WEBVTT

1
00:00:01.159 --> 00:00:06.200
<v Speaker 1>Welcome to the Sentient Code, where intelligence is engineered, autonomy

2
00:00:06.280 --> 00:00:10.439
<v Speaker 1>is emerging, and a line between human and machine grows thinner.

3
00:00:10.800 --> 00:00:15.359
<v Speaker 1>Each episode, we decode the algorithms, explore the robotics, and

4
00:00:15.439 --> 00:00:21.920
<v Speaker 1>examine the ideas shaping the future of artificial minds.

5
00:00:23.839 --> 00:00:27.559
<v Speaker 2>A publication in Cell Reports Medicine dated February seventeenth, twenty

6
00:00:27.640 --> 00:00:32.560
<v Speaker 2>twenty six establishes a comprehensive benchmark for evaluating large language models.

7
00:00:33.000 --> 00:00:37.280
<v Speaker 2>The research specifically focuses on predictive modeling and biomedical research,

8
00:00:37.479 --> 00:00:40.079
<v Speaker 2>with a concentrated application in reproductive health.

9
00:00:40.399 --> 00:00:43.520
<v Speaker 3>The investigation was jointly conducted by the University of California,

10
00:00:43.520 --> 00:00:48.880
<v Speaker 3>San Francisco, specifically the Batcar Computational Health Sciences Institute and

11
00:00:48.920 --> 00:00:52.960
<v Speaker 3>Wayne State University. The core thesis examines the efficacy of

12
00:00:53.000 --> 00:00:57.799
<v Speaker 3>generative artificial intelligence in analyzing highly complex medical data.

13
00:00:57.560 --> 00:01:00.759
<v Speaker 2>Sets, and it compares that efficacy directly again the output

14
00:01:00.799 --> 00:01:04.280
<v Speaker 2>of traditional human research teams. The primary finding of this

15
00:01:04.319 --> 00:01:08.239
<v Speaker 2>benchmarking study is that generative AI tools successfully developed accurate

16
00:01:08.280 --> 00:01:09.840
<v Speaker 2>prediction models for preterm birth.

17
00:01:10.000 --> 00:01:13.159
<v Speaker 3>Furthermore, the AI systems executed the data analysis at a

18
00:01:13.159 --> 00:01:18.439
<v Speaker 3>significantly accelerated rate compared to human teams. In specific evaluated instances,

19
00:01:18.480 --> 00:01:22.680
<v Speaker 3>the AI actually outperformed models generated by human computer scientists.

20
00:01:22.840 --> 00:01:25.079
<v Speaker 2>A critical variable in this execution is what we must

21
00:01:25.079 --> 00:01:28.879
<v Speaker 2>categorize as the junior researcher phenomenon. The AI assisted models

22
00:01:28.879 --> 00:01:31.319
<v Speaker 2>were not generated by senior bioinformaticians.

23
00:01:31.560 --> 00:01:35.359
<v Speaker 3>They were successfully generated by a master's student, Rubin Sarwall

24
00:01:35.719 --> 00:01:37.760
<v Speaker 3>and a high school student, Victor Tarka.

25
00:01:38.439 --> 00:01:42.599
<v Speaker 2>This contrasts fundamentally with the decades of senior expertise typically

26
00:01:42.640 --> 00:01:46.719
<v Speaker 2>required to execute multi dimensional data analysis and bioinformatics.

27
00:01:47.359 --> 00:01:50.840
<v Speaker 3>To contextualize the significance of this methodology, you must first

28
00:01:50.920 --> 00:01:54.799
<v Speaker 3>understand the clinical challenge. The medical problem being modeled is

29
00:01:54.959 --> 00:01:55.879
<v Speaker 3>preterm birth.

30
00:01:56.079 --> 00:02:00.640
<v Speaker 2>Epidemiologically, preterm birth remains the leading cause of newborn mortality globally.

31
00:02:00.920 --> 00:02:03.719
<v Speaker 3>It is also the primary driver of long term cognitive

32
00:02:03.760 --> 00:02:07.840
<v Speaker 3>and motor disabilities in surviving infants. In terms of statistical scope,

33
00:02:07.959 --> 00:02:11.280
<v Speaker 3>we observe approximately one thousand incidences per day in the

34
00:02:11.360 --> 00:02:12.080
<v Speaker 3>United States.

