1
00:00:06,367 --> 00:00:12,809
Welcome to LawNext PR, the podcast where we put a spotlight on the latest news coming out
of the legal tech industry.

2
00:00:12,889 --> 00:00:22,232
This is Bob Ambrosie and in each sponsored episode of LawNext PR, I interview a legal tech
company about its latest news and developments.

3
00:00:22,492 --> 00:00:31,066
topic today is the launch of Lighthouse AI Search by the global e-discovery and
information governance company, Lighthouse.

4
00:00:31,066 --> 00:00:38,894
And joining me to discuss this news is Dan Brassil executive director, search and
information retrieval at Lighthouse.

5
00:00:38,894 --> 00:00:40,531
Dan, Welcome to the show.

6
00:00:41,344 --> 00:00:42,576
Thank you for having me.

7
00:00:43,209 --> 00:00:53,337
So before we get into talking about what you've announced here, can I ask you to kind give
me in a nutshell what Lighthouse is for anybody who might not be familiar with the

8
00:00:53,337 --> 00:00:54,187
company?

9
00:00:54,640 --> 00:00:57,140
Yeah, I think you did a great job at the very beginning, right?

10
00:00:57,140 --> 00:01:03,820
We're a global e-discovery and information governance company.

11
00:01:03,820 --> 00:01:07,760
We've been around for, I think, like since 1995.

12
00:01:07,920 --> 00:01:20,292
We've been in this space for a really long time and really evolving with the technology,
but also pushing forward, being on the vanguard of what's possible as we incorporate new

13
00:01:20,292 --> 00:01:22,301
and different types of technology like

14
00:01:22,301 --> 00:01:23,231
AI.

15
00:01:23,419 --> 00:01:24,240
Yeah.

16
00:01:25,145 --> 00:01:27,313
And what about your own background?

17
00:01:27,313 --> 00:01:29,659
What's, how did you get into all of this?

18
00:01:29,659 --> 00:01:37,670
Yeah, well, so actually, this month, it will be my 20th anniversary in the industry.

19
00:01:37,750 --> 00:01:39,030
Thank you.

20
00:01:39,030 --> 00:01:44,482
Like a lot of people, you you don't choose eDiscovery, eDiscovery chooses Right.

21
00:01:44,482 --> 00:01:47,162
I have a background in linguistics.

22
00:01:47,182 --> 00:01:52,260
And I found my way into at the time back in 2005.

23
00:01:52,260 --> 00:02:02,500
And they were doing really cutting edge things about modeling language from its
information retrieval space for responsiveness and hot docs.

24
00:02:02,560 --> 00:02:15,375
And it was really, really interesting that I was able to take the training I had about
looking at language sort of how variable can be to help lawyers do practical things with

25
00:02:15,375 --> 00:02:16,455
my training.

26
00:02:16,455 --> 00:02:19,815
And I think that's the cool thing about

27
00:02:19,885 --> 00:02:30,676
the AI, right, which is now we have large language models that can also really understand
the variability of human language and being able to melt.

28
00:02:30,676 --> 00:02:37,005
So my expertise with this really powerful analytic tool is really interesting.

29
00:02:37,005 --> 00:02:38,126
Yeah, that is really interesting.

30
00:02:38,126 --> 00:02:43,019
mean, legal work is all about words, ultimately, and it comes down to that.

31
00:02:43,700 --> 00:02:50,476
So Lighthouse has just announced the launch of Lighthouse AI Search, and we can get into
kind of more details about what it is.

32
00:02:50,476 --> 00:02:54,908
But at a high level, why don't you kind of describe what it is that you've just, what
you've launched here.

33
00:02:54,973 --> 00:03:10,640
Yeah, you know, at a high level, Lighthouse AI Search is an AI driven or AI first, as we
like to say, search platform that allows someone to just ask a normal everyday question of

34
00:03:10,640 --> 00:03:11,680
the tool.

35
00:03:11,680 --> 00:03:21,719
And what you get back are answers to that question as opposed to getting back documents
that then you have to find the answers in, right?

36
00:03:21,719 --> 00:03:32,208
It does give you the documents, right, that it generates the answers from so that you can
always, you know, do that sort of level of analysis, but it gives you back an answer to

37
00:03:32,208 --> 00:03:34,599
your question, which is incredibly powerful.

38
00:03:34,599 --> 00:03:35,060
Yeah.

39
00:03:35,060 --> 00:03:39,625
So you're able to ask questions against the corpus of documents that you have.

40
00:03:39,625 --> 00:03:43,748
Can you kind of give me an example, perhaps, of a question you might ask?

