1
00:00:00,080 --> 00:00:02,480
This is Carly Beam with Becker Spine and

2
00:00:02,480 --> 00:00:05,679
Orthopedics podcast. And joining us today is doctor

3
00:00:05,679 --> 00:00:09,059
Morgan L'Oreal, a spine surgeon and past president

4
00:00:09,279 --> 00:00:12,240
of the International Society for the Advancement of

5
00:00:12,240 --> 00:00:13,059
Spine Surgery.

6
00:00:13,439 --> 00:00:15,359
And today, we're gonna have a great chat

7
00:00:15,359 --> 00:00:15,859
about,

8
00:00:16,454 --> 00:00:19,574
AI and spine surgery and where AllSource kinda

9
00:00:19,574 --> 00:00:20,635
fits in the larger

10
00:00:21,015 --> 00:00:23,815
health care equation. So, doctor Lorio, thank you

11
00:00:23,815 --> 00:00:25,435
so much for joining us.

12
00:00:25,894 --> 00:00:27,414
Thank you, Carly. It's a pleasure to be

13
00:00:27,414 --> 00:00:29,255
here. Oh, it's great to have you back

14
00:00:29,255 --> 00:00:31,379
on the podcast. And before we dive into

15
00:00:31,379 --> 00:00:33,619
our questions, could you just give a a

16
00:00:33,619 --> 00:00:36,520
quick overview of your background and your work?

17
00:00:37,539 --> 00:00:40,340
Okay. Well, I've, been a spine surgeon and

18
00:00:40,340 --> 00:00:41,239
a hand surgeon

19
00:00:41,780 --> 00:00:42,179
and,

20
00:00:43,059 --> 00:00:45,460
past president of ISAS. And,

21
00:00:46,315 --> 00:00:47,435
I did a lot of work,

22
00:00:47,835 --> 00:00:48,655
for policy,

23
00:00:49,274 --> 00:00:51,755
and I chaired coding and reimbursement task force

24
00:00:51,755 --> 00:00:54,075
for ISAS as well. So that's pretty much

25
00:00:54,075 --> 00:00:55,454
been my interest, and

26
00:00:55,914 --> 00:00:56,975
AI kind of,

27
00:00:57,354 --> 00:00:59,759
fits in here because of its impact on

28
00:01:00,079 --> 00:01:01,539
policies is is it impacts,

29
00:01:02,239 --> 00:01:03,780
both health care delivery

30
00:01:04,079 --> 00:01:04,579
and,

31
00:01:05,439 --> 00:01:05,939
frankly,

32
00:01:07,599 --> 00:01:08,560
what we write,

33
00:01:08,959 --> 00:01:10,819
and what we do moving forward.

34
00:01:11,280 --> 00:01:12,899
So, hopefully, this is relevant.

35
00:01:13,614 --> 00:01:16,094
Absolutely. And, I mean, obviously, AI is is

36
00:01:16,094 --> 00:01:16,834
is everywhere

37
00:01:17,295 --> 00:01:19,715
in health care and health care conversations.

38
00:01:20,254 --> 00:01:22,734
I'd love to hear start off how you

39
00:01:22,734 --> 00:01:24,894
think about its role when it comes to

40
00:01:24,894 --> 00:01:25,715
spine surgery.

41
00:01:27,130 --> 00:01:30,090
Well, I I use AI, and I value

42
00:01:30,090 --> 00:01:31,390
it. It's an extraordinary

43
00:01:31,689 --> 00:01:32,750
analytical tool.

44
00:01:33,370 --> 00:01:35,290
If you step back for a moment, the

45
00:01:35,290 --> 00:01:36,430
scale of this technology,

46
00:01:37,530 --> 00:01:38,510
becomes clearer.

47
00:01:38,864 --> 00:01:41,344
We remember the Internet took decades to reshape

48
00:01:41,344 --> 00:01:41,844
society,

49
00:01:42,704 --> 00:01:43,364
and today,

50
00:01:43,745 --> 00:01:46,564
we recognize the lack of Internet access represents

51
00:01:46,625 --> 00:01:47,284
a form

52
00:01:47,665 --> 00:01:50,805
of inequality. And so artificial intelligence is arriving

53
00:01:51,379 --> 00:01:52,200
very differently.