35
00:02:12.319 --> 00:02:15.879
<v Speaker 2>Despite that high incidence rate, the etiological obscurity of preterm

36
00:02:15.919 --> 00:02:20.199
<v Speaker 2>birth presents a massive clinical hurdle. The underlying biological triggers

37
00:02:20.199 --> 00:02:23.159
<v Speaker 2>that initiate early labor remains scientifically opaque.

38
00:02:23.319 --> 00:02:26.759
<v Speaker 3>Because the mechanisms are not fully understood, clinicians cannot simply

39
00:02:26.800 --> 00:02:30.919
<v Speaker 3>observe standard vital signs to predictorily labor. This necessitates highly

40
00:02:30.919 --> 00:02:35.080
<v Speaker 3>advanced data analysis to identify subtle, complex predictive markers within

41
00:02:35.159 --> 00:02:36.120
<v Speaker 3>maternal biology.

42
00:02:36.280 --> 00:02:39.439
<v Speaker 2>The data set utilized for this benchmarking standard represents a

43
00:02:39.479 --> 00:02:43.919
<v Speaker 2>significant layer of complexity. The model's evaluated microbiome.

44
00:02:43.439 --> 00:02:48.680
<v Speaker 3>Data, specifically vaginal microbiome data collected from approximately twelve hundred

45
00:02:48.719 --> 00:02:52.479
<v Speaker 3>pregnant women. The objective was to track the microbial environment

46
00:02:52.520 --> 00:02:54.439
<v Speaker 3>throughout the pregnancy through to delivery.

47
00:02:54.800 --> 00:02:57.680
<v Speaker 2>The primary challenge of this data set is its heterogeneity.

48
00:02:58.039 --> 00:03:01.439
<v Speaker 2>The data was not collected in a singles standardized clinical trial.

49
00:03:01.840 --> 00:03:04.360
<v Speaker 2>It was aggregated from nine separate studies.

50
00:03:04.479 --> 00:03:07.800
<v Speaker 3>When you aggregate data from nine distinct studies, you introduce

51
00:03:07.879 --> 00:03:12.560
<v Speaker 3>massive variance in collection protocols, sequencing technologies, and temporal measurements.

52
00:03:12.960 --> 00:03:16.919
<v Speaker 3>Doctor Timiko Tioskotsky asserts that pooling experiences in sharing open

53
00:03:17.000 --> 00:03:20.240
<v Speaker 3>data is an absolute requirement to achieve statistical power in

54
00:03:20.280 --> 00:03:21.240
<v Speaker 3>reproductive research.

55
00:03:21.439 --> 00:03:25.319
<v Speaker 2>However, that pooled open data generates a computational bottleneck. The

56
00:03:25.360 --> 00:03:29.280
<v Speaker 2>sheer volume and the architectural complexity of vaginal microbiome data

57
00:03:29.560 --> 00:03:34.199
<v Speaker 2>render traditional manual analysis methods intensely slow and resource heavy.

58
00:03:34.280 --> 00:03:38.080
<v Speaker 3>Doctor Marina Serota accurately identifies the process of building analysis

59
00:03:38.120 --> 00:03:41.360
<v Speaker 3>pipelines as a primary bottleneck in modern data science.

60
00:03:41.639 --> 00:03:45.280
<v Speaker 2>A pipeline in this context refers to the sequential series

61
00:03:45.319 --> 00:03:49.039
<v Speaker 2>of programmatic instructions required to clean the data, normalize the

62
00:03:49.120 --> 00:03:53.120
<v Speaker 2>variables across those nine disparate studies, and extract the relevant

63
00:03:53.120 --> 00:03:54.319
<v Speaker 2>biological features.

64
00:03:54.560 --> 00:03:59.439
<v Speaker 3>Constructing those pipelines manually requires extensive coding, iterative debugging, and

65
00:03:59.520 --> 00:04:04.879
<v Speaker 3>constant syntax correction. This procedural latency directly delays the translation

66
00:04:04.960 --> 00:04:08.879
<v Speaker 3>of raw data into clinical applications, thereby delaying patient care.

67
00:04:09.280 --> 00:04:13.560
<v Speaker 2>To measure the AIS capability against this bottleneck, the researchers

68
00:04:13.639 --> 00:04:17.800
<v Speaker 2>utilized a rigorous benchmarking standard known as the DREAM Challenge.

69
00:04:18.000 --> 00:04:21.519
<v Speaker 3>DREAM stands for the Dialogue on Reverse Engineering Assessment and Methods.

70
00:04:21.959 --> 00:04:24.439
<v Speaker 3>It is a highly respected computational competition.

