in the context of risk management.

310
00:18:32.119 --> 00:18:36.039
<v Speaker 2>Well again, Josh, thanks for coming back. Do me a favor.

311
00:18:36.279 --> 00:18:38.640
<v Speaker 2>Let's not let it be another one hundred or so

312
00:18:38.759 --> 00:18:42.000
<v Speaker 2>episodes until you come back and do another one. Sure

313
00:18:42.160 --> 00:18:46.400
<v Speaker 2>fair deal. Yeah sounds good to me and for everyone listening.

314
00:18:46.599 --> 00:18:50.079
<v Speaker 2>This has been the seven Minute Leadership Podcast and we'll

315
00:18:50.119 --> 00:18:50.920
<v Speaker 2>see you next time.

316
00:18:51.319 --> 00:18:55.920
<v Speaker 1>For more Paul Fell of Alito Podcasts, visit paulfellowalito dot com.