54
00:01:52,659 --> 00:01:55,780
Its growth isn't gradual. It's almost vertical. And

55
00:01:55,780 --> 00:01:58,439
now that capability sits inside the devices

56
00:01:58,899 --> 00:02:00,200
we carry in our pockets.

57
00:02:00,500 --> 00:02:02,359
But I think the real risk,

58
00:02:03,140 --> 00:02:04,359
is often misunderstood.

59
00:02:05,060 --> 00:02:06,905
By that, I mean, the real

60
00:02:07,364 --> 00:02:09,944
risk isn't that AI will get things wrong,

61
00:02:10,405 --> 00:02:11,544
that it will succeed

62
00:02:11,925 --> 00:02:12,905
where we

63
00:02:13,284 --> 00:02:14,185
while we slowly

64
00:02:14,485 --> 00:02:15,465
forget how.

65
00:02:16,485 --> 00:02:19,705
In spine care, the the highest value work

66
00:02:19,925 --> 00:02:20,824
often happens

67
00:02:21,360 --> 00:02:23,539
before any incision is made.

68
00:02:24,080 --> 00:02:26,419
That's when a surgeon integrates imaging,

69
00:02:27,120 --> 00:02:27,620
biomechanics,

70
00:02:28,879 --> 00:02:29,780
patient history,

71
00:02:30,479 --> 00:02:30,979
comorbidities,

72
00:02:32,319 --> 00:02:32,819
expectations,

73
00:02:33,680 --> 00:02:35,539
and risk tolerance into a coherent

74
00:02:35,840 --> 00:02:36,900
clinical judgment.

75
00:02:37,495 --> 00:02:41,034
Mhmm. Those cognitive steps are difficult to measure

76
00:02:41,495 --> 00:02:43,354
and rare or rarely reimbursed

77
00:02:43,655 --> 00:02:45,435
in proportion to their importance,

78
00:02:46,455 --> 00:02:47,754
but they determine

79
00:02:48,534 --> 00:02:49,995
whether the right patient

80
00:02:50,375 --> 00:02:51,915
receives the right intervention

81
00:02:52,810 --> 00:02:53,870
at the right time.

82
00:02:54,489 --> 00:02:56,329
It's really fascinating way of looking at it

83
00:02:56,329 --> 00:02:58,269
too, just the idea that AI

84
00:02:58,569 --> 00:03:00,030
is coming in before,

85
00:03:00,969 --> 00:03:04,189
the actual surgery. But, obviously, some people

86
00:03:04,489 --> 00:03:07,049
are worried that AI might replace their jobs

87
00:03:07,049 --> 00:03:07,709
and including,

88
00:03:08,125 --> 00:03:10,925
you know, physicians. Do you see that being

89
00:03:10,925 --> 00:03:11,504
a concern?

90
00:03:12,284 --> 00:03:14,764
Well, I I don't think that replacement's the

91
00:03:14,764 --> 00:03:15,504
real issue,

92
00:03:16,125 --> 00:03:17,344
specifically for surgeons.

93
00:03:18,125 --> 00:03:18,944
Health care

94
00:03:19,245 --> 00:03:20,944
already operates through algorithms,

95
00:03:21,485 --> 00:03:21,985
reimbursement

96
00:03:22,364 --> 00:03:22,864
rules,

97
00:03:23,340 --> 00:03:24,080
prior authorization

98
00:03:24,379 --> 00:03:24,879
systems,

99
00:03:25,420 --> 00:03:26,480
coverage policies,

100
00:03:27,099 --> 00:03:28,960
and outcome reporting structures.

101
00:03:30,219 --> 00:03:32,780
But I can say this, AI will not

102
00:03:32,780 --> 00:03:34,560
enter medicine neutrally.

103
00:03:35,474 --> 00:03:38,914
Artificial intelligence will likely arrive through those same

104
00:03:38,914 --> 00:03:40,134
systems I mentioned.

105
00:03:40,914 --> 00:03:41,414
Unfortunately,

106
00:03:42,275 --> 00:03:43,014
skill erosion

107
00:03:43,314 --> 00:03:46,375
rarely shows up in the metrics we measure.