41
00:03:44,461 --> 00:03:48,192
Yeah, so again, like I said, you can ask a simple question.

42
00:03:48,192 --> 00:03:59,359
It can be as simple as what are Juul's products, again, I'm talking about what is the
product we're talking about, the Juul data that we were using to show a client as publicly

43
00:03:59,359 --> 00:03:59,951
available.

44
00:03:59,951 --> 00:04:06,476
And then we asked other questions like what were the first flavors that Juul had, right?

45
00:04:06,497 --> 00:04:17,754
Again, these are things that if you didn't know the answer to them, you would not, it
would be a hard time figuring out how to craft a search to find all that information,

46
00:04:17,754 --> 00:04:17,994
right?

47
00:04:17,994 --> 00:04:26,279
You'd have to sort of dig through the documents that a search like flavor within five of
Juul would give you, right?

48
00:04:26,279 --> 00:04:30,011
So this just gives you a list and says, these were the flavors at launch.

49
00:04:30,011 --> 00:04:32,813
If that's your question, what were the flavors at launch?

50
00:04:32,813 --> 00:04:34,029
So again, it's

51
00:04:34,029 --> 00:04:45,208
It's sort of what it allows you to do is just ask a question as if you're asking a
colleague or if you are a senior associate or a partner at a law firm, what you would ask

52
00:04:45,208 --> 00:04:46,923
a junior associate to do.

53
00:04:46,923 --> 00:04:47,655
Right.

54
00:04:47,655 --> 00:04:49,127
I need to find out this.

55
00:04:49,127 --> 00:04:50,837
Can you go find it for me?

56
00:04:50,837 --> 00:04:53,923
it sounds like it may be a situate like the example you just gave.

57
00:04:53,923 --> 00:04:59,457
You probably could have found that answer eventually without AI search, but it would have
taken you a long time.

58
00:04:59,457 --> 00:05:04,511
And maybe you wouldn't have been absolutely sure that you were covering all the bases in
terms of what you were looking for.

59
00:05:04,902 --> 00:05:05,803
Yeah, most definitely.

60
00:05:05,803 --> 00:05:12,620
You know, as someone who built his career on helping people use linguistics to find
things, right?

61
00:05:12,620 --> 00:05:18,385
I, one of my expertise in this field is the crafting of search terms, right?

62
00:05:18,385 --> 00:05:23,310
Using syntax and then you have to know the arcana of syntax, right?

63
00:05:23,310 --> 00:05:27,353
and that can be difficult because sometimes you're like, did I get all the right words,
right?

64
00:05:27,353 --> 00:05:30,367
Because that's how search engines work.

65
00:05:30,367 --> 00:05:40,773
and then what you get back when you write when you do a search like that you get that
document again you don't get back the answer to your question with Lighthouse AI search.

66
00:05:40,773 --> 00:05:42,101
You don't need to know.

67
00:05:42,101 --> 00:05:54,114
search syntax you just need to ask your question in a normal way then again what you get
back is the answer to your question backed up by the documents that.

68
00:05:54,114 --> 00:05:57,878
the AI search was able to use to generate your question.

69
00:05:57,878 --> 00:06:00,101
Again, I think it's really powerful,

70
00:06:00,204 --> 00:06:09,692
So, I mean, you know, traditionally in e-discovery, you would have keyword search, Boolean
search, as you say, there are also tools like technology assisted review or predictive

71
00:06:09,692 --> 00:06:11,514
coding or those kinds of things.

72
00:06:11,514 --> 00:06:15,268
So where does this fit in that universe?

73
00:06:15,268 --> 00:06:17,170
Does this replace those kinds of search?

74
00:06:17,170 --> 00:06:19,242
Is it a supplement to those kinds of search?

75
00:06:19,242 --> 00:06:22,662
I think it works in tandem with a lot of different types of things.

76
00:06:22,922 --> 00:06:41,582
So first and foremost, this is a tool that allows litigation teams to get early insight to
query their documents when they're doing sort of like investigations or litigation.

77
00:06:41,582 --> 00:06:48,586
It's not just an e-discovery tool, more of a, you know, again, I think investigation is a
great example.

78
00:06:48,586 --> 00:06:52,186
or early case assessment, is there a there there, right?

79
00:06:52,186 --> 00:06:59,261
Even prior to sort of litigation or anything like that, you can say, what is my exposure,
right?

80
00:06:59,261 --> 00:07:02,781
And you can just ask the tool this.

81
00:07:02,781 --> 00:07:12,721
Now, going back to your question, how does this sit along, sort of say predictive AI or
predictive coding, I think you said, and other types of tools.

