Nathan Benaich and Ian Hogarth · consulting-deck
2019 Air Street Capital The State of AI Report 2019
136 pages · 3 arc beats · 21 loops
2019 Air Street Capital The State of AI Report 2019
Nathan Benaich and Ian Hogarth arc beats above · slides in the middle · loops below · scroll → 21 LOOPS
1102030405060708090100110120130136
Deck intelligence map
3 coverage by narrative range · generated from this deck JSON
Narrative range 127 total
Metadata
Components
Metrics
Tools
Frameworks
Beats
Loops
The Facts (What) 88 slides 100% 88/88 slides 100% 88/88 slides · 523 hits — 0/88 slides
70.5% 62/88 slides · 112 hits 12.5% 11/88 slides · 12 hits 100% 88/88 slides · 176 hits 85.2% 75/88 slides The Implications (So What) 37 slides 100% 37/37 slides 100% 37/37 slides · 205 hits — 0/37 slides
62.2% 23/37 slides · 37 hits — 0/37 slides
100% 37/37 slides · 74 hits 78.4% 29/37 slides The Action (Now What) 2 slides 100% 2/2 slides 100% 2/2 slides · 3 hits — 0/2 slides
100% 2/2 slides · 3 hits — 0/2 slides
100% 2/2 slides · 4 hits 50% 1/2 slides Slide inventory
136 every slide · same image gating as the playbook
03
The slide serves as a table of contents and mission statement for the report.front_matter
05
The slide provides foundational definitions for the report's subject matter.present_framework
Open slide detailBeat · The Facts (What)
08
The slide uses a color-coded bar (green, orange, red) to represent the status of predictions, acting as a visual indicator of success.summarize
10
The chart shows the progression of various AI models' scores over time, highlighting the jump in performance with RND.illustrate_case
11
The slide highlights the transition from supervised learning to multi-agent reinforcement learning.illustrate_case
12
The chart shows Elo progression over training steps, comparing FTW (For The Win) agents against human benchmarks.illustrate_case
13
Includes a technical diagram of the OpenAI Five training architecture.illustrate_case
14
The chart shows a clear logarithmic relationship between training compute (PF/s days) and model performance.illustrate_case
15
The slide highlights a specific AI research methodology (Play-LMP) from the 2019 State of AI report.illustrate_case
16
The slide highlights two specific research examples: BLUE (UC Berkeley) and OpenAI's robotic hand manipulation system.illustrate_case
17
The slide uses technical diagrams to illustrate RL concepts like memory-based exploration and skill-based discriminator training.present_framework
18
The slide compares PlaNet's performance against established RL benchmarks (A3C, D4PG) using both a summary table and performance charts.illustrate_case
19
The slide describes the technical architecture and workflow of the Horizon platform.illustrate_case
20
The slide uses a process flow diagram to explain the technical methodology of AlphaFold.illustrate_case
Open slide detailBeat · The Facts (What)
21
The slide uses a bar chart with Elmo-themed icons to illustrate performance gains over previous state-of-the-art models.summarize
22
The slide serves as a demonstration of the capabilities of pretrained language models in 2019.illustrate_case
23
The slide tracks the evolution of NLP model performance against the GLUE benchmark, noting the introduction of SUPERglue due to rapid progress.summarize
24
The slide highlights a specific research paper's methodology and performance results.illustrate_case
25
The slide references the Cyc project and provides examples of inferential relations (PersonX leaves without PersonY, PersonX wins the title) and a visual map ofillustrate_case
26
The slide uses a process diagram to explain the technical workflow of Federated Learning rounds.present_framework
Open slide detailBeat · The Facts (What)
27
The diagram shows three layers of ML systems (Generic, Automotive, Network) and their data flow from physical domain to digital representation to model applicatpresent_solution
Open slide detailBeat · The Facts (What)
28
The slide illustrates a specific medical AI application developed by DeepMind and Moorfields Eye Hospital.illustrate_case
29
The slide presents ROC and PR curves comparing model performance against individual and average cardiologist performance across three specific arrhythmia classeillustrate_case
30
The slide uses ROC curves to compare model performance against human radiologists across five medical conditions.analyze_data
31
The slide details a specific medical research breakthrough involving brain-computer interfaces.illustrate_case
32
The slide compares a neural network (uNN) against a Support Vector Machine (SVM) baseline in a medical brain-computer interface application.illustrate_case
33
The slide highlights the application of neural networks and Monte Carlo tree search in chemistry.illustrate_case
34
The slide compares LEAF against existing benchmarks like CheXNet and Google AutoML.present_solution
35
Slide contains two tables comparing various neural network models across metrics like parameters, latency, and accuracy.analyze_data
36
The slide uses a before-and-after comparison to demonstrate technological progress in AI image generation.illustrate_case