108
00:03:47,314 --> 00:03:48,775
Accuracy can improve

109
00:03:49,689 --> 00:03:51,550
while human judgment quietly

110
00:03:52,330 --> 00:03:52,830
declines.

111
00:03:53,209 --> 00:03:56,349
So I guess the real question becomes governance.

112
00:03:57,209 --> 00:03:59,469
Yeah. Can you dive into that more governance

113
00:03:59,689 --> 00:04:00,189
specifically?

114
00:04:01,370 --> 00:04:01,870
Yeah.

115
00:04:02,805 --> 00:04:05,465
Complex patients don't always fit neatly

116
00:04:06,004 --> 00:04:07,544
inside efficient systems.

117
00:04:08,645 --> 00:04:09,145
Historically,

118
00:04:09,925 --> 00:04:11,784
the surgeon in the loop judgment

119
00:04:12,405 --> 00:04:14,425
has allowed physicians to recognize

120
00:04:15,125 --> 00:04:17,944
when a patient doesn't fit the algorithm

121
00:04:18,649 --> 00:04:20,509
and to treat the patient

122
00:04:21,370 --> 00:04:22,589
rather than the pathway.

123
00:04:23,610 --> 00:04:25,949
Medicine requires flexibility because

124
00:04:26,250 --> 00:04:28,829
real patients are real complicated.

125
00:04:29,689 --> 00:04:30,669
Systems, however,

126
00:04:31,129 --> 00:04:32,110
often prefer

127
00:04:33,050 --> 00:04:33,550
predictability.

128
00:04:35,104 --> 00:04:37,204
Maintaining physician oversight, therefore,

129
00:04:37,504 --> 00:04:41,104
becomes critical as AI becomes embedded in those

130
00:04:41,104 --> 00:04:41,604
systems.

131
00:04:42,305 --> 00:04:45,024
Mhmm. It's an interesting balance, it sounds like,

132
00:04:45,024 --> 00:04:47,524
between the algorithms and then

133
00:04:48,000 --> 00:04:50,580
figuring out those one off cases.

134
00:04:51,040 --> 00:04:53,139
Can you give an example of where physician

135
00:04:53,199 --> 00:04:54,259
judgment still

136
00:04:54,800 --> 00:04:57,759
matters even when AI is helping out as

137
00:04:57,759 --> 00:04:58,339
a tool?

138
00:04:58,800 --> 00:04:59,300
Sure.

139
00:05:00,014 --> 00:05:03,375
Well, I I recently explored a cancer some

140
00:05:03,375 --> 00:05:04,514
cancer trend data,

141
00:05:05,055 --> 00:05:06,514
within an AI system.

142
00:05:07,134 --> 00:05:10,414
And when I asked whether cancer incidence had

143
00:05:10,414 --> 00:05:11,714
changed after COVID,

144
00:05:12,689 --> 00:05:13,910
the analysis suggested

145
00:05:14,449 --> 00:05:17,270
that incident rates were largely unchanged.

146
00:05:18,050 --> 00:05:20,470
Mhmm. But when I examined the time series

147
00:05:21,090 --> 00:05:21,910
more closely,

148
00:05:22,770 --> 00:05:24,150
something about the pattern,

149
00:05:25,024 --> 00:05:27,444
didn't quite fit in the data points.

150
00:05:27,985 --> 00:05:30,245
The COVID years were clearly disruptive,

151
00:05:30,865 --> 00:05:33,824
so we treated that period as difficult to

152
00:05:33,824 --> 00:05:34,324
interpret

153
00:05:34,704 --> 00:05:37,444
and step back to examine broader trends.

154
00:05:38,225 --> 00:05:39,604
When we then asked

155
00:05:40,300 --> 00:05:43,120
a different question focusing on cancer mortality

156
00:05:44,139 --> 00:05:45,199
instead of incidence,

157
00:05:45,740 --> 00:05:48,400
the pattern appeared different. Mhmm. Mortality

158
00:05:48,939 --> 00:05:52,639
seemed meaningfully higher than earlier baseline patterns

159
00:05:53,100 --> 00:05:53,600
roughly

160
00:05:54,004 --> 00:05:56,504
on the order of about sixteen percent.

161
00:05:57,125 --> 00:05:58,745
That is a substantial

162
00:05:59,125 --> 00:05:59,625
change.