71
00:04:24.639 --> 00:04:27.439
<v Speaker 2>It serves as the definitive baseline for human performance in

72
00:04:27.439 --> 00:04:29.720
<v Speaker 2>this field. The scope of the human effort in the

73
00:04:29.720 --> 00:04:33.720
<v Speaker 2>original DREAM competition involved over one hundred global research groups.

74
00:04:33.920 --> 00:04:37.360
<v Speaker 3>These groups consisted of highly credential scientists competing to design

75
00:04:37.439 --> 00:04:41.600
<v Speaker 3>machine learning algorithms capable of predicting pre term birth based

76
00:04:41.639 --> 00:04:44.759
<v Speaker 3>on that exact same vaginal microbiome data set.

77
00:04:44.920 --> 00:04:47.800
<v Speaker 2>The temporal metrics for the human baseline are vital to

78
00:04:47.839 --> 00:04:51.920
<v Speaker 2>the comparative analysis. While the initial computational challenges for the

79
00:04:52.000 --> 00:04:55.560
<v Speaker 2>dream teams were completed over a three month period.

80
00:04:55.480 --> 00:05:00.160
<v Speaker 3>The total compilation, verification, and publication of their combined findings

81
00:05:00.199 --> 00:05:03.639
<v Speaker 3>required nearly two years of sustained academic effort.

82
00:05:03.920 --> 00:05:07.600
<v Speaker 2>That two year life cycle establishes the human standard. The

83
00:05:07.680 --> 00:05:12.319
<v Speaker 2>collaborative framework between UCSF, Seroda Lab, and Wayne State University,

84
00:05:12.639 --> 00:05:15.879
<v Speaker 2>led by doctor ad L. Tarka, was designed to test

85
00:05:15.920 --> 00:05:20.319
<v Speaker 2>if AI could independently replicate or exceed that global human effort, and.

86
00:05:20.240 --> 00:05:23.800
<v Speaker 3>Crucially, whether the AI could accomplish this without extensive human

87
00:05:23.839 --> 00:05:28.199
<v Speaker 3>coding intervention. The experimental variables consisted of testing eight distinct

88
00:05:28.360 --> 00:05:29.839
<v Speaker 3>artificial intelligence systems.

89
00:05:30.000 --> 00:05:32.959
<v Speaker 2>These were large language models. The methodological input is the

90
00:05:32.959 --> 00:05:36.120
<v Speaker 2>most critical factor here. The human operators did not feed

91
00:05:36.160 --> 00:05:37.839
<v Speaker 2>the AI system's raw code.

92
00:05:37.959 --> 00:05:41.519
<v Speaker 3>They utilized strictly natural language prompts. The systems were given

93
00:05:41.560 --> 00:05:44.920
<v Speaker 3>plain language instructions combined with precise analytical guidance.

94
00:05:45.240 --> 00:05:49.480
<v Speaker 2>The prompts mimicked the exact objective parameters originally provided to

95
00:05:49.519 --> 00:05:53.439
<v Speaker 2>the Human Dream Challenge teams. The objective was to prompt

96
00:05:53.480 --> 00:05:56.360
<v Speaker 2>the system to generate the analytical pipeline from scratch.

97
00:05:56.600 --> 00:05:59.639
<v Speaker 3>This brings the analysis back to the personnel involved. The

98
00:05:59.680 --> 00:06:03.759
<v Speaker 3>AI assisted research was executed by a UCSF master's student

99
00:06:03.920 --> 00:06:05.680
<v Speaker 3>and a student from here On High school.

100
00:06:05.759 --> 00:06:09.079
<v Speaker 2>The core implication here is the demonstrated ability of non

101
00:06:09.120 --> 00:06:14.120
<v Speaker 2>expert programmers to generate functional, complex algorithmic code within minutes.

102
00:06:14.279 --> 00:06:18.079
<v Speaker 3>The task that as established by the bottleneck analysis typically

103
00:06:18.120 --> 00:06:23.279
<v Speaker 3>requires experienced bioinformatitions hours, days, or even weeks to manually

104
00:06:23.319 --> 00:06:24.199
<v Speaker 3>script and debug.

105
00:06:24.360 --> 00:06:28.319
<v Speaker 2>However, the performance analysis reveals significant failure points in the technology,

106
00:06:28.800 --> 00:06:32.079
<v Speaker 2>we must examine the success rates quantitatively. Of the eight

107
00:06:32.160 --> 00:06:35.680
<v Speaker 2>distinct AI systems tested, only four were capable of generating

108
00:06:35.800 --> 00:06:37.279
<v Speaker 2>usable functional code.