37
The slide uses visual examples to demonstrate the progression of Generative Adversarial Networks (GANs) from facial manipulation to full-body generation.illustrate_case
38
The slide uses a real-world example to demonstrate the advancement of AI voice synthesis technology.illustrate_case
39
The slide showcases Mapillary's computer vision capabilities for autonomous driving/mapping.illustrate_case
40
The slide uses a combination of a bar chart for total volume and a line chart for topic share.analyze_data
Open slide detailBeat · The Facts (What)
43
The chart uses a vertical line to indicate the global average of 88% men.compare_peers
44
The chart illustrates the exponential growth of total compensation as seniority increases, driven primarily by stock grants.quantify_impact
45
Slide from the State of AI 2019 report.illustrate_case
46
The slide features portraits of the three Turing Award winners.illustrate_case
47
The slide uses two line charts to contrast quantity (papers published) vs quality (citation impact).analyze_data
48
The slide uses a case study approach to highlight institutional responses to AI talent needs.illustrate_case
49
The slide uses two line charts to compare growth rates relative to a base year (2012 for US, 2010 for non-US).analyze_data
50
The slide highlights a significant gender imbalance in AI academia.analyze_data
51
The chart displays raw applicant counts within a 100% stacked bar format.analyze_data
52
The slide uses a bubble map to visualize the concentration of AI researchers globally.analyze_data
53
The chart uses a 2x2 matrix structure to categorize countries based on talent mobility.compare_peers
54
The slide uses a screenshot of an article from The Information to support the claim.illustrate_case
56
The slide uses a combination of a combo chart (bar + line) and a 100% stacked bar chart to illustrate growth and regional composition.analyze_data
57
The chart uses a stacked area format to show cumulative growth, with individual company totals labeled at the end of the series.quantify_impact
58
The slide acts as a follow-up or validation of previous coverage by showing the financial progress of featured companies.cite_precedent
59
The slide highlights the exponential growth of UiPath as a case study for the RPA industry.illustrate_case
60
The slide uses a photo to illustrate the application of AI robotics in a retail environment.illustrate_case
61
The slide uses a visual comparison to demonstrate technological advancement over a decade.illustrate_case
62
The slide uses a visual representation of a factory floor to illustrate the concept of 'robots making robots'.illustrate_case
63
Part of the State of AI 2019 report.illustrate_case
64
The slide serves as a company spotlight within an industry report.illustrate_case
65
The slide uses a small-multiple chart layout to compare robot adoption across eight different industry sectors.analyze_data
66
The slide uses a combination of a bar chart (left) and a stacked bar chart (right) to illustrate infrastructure growth.quantify_impact
67
Part of the 'state of AI 2019' report.summarize
68
The slide uses a list format with company logos to illustrate the high financial barrier to entry in the autonomous vehicle sector.quantify_impact
69
The slide uses a combination of bar charts for quantitative growth and news headlines/quotes for qualitative industry sentiment.analyze_data
70
The chart uses a combination of bar (miles) and line (vehicles) to show the disparity in testing scale.analyze_data
71
Logarithmic scale used for the y-axis to visualize the extreme disparity.quantify_impact
72
The slide uses a timeline format to track progress against SAE automation levels (0-5).present_framework
73
Includes a Google 'Popular times' bar chart and a generic line chart showing yearly fluctuations.analyze_data
74
Part of a larger report on AI trends.illustrate_case
75
The diagram distinguishes between hydrology (blue) and ML (yellow) components in the flood forecasting process.present_framework
76
The slide uses a list-based structure to categorize industry applications of ML.present_solution
77
The slide uses a navigation bar at the top to indicate the current section is 'Industry'.illustrate_case
78
The chart illustrates the divergence between linear growth in human employment and exponential growth in data volume.size_opportunity
Open slide detailBeat · The Facts (What)
79
The slide uses a specific example (Primer) to illustrate the broader trend of machine learning in professional services.illustrate_case
Open slide detailBeat · The Facts (What)
80
The slide uses a timeline-like list format to present case studies of AI adoption in healthcare.illustrate_case
81
The slide highlights the business model of AI drug discovery startups (Atomwise and Exscientia) through partnerships with Charles River and Celgene.illustrate_case
82
The slide highlights the trend of AI-driven drug discovery through specific industry examples.illustrate_case
83
The slide uses data from the ZOE study (implied by logo) to demonstrate the efficacy of ML in personalized nutrition.analyze_data