163
00:06:00,324 --> 00:06:01,944
The the AI system

164
00:06:02,724 --> 00:06:05,365
hadn't really made a mistake per se. It

165
00:06:05,365 --> 00:06:07,865
had answered the question it was asked,

166
00:06:08,279 --> 00:06:10,699
but but it reminded me of something important.

167
00:06:12,520 --> 00:06:13,980
AI answers questions

168
00:06:14,600 --> 00:06:15,500
very well.

169
00:06:16,040 --> 00:06:17,500
Sometimes it narrows

170
00:06:17,800 --> 00:06:18,300
answers

171
00:06:18,759 --> 00:06:19,259
simultaneously.

172
00:06:20,384 --> 00:06:20,884
Physicians

173
00:06:21,425 --> 00:06:22,485
still have to decide

174
00:06:22,785 --> 00:06:24,164
which questions matter.

175
00:06:24,625 --> 00:06:27,664
Sometimes the insight comes from recognizing that the

176
00:06:27,664 --> 00:06:28,644
question itself

177
00:06:29,185 --> 00:06:30,164
needs to change.

178
00:06:31,024 --> 00:06:33,185
You're a fascinating idea there, and we'll circle

179
00:06:33,185 --> 00:06:34,889
back to that in a moment. I also

180
00:06:34,889 --> 00:06:37,769
wanna pick your brain on health systems and

181
00:06:37,769 --> 00:06:40,170
how many of them often they hope that

182
00:06:40,170 --> 00:06:42,509
AI will make care more efficient.

183
00:06:43,129 --> 00:06:44,430
What are your thoughts on that?

184
00:06:44,970 --> 00:06:48,564
Well, efficiency is important, but efficiency can also

185
00:06:48,564 --> 00:06:50,264
introduce blind spots.

186
00:06:50,805 --> 00:06:53,464
Sometimes I think about this using a GPS

187
00:06:53,524 --> 00:06:54,024
analogy.

188
00:06:54,564 --> 00:06:56,504
Mhmm. You know, navigation systems

189
00:06:57,044 --> 00:06:57,544
optimize

190
00:06:58,004 --> 00:06:59,464
routes based on filters,

191
00:07:00,180 --> 00:07:04,120
minimizing time, avoiding tolls, reducing distance.

192
00:07:04,500 --> 00:07:07,480
But, occasionally, those filters lead you into loops

193
00:07:07,779 --> 00:07:09,779
or take you off the main road only

194
00:07:09,779 --> 00:07:11,480
to bring you back again later.

195
00:07:12,185 --> 00:07:13,725
The system is optimized

196
00:07:14,105 --> 00:07:14,925
or optimizing

197
00:07:15,305 --> 00:07:16,444
based on the parameters

198
00:07:16,824 --> 00:07:17,725
it was given.

199
00:07:18,585 --> 00:07:20,525
AI functions in a similar way.

200
00:07:21,305 --> 00:07:23,485
AI can optimize the route,

201
00:07:24,025 --> 00:07:25,964
but someone still has to decide

202
00:07:26,910 --> 00:07:29,089
where we're going. And as we move forward,

203
00:07:29,789 --> 00:07:32,750
these filters, they're gonna are gonna become more

204
00:07:32,750 --> 00:07:34,050
and more important.

205
00:07:34,910 --> 00:07:36,990
Yeah. And, doctor L'Oreal, can you talk about

206
00:07:36,990 --> 00:07:39,169
the role you physicians should play

207
00:07:39,504 --> 00:07:41,824
as AI is becoming more integrated into the

208
00:07:41,824 --> 00:07:42,805
work they do?

209
00:07:43,345 --> 00:07:43,845
Yeah.

210
00:07:44,225 --> 00:07:47,185
So physicians need to remain actively involved in

211
00:07:47,185 --> 00:07:48,964
the governance of these systems,

212
00:07:49,504 --> 00:07:50,805
not just as users,

213
00:07:51,185 --> 00:07:52,165
but as stewards.

214
00:07:52,930 --> 00:07:55,350
Complex patients often generate more uncertainty

215
00:07:55,970 --> 00:07:57,509
and variability in outcomes,

216
00:07:58,129 --> 00:08:00,689
but those are frequently the patients who need

217
00:08:00,689 --> 00:08:02,310
the most thoughtful care.