109
00:06:37.360 --> 00:06:41.600
<v Speaker 3>A fifty percent failure rate indicates high variability in model reliability.

110
00:06:42.199 --> 00:06:45.399
<v Speaker 3>The systems that failed produced code containing structural errors, or

111
00:06:45.399 --> 00:06:47.120
<v Speaker 3>they suffered from the hallucination risk.

112
00:06:47.439 --> 00:06:52.040
<v Speaker 2>Hallucinations in this context mean the AI generated syntactically plausible

113
00:06:52.120 --> 00:06:56.519
<v Speaker 2>code that called upon non existent software libraries or executed

114
00:06:56.560 --> 00:06:58.680
<v Speaker 2>incorrect mathematical normalizations.

115
00:06:58.879 --> 00:07:02.120
<v Speaker 3>But when evaluating the performance metrics of the four successful

116
00:07:02.120 --> 00:07:06.439
<v Speaker 3>AI models. The results are definitive. The successful generative AI

117
00:07:06.560 --> 00:07:10.759
<v Speaker 3>models matched the predictive accuracy of the consensus models built

118
00:07:10.759 --> 00:07:13.319
<v Speaker 3>by the one hundred human dream teams.

119
00:07:13.000 --> 00:07:17.720
<v Speaker 2>And in documented specific instances, the AI generated models demonstrated

120
00:07:17.759 --> 00:07:22.000
<v Speaker 2>statistically superior performance to the models created by the human scientists.

121
00:07:22.199 --> 00:07:26.240
<v Speaker 3>The computational tasks performed were divided into two primary categories.

122
00:07:26.720 --> 00:07:30.560
<v Speaker 3>Task A required the evaluation of the vaginal microbiome data

123
00:07:30.600 --> 00:07:34.439
<v Speaker 3>to identify specific biological indicators and patterns intrinsically linked to

124
00:07:34.480 --> 00:07:35.319
<v Speaker 3>preterm birth.

125
00:07:35.519 --> 00:07:38.839
<v Speaker 2>Task B shifted focus to a different biological medium. It

126
00:07:38.879 --> 00:07:42.120
<v Speaker 2>required the examination of maternal blood or placental samples.

127
00:07:42.279 --> 00:07:44.839
<v Speaker 3>The objective of task B was to utilize those samples

128
00:07:44.920 --> 00:07:48.120
<v Speaker 3>to estimate gestational age, an application commonly referred to as

129
00:07:48.160 --> 00:07:49.000
<v Speaker 3>pregnancy dating.

130
00:07:49.199 --> 00:07:52.759
<v Speaker 2>The clinical relevance of task B cannot be overstated. Accurate

131
00:07:52.839 --> 00:07:56.160
<v Speaker 2>dating of a pregnancy is the foundational metric for labor preparation.

132
00:07:56.560 --> 00:07:59.959
<v Speaker 3>If a clinician does not have an accurate gestational baseline,

133
00:08:00.279 --> 00:08:04.920
<v Speaker 3>in accuracies severely compromised clinical decision making. Interventions such as

134
00:08:05.040 --> 00:08:09.839
<v Speaker 3>administering quortacosteroids for fetal lung development rely entirely on precise

135
00:08:09.920 --> 00:08:10.879
<v Speaker 3>gestational dating.

136
00:08:11.360 --> 00:08:15.240
<v Speaker 2>Developing molecular models from blood and placental tissue to predict

137
00:08:15.240 --> 00:08:19.759
<v Speaker 2>that date is an extremely complex regression problem. The AI

138
00:08:19.879 --> 00:08:22.519
<v Speaker 2>handled this task with the same proficiency it applied to

139
00:08:22.600 --> 00:08:23.759
<v Speaker 2>the microbiome data.

140
00:08:24.120 --> 00:08:27.800
<v Speaker 3>When we conduct an operational efficiency and temporal analysis, the

141
00:08:27.879 --> 00:08:32.279
<v Speaker 3>contrast is stark. We establish the human timeline the original

142
00:08:32.399 --> 00:08:35.440
<v Speaker 3>Dream challenge process extended over a multi year.

143
00:08:35.320 --> 00:08:39.960
<v Speaker 2>Period, specifically nearly two years to achieve final verified results.

144
00:08:40.279 --> 00:08:43.600
<v Speaker 3>The AI timeline presents a fundamental disruption to that standard.

