Nov 23, 2015 · 22m · mad
Black Boxes and Unicorns - DataRobot CEO Jeremy Achin
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this FirstMark DataDrivenNYC presentation, DataRobot CEO Jeremy Achin presents 'Black Boxes and Unicorns,' explaining how algorithm-agnostic interpretability tools demystify complex machine learning models while automated platforms and practical education can solve the data science talent shortage.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 4.6% of the talking time here. How this is scored →
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Achin explicitly mocks standard industry reliance on linear regression, defining compliance as a desperate attempt to defend outdated habits while acknowledging he may offend audience members.
Hardest push from Matt ▶ 17:30 Prompting Product ExplanationHost Matt Turck intervenes post-presentation to gently push Achin into explaining DataRobot's actual commercial product after Achin intentionally avoided a sales pitch.
Biggest teaching moment ▶ 8:15 Explaining Partial Dependence PlotsAchin educates the room on how partial dependence curves isolate non-linear feature relationships inside black-box models, demonstrating how modern machine learning achieves interpretability without linear assumptions.
Matt holds his own ▶ 17:30 Summarizing Automated Model SelectionHost Matt Turck demonstrates his domain knowledge by accurately summarizing how DataRobot automates testing multiple algorithms against arbitrary dataset types.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Jeremy Achin Background and DataRobot Company Overview | 0 | 2 | 1 | 0 | Achin provides his background as an insurance data scientist and Kaggle competitor while introducing DataRobot and citing Leo Breiman's 2001 paper on statistical modeling. Because this is a keynote monologue, the host does not participate. | |
| Redefining Black Boxes and Model Interpretability | 0 | 4 | 4 | 0 | Achin aggressively redefines 'black box' as a term used by people afraid of technology who want to maintain 20-year-old habits. He outlines technical criteria for model transparency that apply across complex ensemble algorithms. | |
| Case Study: Hospital Readmission and Partial Dependence | 0 | 5 | 1 | 0 | Achin presents a hospital readmission case study to demonstrate how partial dependence plots reveal marginal non-linear feature effects in complex models. He explains how this approach provides model interpretability without relying on simple linear regression. | |
| Critiquing Compliance and Arbitrary Representations of Reality | 0 | 6 | 7 | 0 | Achin heavily critiques traditional statisticians and compliance teams, defining compliance as a last resort excuse to defend linear regression status quo. He uses humorous parables like statisticians in a bar and killer potatoes to show how regression coefficients can create arbitrary representations of reality. | |
| Democratizing Data Science and Presentation Takeaways | 0 | 4 | 3 | 0 | Achin critiques traditional data science education pipelines for dropping students before teaching practical application. He argues that automated modeling tools will solve the data scientist shortage by enabling practical application first. | |
| Host Q&A and Event Conclusion | 3 | 3 | 1 | 2 | Host Matt Turck opens Q&A by concisely summarizing DataRobot's automated model selection value proposition and inviting Achin to explain the product. Achin and the audience discuss Kaggle performance and unexpected algorithmic results. |