218
00:08:02,930 --> 00:08:04,629
If systems begin to prioritize

219
00:08:05,009 --> 00:08:05,509
predictability

220
00:08:05,889 --> 00:08:06,949
above all else,

221
00:08:07,329 --> 00:08:08,230
there's a risk

222
00:08:08,665 --> 00:08:10,845
that the most complex patients

223
00:08:11,305 --> 00:08:13,004
may become harder to treat.

224
00:08:13,625 --> 00:08:14,125
Just

225
00:08:14,504 --> 00:08:17,165
over a year ago, I had the opportunity

226
00:08:17,225 --> 00:08:17,964
to participate

227
00:08:18,745 --> 00:08:19,645
in a global

228
00:08:20,264 --> 00:08:23,485
fifty roundtable organized by the Dubai Future Foundation,

229
00:08:24,100 --> 00:08:26,500
where experts from multiple sectors were asked to

230
00:08:26,500 --> 00:08:28,120
consider possible futures

231
00:08:28,580 --> 00:08:30,600
roughly five decades from now.

232
00:08:31,379 --> 00:08:33,779
My perspective in that discussion came from two

233
00:08:33,779 --> 00:08:34,279
places.

234
00:08:35,139 --> 00:08:38,125
My international policy work in spine surgery through

235
00:08:38,125 --> 00:08:38,625
ISAS,

236
00:08:39,485 --> 00:08:42,384
and my experience as a practicing surgeon,

237
00:08:44,044 --> 00:08:44,865
one idea

238
00:08:46,044 --> 00:08:49,105
emerging from that initiative is a classification framework

239
00:08:49,324 --> 00:08:50,544
describing how humans

240
00:08:50,909 --> 00:08:53,789
and artificial intelligence collaborate in the production of

241
00:08:53,789 --> 00:08:55,730
knowledge and decisions. Mhmm.

242
00:08:56,589 --> 00:08:59,389
The the framework is a simple visual markers

243
00:08:59,389 --> 00:09:01,409
to indicate the degree of human involvement

244
00:09:02,110 --> 00:09:04,829
in AI assisted work ranging from fully human

245
00:09:04,829 --> 00:09:07,214
activity to machine dominant processes

246
00:09:07,835 --> 00:09:11,434
with several intermediate stages representing different levels of

247
00:09:11,434 --> 00:09:12,815
collaboration and oversight.

248
00:09:13,274 --> 00:09:13,934
The purpose

249
00:09:14,394 --> 00:09:15,214
is transparency.

250
00:09:16,554 --> 00:09:19,595
As artificial intelligence becomes more deeply embedded in

251
00:09:19,595 --> 00:09:21,455
research, analysis, and communication,

252
00:09:22,149 --> 00:09:24,709
it will become increasingly difficult to determine how

253
00:09:24,709 --> 00:09:27,909
much of an output reflects human judgment versus

254
00:09:27,909 --> 00:09:28,809
machine computation.

255
00:09:29,750 --> 00:09:32,549
Mhmm. A clear labeling system helps make that

256
00:09:32,549 --> 00:09:33,049
relationship

257
00:09:33,429 --> 00:09:33,929
visible.

258
00:09:34,615 --> 00:09:36,075
From a clinical perspective,

259
00:09:36,695 --> 00:09:39,254
this aligns closely with something we have long

260
00:09:39,254 --> 00:09:40,634
emphasized in medicine,

261
00:09:41,335 --> 00:09:43,195
surgeon in the loop governance.

262
00:09:44,134 --> 00:09:46,634
Artificial intelligence may assist with analysis,

263
00:09:47,440 --> 00:09:48,259
pattern recognition,

264
00:09:48,879 --> 00:09:51,379
and data synthesis, but clinical responsibility

265
00:09:52,399 --> 00:09:53,700
and decision authority

266
00:09:54,240 --> 00:09:55,139
remain human.

267
00:09:56,080 --> 00:09:58,580
In the terminology of the Dubai framework,

268
00:09:59,200 --> 00:10:02,340
this approach would be classified as human led

269
00:10:02,399 --> 00:10:02,899
AI.

