Modern WisdomArtificial Intelligence, Big Data & China | Martin Schmalz | Modern Wisdom Podcast 144
CHAPTERS
- 0:00 – 0:56
How unexpected data (like location) predicts credit risk
The conversation opens with a provocative example: people sleeping in multiple locations can correlate with loan default risk. Schmalz uses it to illustrate a central theme—modern AI is largely about prediction using unconventional behavioral data.
- •Location patterns can be highly predictive of loan default
- •Colorful causal stories (affairs → divorce → default) may be unverifiable, but correlations still get used
- •Data value often comes from signals people don’t realize they’re emitting
- 0:56 – 2:12
Schmalz’s path: from mechanical engineering to finance + AI economics
Schmalz explains his transition from engineering to economics and ultimately becoming a finance professor. He describes why he started teaching Python and data-driven strategy to bridge the gap between business education and industry demands.
- •Shift from engineering to understanding financial/economic systems
- •Motivation to modernize MBA skillsets beyond Excel
- •Focus on economics of data-driven business models and platform success
- 2:12 – 7:21
Why companies need “translators” between data science and strategy
Schmalz argues that technical teams and economic/strategy teams often fail to communicate effectively. Without a bridge, companies build data infrastructure without a clear value proposition or strategy.
- •Data scientists and economists often lack shared language (features vs variables, etc.)
- •Executives may demand “do AI” without knowing what outcome they want
- •Engineers optimize prediction, but business value requires economic framing
- •The book exists to provide structured thinking about AI-era business models
- 7:21 – 10:10
What AI really does: cheap, scalable prediction (not “thinking”)
Schmalz reframes most modern AI as prediction machines trained on past data. He emphasizes the practical drivers—cheap storage, cheap compute, and massive data—rather than sci‑fi notions of computers thinking like humans.
- •95% of deployed ML is prediction from historical data
- •More data + cheaper compute makes prediction faster, better, cheaper than humans
- •AI’s capabilities are narrow; ‘intelligence’ is mostly a misnomer
- •Examples: ad targeting, willingness-to-pay estimates, behavioral inference
- 10:10 – 15:33
Generic vs non-generic prediction: where humans still dominate
They distinguish repeatable, data-rich prediction tasks from novel, unprecedented ones. Schmalz argues that creativity, synthesis, and forecasting disruptions without historical precedent remain human strengths—especially in executive decision-making.
- •Loan officers as an example of a prediction job being automated
- •Generic prediction (repeatable patterns) is easiest to replace
- •Non-generic prediction (new products, new markets) needs human judgment
- •Creativity and synthesis are ‘complements’ to machine prediction
- 15:33 – 17:18
Behavioral micro-signals: typing speed, typos, and insurance/credit models
Schmalz shares examples from China showing how tiny interaction details become predictive features. Banks and insurers may infer fraud risk, intelligence, carefulness, or default probability from how users fill online forms.
- •Form-fill speed and typo rate can predict fraud/default/insurance risk
- •Companies collect new data types once they prove predictive
- •Prediction often matters more than the ‘story’ explaining why it works
- •China is presented as the frontier of aggressive feature collection
- 17:18 – 23:25
Dynamic pricing, willingness to pay, and the ethics of personalization
A Skyscanner anecdote leads into how firms estimate willingness to pay and may adjust prices accordingly. They discuss the consumer backlash risk, and why people react differently to price discrimination versus “personalized discounts.”
- •Sites can infer desperation/interest and adjust pricing
- •Personalized pricing can damage trust and retention
- •People hate paying more than others but love getting a special discount
- •Ethics becomes a business risk, not just a moral question
- 23:25 – 26:37
Location data, ride-hailing, and why Uber/Didi move into lending
Schmalz explains how ride-hailing platforms can infer income, lifestyle, and liquidity from pickup/drop-off behavior. This makes expanding into financial products (like lending) economically logical—and China often previews what the West does later.
- •Sleeping location patterns as a strong default predictor
- •Home/work and neighborhood shifts reveal income changes
- •Restaurants visited can act as proxies for spending power
- •Didi launched lending; Uber followed similar logic
- •“Watch China to predict the future”
- 26:37 – 36:42
Why China is ahead: super-apps, lax constraints, and engineering scale
Schmalz outlines several reasons China leads in AI deployment: population-scale data, integrated super-app ecosystems like WeChat, fewer privacy roadblocks, and huge engineering investment. Cross-domain data fusion creates major predictive advantages.
- •Large population enables massive datasets in a unified market
- •WeChat-style ‘super apps’ merge payments, healthcare, insurance, messaging
- •Cross-business data combination improves prediction (e.g., doctor bookings → life expectancy)
- •Fewer GDPR-like constraints; faster experimentation
- •Enormous engineering headcount (e.g., Ping An’s AI workforce)
- 36:42 – 41:35
Antitrust, GDPR, and regulators blocking data mergers (Facebook example)
They shift to how competition law intersects with privacy. Schmalz explains the German competition authority’s argument that Facebook’s dominance limits privacy-respecting alternatives, justifying restrictions on merging WhatsApp/Instagram/Facebook data.
- •Antitrust focuses on ‘abuse of dominance,’ not dominance itself
- •Privacy can be framed as a competition dimension (users would choose privacy if real alternatives existed)
- •Bundeskartellamt action restricting cross-platform data merging
- •Europe’s regulatory posture differs sharply from the US
- 41:35 – 51:37
The privacy–convenience tradeoff and the coming societal decision
They explore how user preferences shift when convenience is high (biometrics at borders, frictionless payments, face-scan checkout). Schmalz argues society will likely move toward more data-driven convenience, but speed and outcomes depend on politics, competition, and public backlash.
- •Convenience often outweighs abstract privacy concerns in practice
- •Transparency matters: users react strongly when they ‘see behind the curtain’
- •Data games are dynamic: people will ‘game’ incentives (Fitbit hacks), firms respond with new signals
- •More winners and losers → inequality and lobbying pressures
- •China as a preview: face-scan payments and low-friction daily life
- 51:37 – 57:36
AGI vs reality: why ‘boring statistics’ drives profits (and jobs) now
They contrast sensational AGI narratives with what’s actually economically rewarded: large-scale prediction, econometrics, and data science that improve pricing and decision-making. Schmalz highlights how market caps reflect expectations that these business models will keep paying off.
- •AGI timelines keep slipping; speculation dominates headlines
- •Real-world impact is prediction on behavioral datasets
- •Amazon hires economists to estimate demand elasticities and optimize pricing
- •Financial markets already transformed by model-driven prediction
- •AI is not a fad unless data collection/processing becomes expensive (mainly via regulation)
- 57:36 – 1:09:34
Business models powered by data extraction—and closing thoughts
Schmalz explains how many modern products function as data collection fronts, with value created by selling or leveraging data rather than the nominal service (e.g., scooters, apps). They close with where to find the book and acknowledgments to co-author Yuri.
- •Data itself becomes the product; nominal products can be loss leaders
- •Data aggregators create a value chain feeding ad targeting and pricing models
- •Legal constraints (GDPR) can reshape which models survive
- •Practical takeaway: you’re not ‘immune,’ just potentially behind the curve
- •Wrap-up: book plug and credits