82
00:07:12,721 --> 00:07:17,121
Again, it has a different use case.

83
00:07:17,549 --> 00:07:23,222
I would say this about what AI search or Lighthouse AI search does, right?

84
00:07:23,222 --> 00:07:37,600
It is a combination of predictive AI to find the information you're looking for in your
question and then generative AI to generate the answer to the question you ask based off

85
00:07:37,600 --> 00:07:41,742
of the information it found using the predictive AI.

86
00:07:41,742 --> 00:07:46,851
I know I got into the weeds there, but I think it's really important to understand that
we're using

87
00:07:46,851 --> 00:07:51,873
the best of both worlds when it comes to the different types of AIs that are out there.

88
00:07:52,044 --> 00:07:52,305
Right.

89
00:07:52,305 --> 00:07:53,448
That makes a lot of sense.

90
00:07:53,448 --> 00:07:59,735
You talked about the fact that this search not only gives you the answer, but gives you.

91
00:08:00,018 --> 00:08:05,453
the sources that it relied on in terms of creating or generating that answer.

92
00:08:05,453 --> 00:08:13,874
And, you know, I wonder if you could talk a little bit more about that because again, we
all know that lawyers are skittish about generative AI because of their fear that it's

93
00:08:13,874 --> 00:08:16,942
gonna make stuff up or hallucinate answers.

94
00:08:16,942 --> 00:08:23,628
So how confident can they be of the kinds of answers that they're getting using this AI
search?

95
00:08:24,107 --> 00:08:25,198
That's a great question.

96
00:08:25,198 --> 00:08:27,959
So yeah, we all know about hallucinations, right?

97
00:08:27,959 --> 00:08:30,881
Especially when it comes to generative AI.

98
00:08:30,881 --> 00:08:34,985
What I think is really special about Lighthouse AI Search is it's

99
00:08:34,985 --> 00:08:40,027
a RAG model or RAG stands for Retrieval Augmented Generation.

100
00:08:40,148 --> 00:08:47,093
So what that allows you to do is instead of looking for answers out in the internet,
right?

101
00:08:47,093 --> 00:08:52,835
It restricts its answers to the corpus of documents you point the system at.

102
00:08:53,036 --> 00:08:58,195
So it won't go outside if it doesn't find the answer within the document population.

103
00:08:58,195 --> 00:09:06,449
that you're querying, it will say there is no information that supports that answer.

104
00:09:06,449 --> 00:09:08,070
there is, I can't find an answer.

105
00:09:08,070 --> 00:09:10,616
It will say use better words than that, obviously.

106
00:09:10,616 --> 00:09:18,700
But yes, it really reduces the likelihood of a hallucination, like rape date.

107
00:09:18,700 --> 00:09:22,759
In fact, I haven't exactly, right.

108
00:09:22,759 --> 00:09:27,400
document that I used or here's the sources and you can go check them yourself basically,
right?

109
00:09:27,400 --> 00:09:27,775
Yeah.

110
00:09:27,775 --> 00:09:28,456
Most definitely.

111
00:09:28,456 --> 00:09:34,917
And then if it doesn't, if it can't find an answer, it will say, there is no answer and
there will be no documents.

112
00:09:34,917 --> 00:09:35,157
Right.

113
00:09:35,157 --> 00:09:37,994
So it won't, it won't make something up.

114
00:09:38,313 --> 00:09:39,173
Yeah.

115
00:09:39,335 --> 00:09:40,036
Yeah.

116
00:09:40,036 --> 00:09:40,841
Very good thing.

117
00:09:40,841 --> 00:09:41,661
Yeah.

118
00:09:42,121 --> 00:09:46,281
So you talked a little bit about use cases, but I wonder if you could expand on that a
little bit.

119
00:09:46,281 --> 00:09:52,981
mean, where are the appropriate use cases for AI search in the eDiscovery process?

120
00:09:53,490 --> 00:10:01,070
Right, yeah, so as I said, you know, the first one that comes to mind is any form of
internal investigation, right?

121
00:10:01,070 --> 00:10:03,770
Where you're like, is there a there there?

122
00:10:03,770 --> 00:10:07,990
Right, and you want to point it at a small subset of data to understand your exposure.

123
00:10:08,010 --> 00:10:10,570
The same thing with PCA, right?

124
00:10:10,570 --> 00:10:12,230
Early case assessment, right?

125
00:10:12,230 --> 00:10:22,738
Where you're, you don't need to find all of something, you need to find information that
lets you make a strategic decision about what to do next.

126
00:10:22,738 --> 00:10:23,078
Right.