270
00:10:03,894 --> 00:10:06,315
A fascinating way to describe it, human led

271
00:10:06,375 --> 00:10:06,875
AI.

272
00:10:07,414 --> 00:10:08,554
And can you talk about

273
00:10:09,495 --> 00:10:10,475
how AI

274
00:10:10,774 --> 00:10:13,894
might influence the way medical knowledge itself is

275
00:10:13,894 --> 00:10:14,394
developing?

276
00:10:15,335 --> 00:10:18,669
Yeah. There's another dimension to this conversation that

277
00:10:18,669 --> 00:10:19,889
we don't always acknowledge.

278
00:10:20,589 --> 00:10:21,089
AI

279
00:10:21,389 --> 00:10:23,649
increasingly influences how we write,

280
00:10:24,110 --> 00:10:27,330
whether drafting clinical notes, summarizing literature,

281
00:10:27,789 --> 00:10:29,889
or producing scientific manuscripts.

282
00:10:30,764 --> 00:10:31,904
When technology

283
00:10:32,284 --> 00:10:34,065
influences how we write,

284
00:10:34,524 --> 00:10:35,424
it inevitably

285
00:10:36,044 --> 00:10:37,824
influences what we write.

286
00:10:38,524 --> 00:10:39,504
Knowing who

287
00:10:39,964 --> 00:10:40,704
or what

288
00:10:41,084 --> 00:10:42,784
produced it is therefore important.

289
00:10:43,639 --> 00:10:46,299
This can be categorized across six levels,

290
00:10:47,320 --> 00:10:48,220
human only,

291
00:10:49,080 --> 00:10:49,980
human directed,

292
00:10:50,840 --> 00:10:51,820
human led,

293
00:10:52,440 --> 00:10:53,340
machine assisted,

294
00:10:54,120 --> 00:10:55,179
machine dominant,

295
00:10:55,804 --> 00:10:56,865
and machine only.

296
00:10:57,324 --> 00:11:00,304
In fact, journals are already starting to formalize

297
00:11:00,524 --> 00:11:01,264
this principle.

298
00:11:01,804 --> 00:11:03,985
At the International Journal of Spine Surgery,

299
00:11:04,524 --> 00:11:07,084
new author guidelines make it clear that AI

300
00:11:07,084 --> 00:11:07,904
can assist

301
00:11:08,259 --> 00:11:09,879
with drafting or organization,

302
00:11:10,500 --> 00:11:11,319
but scientific

303
00:11:11,620 --> 00:11:12,120
judgment

304
00:11:12,579 --> 00:11:13,879
and clinical interpretation

305
00:11:14,659 --> 00:11:16,679
must remain human led.

306
00:11:17,459 --> 00:11:18,679
And over that time,

307
00:11:19,139 --> 00:11:21,319
that shapes will be collectively known,

308
00:11:21,975 --> 00:11:22,954
What we believe

309
00:11:23,654 --> 00:11:25,995
and ultimately how we practice medicine.

310
00:11:27,174 --> 00:11:29,195
AI systems apply filters,

311
00:11:30,454 --> 00:11:30,954
optimization

312
00:11:31,334 --> 00:11:31,834
rules,

313
00:11:32,695 --> 00:11:34,235
and administrative constraints.

314
00:11:35,570 --> 00:11:36,070
Capability

315
00:11:36,529 --> 00:11:37,269
will proceed

316
00:11:38,210 --> 00:11:38,710
consent,

317
00:11:39,090 --> 00:11:39,910
and adoption

318
00:11:40,370 --> 00:11:41,830
will proceed reflection.

319
00:11:42,690 --> 00:11:43,509
Most importantly,

320
00:11:44,129 --> 00:11:47,250
those filters can gradually direct the flow of

321
00:11:47,250 --> 00:11:49,509
conversation, emphasizing some questions

322
00:11:50,044 --> 00:11:53,404
while quietly discouraging others. I can't emphasize how

323
00:11:53,404 --> 00:11:55,725
important and dangerous these filters are and may

324
00:11:55,725 --> 00:11:56,225
become.

325
00:11:56,605 --> 00:12:00,044
When systems rewrite rather than carefully amend existing

326
00:12:00,044 --> 00:12:03,105
work, portions of the original meaning can quietly

327
00:12:03,245 --> 00:12:03,745
disappear.

