537: Data Science Trends for 2022

Episode Summary

Executive Summary: Sadie St. Lawrence and John Crohn review how 2021 predictions played out and forecast 2022 data science trends: AutoML adoption, deepfake risks and uses, scalable AI production practices, the remote-work talent market, blockchain/NFTs as data infrastructure, and broadening data literacy. The episode frames data science as increasingly automated, distributed, and socially consequential.

Main Topics: 2021 Prediction Retrospective: They revisit prior predictions and note strong accuracy on explainable AI growth, federated learning/privacy shifts, and PyTorch overtaking TensorFlow in popularity. AutoML and Citizen Data Scientists: AutoML is expected to accelerate, enabling non-experts to build models while raising concerns about bias, governance, and production deployment. Deepfakes and AI Ethics: GAN-powered deepfakes are becoming easier and more realistic, with positive creative uses but major risks around identity theft, misinformation, and non-consensual content. Scalable AI in Production: Four core principles are emphasized: cloud-first infrastructure, standardized/automated workflows, performance monitoring, and traceability for models in production. Remote Economy and Talent Competition: Remote work has expanded access to global talent and increased competition for data science jobs, making retention, one-on-ones, and flexible hiring strategies critical. Blockchain, NFTs, and Data Provenance: Blockchain is presented as an immutable ledger that could improve auditability of data, digital ownership, and even help combat identity theft in a metaverse-like world. Data Literacy as a Workforce Necessity: Data literacy is framed as essential beyond technical teams, extending to farmers, manufacturing workers, and business roles as AI and sensors permeate every industry.

Key Arguments: Explainable AI and bias-focused research are no longer niche; they are becoming mainstream, evidenced by conference policies and a surge in publications. AutoML will broaden access to machine learning, but organizations must guard against biased inputs, poor governance, and careless deployment. Data scientists should rethink their value proposition as model selection becomes automated; business understanding, data quality, and ethical oversight become more important. Deepfake technology has legitimate creative and educational uses, but its ease of creation and realism create urgent social and legal harms. Scalable AI requires operational discipline: cloud infrastructure, automation, monitoring, and traceability are as important as model accuracy. Remote work has permanently widened the talent market, intensifying competition and making employee retention and distributed collaboration central to success. Blockchain could provide immutable provenance for digital assets and data, creating richer, more trustworthy records for analysis and ownership. Data literacy will determine who can benefit from analytics and AI; it must extend to the entire workforce, not just data teams. As compute scaling slows, AI progress may depend more on smarter methods (Bayesian/causal approaches) and eventually quantum computing.

Data Points: Explainable AI papers since 2020: 17,000 - Google Scholar search mentioned by John when assessing the growth of explainable AI research. Explainable AI papers in last year: 12,000 - Of the 17,000 explainable AI papers since 2020, 12,000 were in the most recent year. Women in Data community size: 20,000+ women across 17 countries - Sadie described the reach of WomenInData.org. Students reached by Sadie's content: 300,000+ students - Used to introduce Sadie’s influence as a data science educator. Global workforce data literacy: 21% - The episode cites this as the current level of data literacy worldwide. US job openings: 11 million - Used to illustrate the post-pandemic labor market and remote-work competition. Pre-pandemic peak US job openings: 7 million - Compared with the current 11 million openings to show a ~50% increase. US monthly quits record: 4.4 million - September U.S. quits were cited as a record-high indicator of labor churn. AutoML search results: First 6 results were ads - Sadie’s Google search on AutoML showed heavy vendor competition and paid promotion. Minimum images for deepfake training: ~300 images - Sadie noted that relatively few images can already be enough to train a deepfake model.

Pivotal Quotes: "We’re automating our job at the same time as we’re doing it." — Sadie St. Lawrence: On AutoML replacing parts of the traditional data scientist workflow and shifting the skill set needed in the field. "The biggest factor to affect weather on our planet is from the sun." — Sadie St. Lawrence: Her analogy for separating micro-trends from macro-trends when thinking about long-term data science shifts. "This is not just like your marketing friend needs to have data literacy. No, farmers need to have data literacy." — Sadie St. Lawrence: On why data literacy must extend across blue-collar and white-collar roles, not only business teams.

Implications: Listeners should expect more automation, stronger ethical scrutiny, and broader access to analytics, but also greater need for governance, digital identity protections, and workforce upskilling as data science becomes embedded everywhere.