145
00:08:44.440 --> 00:08:48.720
<v Speaker 3>The entire generitive AI project, encompassing the initial conceptualization, the

146
00:08:48.759 --> 00:08:52.519
<v Speaker 3>pipeline generation, the data analysis, and the final submission of

147
00:08:52.519 --> 00:08:54.240
<v Speaker 3>the manuscript to sell reports.

148
00:08:53.919 --> 00:08:56.559
<v Speaker 2>Medicine, was completed in its entirety in six months.

149
00:08:56.799 --> 00:08:59.399
<v Speaker 3>This compression of the research life cycle is driven entirely

150
00:08:59.440 --> 00:09:01.720
<v Speaker 3>by the coach generation velocity of the language models.

151
00:09:01.840 --> 00:09:05.480
<v Speaker 2>In a manual coding process, the scientist writes syntax, encounters

152
00:09:05.480 --> 00:09:09.480
<v Speaker 2>in error, diagnoses the bug, rewrites the code, and re executes.

153
00:09:09.919 --> 00:09:13.759
<v Speaker 2>This cycle represents a massive expenditure of cognitive load and time.

154
00:09:14.000 --> 00:09:17.799
<v Speaker 3>The AI capability demonstrated in this study allows researchers to

155
00:09:17.879 --> 00:09:22.799
<v Speaker 3>bypass that latency. They generate complex analytical pipelines which short

156
00:09:22.840 --> 00:09:24.399
<v Speaker 3>technical prompts in minutes.

157
00:09:25.000 --> 00:09:28.080
<v Speaker 2>If an error occurs, the junior researchers simply fed the

158
00:09:28.240 --> 00:09:30.799
<v Speaker 2>error output back into the AI system via a natural

159
00:09:30.879 --> 00:09:35.080
<v Speaker 2>language prompt, and the system autonomously generated the corrected syntax.

160
00:09:35.480 --> 00:09:39.840
<v Speaker 3>This immediate syntax resolution validated their ability to run massive

161
00:09:39.840 --> 00:09:44.679
<v Speaker 3>statistical experiments, verify their results, and formalize their findings rapidly.

162
00:09:45.240 --> 00:09:48.080
<v Speaker 3>The coding latency was effectively eliminated from the workflow.

163
00:09:48.240 --> 00:09:52.480
<v Speaker 2>The implications for biomedical research are systemic. Doctor Auditarca observes

164
00:09:52.480 --> 00:09:55.399
<v Speaker 2>that this methodology drives a democratization of data science.

165
00:09:55.639 --> 00:09:59.399
<v Speaker 3>The requisite skills for biological data analysis are fundamentally shifting.

166
00:10:00.000 --> 00:10:02.879
<v Speaker 3>S urchre Possessing limited formal data science backgrounds will no

167
00:10:02.919 --> 00:10:07.279
<v Speaker 3>longer require extensive, heavily funded collaborations to process their data.

168
00:10:07.360 --> 00:10:10.720
<v Speaker 2>They will not require deep coding knowledge to extract clinical

169
00:10:10.799 --> 00:10:16.320
<v Speaker 2>value from multidimensional sets. The technology facilitates a cognitive focus shift.

170
00:10:16.960 --> 00:10:20.559
<v Speaker 3>Scientists can now shift their cognitive resources away from debugging

171
00:10:20.639 --> 00:10:25.679
<v Speaker 3>syntax algorithms and reallocate that mental bandwidth toward answering complex

172
00:10:25.759 --> 00:10:29.799
<v Speaker 3>biomedical questions and interpreting the biological relevance of the findings.

173
00:10:29.960 --> 00:10:34.159
<v Speaker 2>This shift, however, necessitates strict operational oversight. We noted the

174
00:10:34.200 --> 00:10:37.720
<v Speaker 2>fifty percent failure rate of the tested models. The hallucination

175
00:10:37.919 --> 00:10:39.960
<v Speaker 2>risk mandates rigorous protocols.

176
00:10:40.360 --> 00:10:43.919
<v Speaker 3>Acknowledging that AI systems can and will produce misleading results

177
00:10:44.039 --> 00:10:48.720
<v Speaker 3>or fail entirely. Makes human oversight a non negotiable component

178
00:10:48.919 --> 00:10:50.200
<v Speaker 3>of the biomedical workflow.