328
00:12:04,139 --> 00:12:06,399
The danger isn't censorship or deception.

329
00:12:06,779 --> 00:12:10,059
The danger is continuous optimization where erosion of

330
00:12:10,059 --> 00:12:12,879
skills slowly becomes framed as efficiency.

331
00:12:13,580 --> 00:12:16,220
And that matters in medicine because the way

332
00:12:16,220 --> 00:12:18,399
knowledge is framed ultimately shapes

333
00:12:18,764 --> 00:12:20,545
how diseases are defined,

334
00:12:21,004 --> 00:12:21,504
classified,

335
00:12:22,045 --> 00:12:22,785
and treated.

336
00:12:23,884 --> 00:12:26,365
Definitely. And I like what you said about

337
00:12:26,365 --> 00:12:27,425
what you're doing

338
00:12:27,965 --> 00:12:29,884
at ISAS and then you in the journal

339
00:12:29,884 --> 00:12:30,865
and how you're,

340
00:12:31,644 --> 00:12:32,384
kind of

341
00:12:32,759 --> 00:12:35,399
making guidelines based off of what we know

342
00:12:35,399 --> 00:12:38,040
now with AI. Can you dive deeper into

343
00:12:38,040 --> 00:12:41,019
how all this relates specifically to spine care?

344
00:12:41,559 --> 00:12:44,519
Sure. Well, in spine care, this matters because

345
00:12:44,519 --> 00:12:47,065
chronic low back pain, as I mentioned, is

346
00:12:47,065 --> 00:12:49,384
not a single disease, for instance. It includes

347
00:12:49,384 --> 00:12:50,365
multiple phenotypes,

348
00:12:51,304 --> 00:12:55,245
discogenic pain, retibrogenic pain, sacroiliac dysfunction,

349
00:12:55,784 --> 00:12:57,325
multifidus muscle failure,

350
00:12:58,184 --> 00:13:00,205
neuropathic and radicular pain,

351
00:13:00,730 --> 00:13:02,830
and central sensitization patterns.

352
00:13:03,529 --> 00:13:06,830
AI can help us recognize these phenotypes earlier

353
00:13:06,889 --> 00:13:07,870
and more consistently.

354
00:13:08,730 --> 00:13:12,910
But if surrounding systems favor simplified treatment pathways

355
00:13:13,674 --> 00:13:15,934
or if they narrow outcome metrics,

356
00:13:16,475 --> 00:13:18,495
those distinctions can gradually

357
00:13:18,795 --> 00:13:20,095
disappear from the conversation.

358
00:13:20,795 --> 00:13:23,855
So the issue isn't simply whether AI helps

359
00:13:23,915 --> 00:13:27,695
analyze data. It's whether physicians remain actively

360
00:13:28,154 --> 00:13:28,654
engaged

361
00:13:29,409 --> 00:13:32,370
in shaping how knowledge itself is generated and

362
00:13:32,370 --> 00:13:32,870
applied.

363
00:13:33,889 --> 00:13:36,129
Got it. And doctor Lohr, I want to

364
00:13:36,129 --> 00:13:38,870
also touch on this idea of machine memory

365
00:13:39,169 --> 00:13:42,595
versus human meaning. Can you dive into that

366
00:13:42,595 --> 00:13:44,194
and just, you know, the kind of the

367
00:13:44,194 --> 00:13:46,375
tensions that exist in that realm?

368
00:13:46,834 --> 00:13:47,975
Okay. Well,

369
00:13:48,355 --> 00:13:51,095
speaking to that tension as well, artificial,

370
00:13:51,954 --> 00:13:54,214
intelligence systems accumulate information

371
00:13:54,754 --> 00:13:55,254
indefinitely.

372
00:13:56,115 --> 00:13:57,334
They do not forget.

373
00:13:58,410 --> 00:13:58,910
Humans

374
00:13:59,690 --> 00:14:00,669
are very different.

375
00:14:01,610 --> 00:14:03,070
Our memory is finite.

376
00:14:03,769 --> 00:14:06,730
We cannot retain everything, and so we must

377
00:14:06,730 --> 00:14:08,990
choose what is worth remembering.

