Summary
Artificial intelligence (AI) has graduated from science fiction to commoditized widget form, readily able to snap into many processes of daily life. Hence, enterprises of all maturity levels are increasingly eager to explore AI’s roles in their innovation, or outright survival strategies. Concurrently, ethical and responsible development and execution of AI-based solutions will increasingly become critical for purposes of safety and fairness. Ensuring that AI proliferates along the right path will require the infusion of multi-faceted research activities along the entire AI lifecycle. We will discuss the challenges and opportunities regarding this topic in this presentation.
Key Insights
-
•
Despite AI hype, up to 90% of enterprise data science projects never deploy into production.
-
•
Humanized AI requires governance frameworks that include non-engineering roles like product researchers and AI ethicists.
-
•
Black-box AI decisions limit trust and adoption due to opaque, probabilistic model reasoning.
-
•
The Gender Shades study revealed up to a 33% error rate disparity in facial recognition accuracy by gender and skin tone.
-
•
ML ops extends DevOps principles to AI, ensuring governance, accountability, validation, and feedback loops in model development.
-
•
Visualization research is critical to exploring high-dimensional data and identifying intersectional biases within training sets.
-
•
Explainability must evolve from feature importance scores towards causal, natural language rationales understandable to non-technical stakeholders.
-
•
AI trust depends on the system’s ability to engage in human-like back-and-forth explanations that users can scrutinize.
-
•
No-code and low-code AI tools are lowering the barrier to AI adoption but require careful governance to avoid perpetuating biases.
-
•
Cross-disciplinary, multi-faceted research collaborations between industry and academia remain essential as AI ethics and policies mature.
Notable Quotes
"I observe that AI development is often uninformed and hurried, resulting in deployments that don’t operate well in the real world."
"Humanizing AI is not just about creating human-like interfaces but about governance frameworks to democratize safe, effective AI solutions."
"Almost 90% of data science projects do not make it into production—they die on the vine."
"Black box decision making means information goes in, something comes out, and we really have no clue why."
"The Gender Shades project spurred Microsoft and IBM to revisit and correct bias issues in their facial recognition products."
"ML ops borrowed from DevOps offers governance, accountability, validation, and stakeholder responsibilities for AI development."
"Visualization helps analyze multi-dimensional data so non-engineers can identify wrong or biased datasets to avoid."
"Explainability techniques aim to turn numerical AI results into causally understandable, human-readable rationales."
"AI trust comes when AI can provide convincing explanations and engage in back-and-forth like a human advisor."
"Humanizing AI is necessary for safe, enjoyable, and competitive AI solutions, especially as governmental ethics policies mature."
Or choose a question:
More Videos
"Don't be cool, be good—work hard at managing your teams because they need you to nail it."
Adam Cutler Karen Pascoe Ian Swinson Susan WorthmanDiscussion
June 8, 2016
"Middle managers are responsible for the how—process, coordination, and communication—and you don’t see the value of that until it’s missing."
Peter MerholzThe Trials and Tribulations of Directors of UX (Videoconference)
July 13, 2023
"If you haven’t designed who your teams are and who your players are, expecting people to comply with standards won’t work."
Lisa WelchmanCleaning Up Our Mess: Digital Governance for Designers
June 14, 2018
"Investing in sustainability today will yield dividends for future generations."
Vincent BrathwaiteOpener: Past, Present, and Future—Closing the Racial Divide in Design Teams
October 22, 2020
"Growth and learning is your long term change management plan."
Brenna FallonLearning Over Outcomes
October 24, 2019
"Treat identities as elastic, not as fixed personas, because people’s needs and roles are complex and changing."
Tricia WangSpatial Collapse: Designing for Emergent Culture
January 8, 2024
"Diverging and converging around the business model canvas helped us test and prototype delivery methods for the value propositions."
Edgar Anzaldua MorenoUsing Research to Determine Unique Value Proposition
March 11, 2021
"We want to build teams with diverse skill sets so we can create a full picture during the knowledge creation phase."
Designing Systems at Scale
November 7, 2018
"You’re never testing an idea purely; you’re always testing the implementation of that idea."
Erin WeigelGet Your Whole Team Testing to Design for Impact
July 24, 2024