84
The chart shows a significant divergence in growth rates between scientific publications and patent families starting around 2012.analyze_data
Open slide detailBeat · The Facts (What)
85
Data source: WIPO. The chart shows a significant divergence in patent volume for computer vision compared to other AI subfields.analyze_data
Open slide detailBeat · The Facts (What)
86
The slide highlights the disparity between mature cloud infrastructure revenue and nascent AI service revenue.analyze_data
87
The slide uses a logo-grid layout to categorize players into Tech Giants, IC Vendors, IP/Design Services, Startups (China), and Startups (Worldwide).present_framework
88
The table includes technical specifications (cores, year) and various performance metrics (CPU, Quant, FP16, FP32, Accuracy, AI-Score).compare_peers
89
The table displays 10 rows of data with 13 columns each, focusing on CPU, RAM, and various AI-specific performance metrics.compare_peers
90
The slide highlights the shift of AI computation to edge devices.illustrate_case
92
Includes a world map of 5G testing/research and a table showing mobile network evolution from 1G to 5G.establish_context
94
The slide acts as a section divider/intro for the 'Politics' section of the report.establish_context
95
Data source: Brookings Institute, State of AI 2019 report.analyze_data
96
Data from State of AI 2019 report by Brookings Institute.analyze_data
98
The slide highlights the failure of internal ethics boards due to lack of power, independence, and transparency.frame_problem
99
The chart shows probability curves over time, comparing public aggregate forecasts with those of CS/engineering degree holders.analyze_data
100
The chart uses a diverging color scheme to represent sentiment (blue for support, red for opposition, grey for neutral/unknown).analyze_data
101
The chart is a 100% stacked bar chart showing confidence levels across different actors.analyze_data
102
The chart uses ellipses to group data points, likely representing confidence intervals or clusters.analyze_data
103
Includes a quote from Angela Merkel regarding AI leadership and European values.establish_context
104
The slide highlights a specific national initiative as a case study for AI policy.illustrate_case
105
The slide highlights the EU's attempt to catch up to the US and China through specific funding programs and research initiatives.establish_context
106
Slide from the State of AI 2019 report.establish_context
107
The slide uses a two-column layout to explain the context of US trade disputes and the specific operational impact on the CVPR conference.illustrate_case
Open slide detailBeat · The Implications (So What)
108
The slide uses a military drone footage still to illustrate the context of Project Maven.illustrate_case
Open slide detailBeat · The Implications (So What)
109
The slide uses a photo to illustrate the context of surveillance in Xinjiang.frame_problem
110
The slide uses a visual example of high-zoom surveillance and a technical process diagram for lip-reading/AV analysis.diagnose
111
Includes a screenshot of a machine-generated text sample demonstrating the capability to generate misleading content.diagnose
112
Includes a reference to the 2019 Gabon presidential controversy.establish_context
114
Slide from State of AI 2019 report.illustrate_case
115
The slide uses a combination of a photo grid (illustrating the 28 members of Congress) and a bar chart to quantify the bias.illustrate_case
116
Includes a meme-style illustration featuring an anime character wearing a Yang 2020 hat.illustrate_case
Open slide detailBeat · The Implications (So What)
117
The slide serves as a visual catalog of organizations in the AI ethics space.establish_context
Open slide detailBeat · The Implications (So What)
120
Part of the 'China' section of the State of AI 2019 report.illustrate_case
Open slide detailBeat · The Implications (So What)
121
The slide highlights specific use cases of AI in agriculture (chicken, pig, and cockroach farming).illustrate_case
122
The slide uses a line chart to show historical growth and a stacked bar chart to show industry-specific regional market share.analyze_data
123
The chart illustrates the widening trade deficit in semiconductors for China, highlighting the strategic importance of onshoring.analyze_data
124
The chart shows a clear divergence in robot installation volume for China compared to other major regions between 2012 and 2017.analyze_data
125
The chart shows a dual-axis representation of growth between Q1 2017 and Q4 2018.illustrate_case
126
Includes a secondary visualization below the bar chart representing the value of deals using bubble sizes.analyze_data
127
The slide explains the 'dual cast' patent system in China which inflates total patent numbers compared to other countries.analyze_data
128
The slide uses two stacked bar charts to visualize the decay of patent maintenance over time.analyze_data
129
Data source: The Allen Institute. Includes trend projections.analyze_data
131
The slide uses a numbered list format to present forward-looking statements.summarize