127
00:10:23,078 --> 00:10:24,958
You need information.

128
00:10:24,958 --> 00:10:26,398
You don't need documents.

129
00:10:26,398 --> 00:10:26,538
Right.

130
00:10:26,538 --> 00:10:28,538
That's a quick way of putting it.

131
00:10:28,538 --> 00:10:40,698
Another case is when you're doing, you know, motion prep or you are looking to see what's
coming in from incoming production.

132
00:10:41,358 --> 00:10:42,058
Right.

133
00:10:42,058 --> 00:10:51,358
All these things were again where you don't need to find all of something, but rather you
need to you need to get grounded in what's in this data.

134
00:10:51,714 --> 00:10:52,976
What do I need to know?

135
00:10:52,976 --> 00:10:56,342
And then you can then make strategic decisions based off of what you learn.

136
00:10:56,342 --> 00:11:02,577
Traditionally, what people often do is do just a full man review of all the data, right?

137
00:11:02,577 --> 00:11:12,937
They might use keywords to help navigate what to look at first, but they do tend to do
sort of, if not exhaustive, quasi-exhaustive review of that data.

138
00:11:13,119 --> 00:11:21,415
This is a situation where you can just start interrogating the information right away or
the data right away to get at the information you need.

139
00:11:21,415 --> 00:11:29,087
And then if you decide based off of what you learn to do a more fulsome review, you can,
but you don't need to start there.

140
00:11:29,087 --> 00:11:32,327
Does it matter the type of data in a sense?

141
00:11:32,327 --> 00:11:43,127
If I've got a collection of documents and I've got say medical records in there or
deposition transcripts or that sort of thing, can it be used against those kinds of

142
00:11:43,127 --> 00:11:43,927
documents?

143
00:11:44,796 --> 00:11:45,497
It can, right?

144
00:11:45,497 --> 00:11:49,311
I mean, it's based off of, you know, it's like all AIs, right?

145
00:11:49,311 --> 00:11:51,804
It's based off of large language models.

146
00:11:51,804 --> 00:11:58,262
And so it does require text, but as long as you have text, you can use it.

147
00:11:58,837 --> 00:12:00,129
Yeah.

148
00:12:00,129 --> 00:12:04,032
And what about in the in the litigation context?

149
00:12:04,032 --> 00:12:05,954
mean, kind of post discovery a little bit.

150
00:12:05,954 --> 00:12:06,486
I don't know.

151
00:12:06,486 --> 00:12:11,238
I don't know whether you're you've talked to any of your customers who are kind of using
it in this way.

152
00:12:11,238 --> 00:12:21,590
But I'm curious whether you can do things like go in and find out whether I don't know
whether something a witness just said might have ever been contradicted in something they

153
00:12:21,590 --> 00:12:24,001
said before that sort of an application.

154
00:12:24,001 --> 00:12:25,421
mean, that's funny.

155
00:12:25,421 --> 00:12:31,601
You should mention that we were just talking about that yesterday amongst my internal team
about what are some other use cases.

156
00:12:31,601 --> 00:12:36,281
And that was one of them, which is someone said something.

157
00:12:36,861 --> 00:12:38,721
Can we find contradictory information?

158
00:12:38,721 --> 00:12:39,361
Right.

159
00:12:39,361 --> 00:12:50,201
And that's a great use case because here you have something very specific and then you can
say, is there any information in this data that contradicts?

160
00:12:50,201 --> 00:12:52,689
You can literally just say that that contradicts

161
00:12:52,689 --> 00:12:54,771
this and whatever this is, right?

162
00:12:54,771 --> 00:12:56,231
So yeah.

163
00:12:56,231 --> 00:13:07,558
And again, you can just ask it that type of question as opposed to thinking about how do I
graph Boolean searches or proximity searches to try to get at this sort of information.

164
00:13:07,558 --> 00:13:11,002
Dan I've been asking you lots of questions but I wanted to give you an opportunity.

165
00:13:11,002 --> 00:13:16,217
there other features you wanted to point out or anything else you wanted to say about this
that we haven't talked about?

166
00:13:16,217 --> 00:13:23,545
Yeah, I think buried in sort of what we've been talking about, right, is sort of the value
proposition, right?

167
00:13:23,545 --> 00:13:30,580
And I think the value proposition for Lighthouse AI Search is speed to knowledge, right?

168
00:13:30,580 --> 00:13:32,912
That is incredibly valuable.

169
00:13:32,912 --> 00:13:39,430
Again, as I said, I've been working in this space for 20 years, and that is the one thing
that case teams are often sort of,

170
00:13:39,430 --> 00:13:42,402
concerned with, which is like, need to know something.