179
00:10:50.360 --> 00:10:53.759
<v Speaker 2>The generative AI does not replace the scientific method, nor

180
00:10:53.799 --> 00:10:56.679
<v Speaker 2>does it replace the scientist. It acts exclusively as an

181
00:10:56.679 --> 00:10:59.279
<v Speaker 2>accelerator for the technical execution of the analysis.

182
00:10:59.440 --> 00:11:02.320
<v Speaker 3>Human veryation is required to confirm that the generated code

183
00:11:02.360 --> 00:11:06.399
<v Speaker 3>is not only functionally executable, but biologically and statistically sound.

184
00:11:06.600 --> 00:11:10.039
<v Speaker 2>Looking toward future applications, the scalability of this methodology is

185
00:11:10.120 --> 00:11:13.320
<v Speaker 2>highly apparent. The prompt driven analysis is not restricted to

186
00:11:13.360 --> 00:11:14.320
<v Speaker 2>reproductive data.

187
00:11:14.519 --> 00:11:17.960
<v Speaker 3>There is immense potential to apply this exact framework to

188
00:11:18.039 --> 00:11:23.600
<v Speaker 3>other complex medical data sets, oncology, neurology, immunology, any field

189
00:11:23.639 --> 00:11:25.840
<v Speaker 3>constrained by high dimensional data pipelines.

190
00:11:26.200 --> 00:11:29.639
<v Speaker 2>The ultimate goal is direct patient impact. By removing the

191
00:11:29.639 --> 00:11:34.440
<v Speaker 2>computational barriers that delay analysis, researchers can significantly accelerate the

192
00:11:34.480 --> 00:11:37.600
<v Speaker 2>discovery of diagnostic tools for vulnerable populations.

193
00:11:38.279 --> 00:11:42.600
<v Speaker 3>Specifically in context like newborn health, where early intervention dictates

194
00:11:42.639 --> 00:11:47.120
<v Speaker 3>long term survivability. Accelerating the data to insight pipeline directly

195
00:11:47.159 --> 00:11:48.759
<v Speaker 3>accelerates clinical implementation.

196
00:11:49.039 --> 00:11:54.279
<v Speaker 2>Synthesizing the core findings of this evaluation yields two primary conclusions. First,

197
00:11:54.360 --> 00:11:59.279
<v Speaker 2>the application of generative AI facilitates massive efficiency gains, reducing

198
00:11:59.320 --> 00:12:02.399
<v Speaker 2>standard bio medical research timelines from years to a matter

199
00:12:02.440 --> 00:12:03.000
<v Speaker 2>of months.

200
00:12:03.279 --> 00:12:07.559
<v Speaker 3>Second, the accuracy validation is robust. The AI models prove

201
00:12:07.679 --> 00:12:12.639
<v Speaker 3>comparable and in specific methodological instances, superior to human generated

202
00:12:12.639 --> 00:12:14.399
<v Speaker 3>models regarding pre term birth prediction.

203
00:12:14.559 --> 00:12:17.120
<v Speaker 2>The execution of this benchmark was supported by an academic

204
00:12:17.200 --> 00:12:20.000
<v Speaker 2>framework including the March of Dimes Prematurity Research.

205
00:12:19.679 --> 00:12:23.360
<v Speaker 3>Center, along with funding and structural support from IMPORT and

206
00:12:23.399 --> 00:12:25.799
<v Speaker 3>the NICHD Pregnancy Research Branch.

207
00:12:26.240 --> 00:12:31.200
<v Speaker 2>The integration of generative artificial intelligence into biomedical workflows represents

208
00:12:31.200 --> 00:12:34.799
<v Speaker 2>a fundamental structural shift in the methodology of medical research.

209
00:12:35.000 --> 00:12:40.080
<v Speaker 3>It initiates a permanent transition from labor intensive manual algorithmic

210
00:12:40.120 --> 00:12:43.480
<v Speaker 3>coding to high velocity, prompt driven data analysis.

211
00:12:43.639 --> 00:12:47.960
<v Speaker 2>Provided crucially that rigorous human verification protocols are maintained at

212
00:12:47.960 --> 00:12:49.279
<v Speaker 2>every stage of the pipeline.

213
00:12:49.320 --> 00:12:52.399
<v Speaker 3>When you evaluate the trajectory of clinical data science. The

214
00:12:52.440 --> 00:12:55.159
<v Speaker 3>removal of the coding barrier means that limitation is no

215
00:12:55.240 --> 00:12:59.559
<v Speaker 3>longer technical execution, but rather the quality of the biological hypothesis.