378
00:14:09,850 --> 00:14:12,269
What matters is remembering the sequence.

379
00:14:13,194 --> 00:14:15,134
What humans did before systems,

380
00:14:15,674 --> 00:14:17,615
when judgment began to be deferred,

381
00:14:18,154 --> 00:14:19,134
and when dependence

382
00:14:19,434 --> 00:14:19,934
quietly

383
00:14:21,115 --> 00:14:22,095
replace skill.

384
00:14:22,475 --> 00:14:24,954
Maintaining skill will be crucial if humans are

385
00:14:24,954 --> 00:14:26,014
to provide resilience

386
00:14:26,660 --> 00:14:28,680
for this new AI ecosystem.

387
00:14:30,180 --> 00:14:31,000
That act

388
00:14:31,379 --> 00:14:32,120
of selection

389
00:14:32,500 --> 00:14:33,000
deciding

390
00:14:33,460 --> 00:14:34,600
which observations

391
00:14:35,220 --> 00:14:37,860
matter enough to carry forward is actually a

392
00:14:37,860 --> 00:14:41,000
fundamental part of how knowledge develops in medicine.

393
00:14:42,075 --> 00:14:45,274
AI may store and synthesize vast amounts of

394
00:14:45,274 --> 00:14:45,774
information,

395
00:14:46,475 --> 00:14:49,134
but physicians still determine which patterns are meaningful

396
00:14:49,195 --> 00:14:50,415
and which insights

397
00:14:50,795 --> 00:14:54,075
deserve to shape future care. AI remembers everything,

398
00:14:54,075 --> 00:14:54,894
but bureaucracies

399
00:14:56,240 --> 00:14:59,700
bureaucracies, they decide what is allowed to matter.

400
00:15:00,320 --> 00:15:01,460
Artificial intelligence

401
00:15:02,159 --> 00:15:04,019
didn't invent these these tendencies.

402
00:15:04,559 --> 00:15:06,339
It simply accelerated them.

403
00:15:06,639 --> 00:15:09,274
That's very well said, doctor L'Oreal. And before

404
00:15:09,274 --> 00:15:11,134
we wrap up, if you had to summarize

405
00:15:11,195 --> 00:15:13,134
a key lesson for physicians

406
00:15:13,674 --> 00:15:15,615
about AI, what would it be?

407
00:15:15,915 --> 00:15:19,134
Okay. Well, artificial intelligence is gonna become

408
00:15:19,754 --> 00:15:22,095
extraordinarily good in answering questions.

409
00:15:23,110 --> 00:15:25,290
The future of medicine will depend on whether

410
00:15:25,429 --> 00:15:28,009
physicians continue to ask the right ones.

411
00:15:28,550 --> 00:15:30,570
Technology can analyze the data.

412
00:15:31,029 --> 00:15:33,210
Physicians still will have to recognize,

413
00:15:34,470 --> 00:15:35,210
the meaning.

414
00:15:36,065 --> 00:15:38,325
The real risk isn't that AI

415
00:15:38,865 --> 00:15:39,605
will replace

416
00:15:40,144 --> 00:15:40,644
physicians.

417
00:15:41,585 --> 00:15:42,485
The real risk

418
00:15:43,184 --> 00:15:44,965
is that it will slowly replace

419
00:15:45,585 --> 00:15:46,884
the habits of thinking

420
00:15:47,519 --> 00:15:48,820
that make physicians

421
00:15:50,000 --> 00:15:50,500
necessary.

422
00:15:51,360 --> 00:15:52,660
And in that vacuum,

423
00:15:53,679 --> 00:15:56,160
no one will be left to recognize or

424
00:15:56,160 --> 00:15:57,700
remember the meaning.

425
00:15:58,559 --> 00:16:01,095
Very well said. Well, doctor Lorio, thank you

426
00:16:01,095 --> 00:16:03,274
again for joining us in the podcast today.

427
00:16:03,815 --> 00:16:05,815
It's a great conversation, and I hope to

428
00:16:05,815 --> 00:16:07,894
connect again in the future. Thank you, Carly.

429
00:16:07,894 --> 00:16:09,815
I appreciate this conversation and time to be

430
00:16:09,815 --> 00:16:10,475
with you.