171
00:13:42,402 --> 00:13:45,585
How long is it going to take me to find it out?

172
00:13:45,585 --> 00:13:48,513
This is a tool to really speed that up.

173
00:13:48,513 --> 00:13:50,450
mean, greatly so.

174
00:13:50,450 --> 00:14:02,469
We think about using it all the time to answer basic questions that in the past, when we
worked with clients, we would have crafted Boolean searches and then said, OK, within this

175
00:14:02,469 --> 00:14:04,207
set, here's the answer.

176
00:14:04,207 --> 00:14:07,416
we could just now say, here's the answer.

177
00:14:07,416 --> 00:14:15,416
Yeah, seems like, I mean, you mentioned early case assessment, seems like it would be
incredibly powerful at that stage of a case when you're still not quite sure what you have

178
00:14:15,416 --> 00:14:19,896
or what you're working with, but where you can dive in and find something really quickly.

179
00:14:20,249 --> 00:14:21,789
Yes, exactly.

180
00:14:21,789 --> 00:14:29,774
You know, we have we have one client who's thinking about using Lighthouse AI Search to do
QC over production.

181
00:14:29,774 --> 00:14:31,656
That's going to be an interesting use case.

182
00:14:31,656 --> 00:14:33,577
We'll be helping them with that as well.

183
00:14:33,577 --> 00:14:43,243
We have another client who, again, is thinking about they need to respond to
interrogatories again.

184
00:14:43,359 --> 00:14:44,710
great use case.

185
00:14:45,426 --> 00:14:46,546
Right.

186
00:14:46,826 --> 00:14:52,446
And I guess the other thing we didn't really mention, but it probably is obvious to people
who listening who work in work in the discovery.

187
00:14:52,646 --> 00:15:03,905
But I'm presuming this can be used as effectively against your own documents and against
incoming production from from another side, from their side.

188
00:15:03,905 --> 00:15:08,269
yeah, incoming production is just a primary use case, right?

189
00:15:08,269 --> 00:15:13,002
Because this is a situation where you don't need to review all of them, right?

190
00:15:13,002 --> 00:15:15,483
You have very specific questions.

191
00:15:15,483 --> 00:15:18,877
Or you might actually have a real sort of what's in here, right?

192
00:15:18,877 --> 00:15:21,629
You know what the contours of the matter are.

193
00:15:21,629 --> 00:15:25,112
You can just start interrogating what did they produce to me, right?

194
00:15:25,112 --> 00:15:27,653
Does it have the answers I expected to have?

195
00:15:27,813 --> 00:15:31,473
Are there documents that are expected to be produced?

196
00:15:31,753 --> 00:15:33,273
Are they in here?

197
00:15:33,273 --> 00:15:34,233
Right?

198
00:15:34,373 --> 00:15:38,880
And then you can, again, decide what to do based off of what you find out.

199
00:15:38,880 --> 00:15:42,529
And Lighthouse AI Search, this is now generally available.

200
00:15:42,529 --> 00:15:43,351
Is it out there?

201
00:15:43,351 --> 00:15:46,262
It's out there, but it's not in general availability just yet.

202
00:15:46,262 --> 00:15:49,894
are, yes, it's a great question, right?

203
00:15:49,894 --> 00:15:52,637
We are working with some early access clients.

204
00:15:52,637 --> 00:16:04,956
And then, but we fully expect to, as the summer and fall roll around, we'll really start
opening it up to general availability come the end of the year Q4.

205
00:16:04,956 --> 00:16:11,336
Okay, so by the end of the year, he'll be out there and he'll be kind of testing it with
customers until then.

206
00:16:11,376 --> 00:16:16,936
Well, sounds really cool and I really appreciate your taking the time to come on the show
and talk about it.

207
00:16:16,981 --> 00:16:18,681
Yeah, thank you.

208
00:16:18,681 --> 00:16:19,681
It's been a pleasure.

209
00:16:20,029 --> 00:16:22,130
That's it for today's episode.

210
00:16:22,130 --> 00:16:28,842
If you enjoyed it, please subscribe to this podcast wherever you get your podcasts, or you
can also find us on YouTube.

211
00:16:28,842 --> 00:16:31,232
Just search for LawnexPR there.

212
00:16:31,329 --> 00:16:36,234
You can also find it all in the Lawnex Legal Tech directory under the Resources tab.

213
00:16:36,574 --> 00:16:37,705
This is Bob Ambrogi.

214
00:16:37,705 --> 00:16:38,895
Thanks for joining us today.