- • Preamble
- • Introduction - Nike Tech Talk
- • Question: Company experience with ML in production?
- • Question: Impact on culture and organization?
- • Question: Who builds the ML models?
- • Question: Which methodologies apply?
- • Question: How are decisions and priorities determined?
- • Question: Which metrics get used to evaluate success?
- • Question: What's on your model-building checklist?
- • Question: How do the more sophisticated practices differentiate?
- • Question: What company changes are required for AI adoption?
- • Outro
Recently, Ben Lorica and I completed a study for O'Reilly Media about the state of AI adoption in enterprise. The study attempted to identify differences between firms with significant experience deploying machine learning in production, versus firms which were just beginning. Had there been impact on culture and organization, or differences in job titles used? We received more than 11,000 respondents worldwide, and noticed some unexpected results. For example, 40% indicated that their organizations check for ML model fairness and bias. Also, the sophisticated teams tend to follow different process than one might expect for 'Agile' product development teams in more traditional software engineering. Other recent analysis shows how (in light of GDPR) open source development is gaining momentum in highly regulated markets -- which again was unexpected. This talk explores those results and more, to examine how AI and other current factors are changing engineering process. If the default is no longer 'Agile' then what comes next?