Select Page
Blog - Aktuelle Umfrage von Applause demonstriert Bias

New Applause Survey Shows How Bias Can Infiltrate AI

A few weeks ago, the Applause team attended the North American AI & Big Data Expo in California. Having just announced our new AI training and testing solution, the event was the perfect time to illustrate how easily bias can permeate an AI algorithm, and how a large and diverse data set is the pathway to avoid this problem.

Our hypothesis was that a global – and more diverse – group of individuals would produce data that was less biased. To test this theory, we created a simple one-question survey: What is your favorite brand of coffee?

Two different data sources

We first surveyed our global community of vetted real-world testers, then asked the question to attendees of the AI & Big Data Expo to see how the results compared. We expected that responses at the tradeshow would be more homogeneous due to event-goers coming primarily from the West Coast and working in technology roles. Our community, on the other hand, is universal and features people from all backgrounds, genders, races, work roles and locations. We hypothesized more diverse answers from them.

Though we expected more varied answers from the community, we were still surprised by the difference in responses between the community and the audience at the AI & Big Data Expo.

The West Coast bias was clear from the get-go in the show floor results.

West Coast bias in effect

The West Coast bias was immediately clear from the show floor results. Starbucks was far and away the favorite, accounting for 37% of total responses, followed by California-based chains Peet’s and Blue Bottle. Those three West Coast brands accounted for 65.5% of total responses, while only 17 brands received at least one response.

Comparatively, 87 different brands received at least one vote in the community survey. Starbucks was still a top choice, but it ranked second to Nestle and was followed closely by Lavazza – two companies headquartered outside of the United States. The top 10 companies in the community survey accounted for only 62.5% of responses.

Survey results: What is your favorite brand of coffee?

 

AI & Big Data Expo

 

Applause Community

The diversity of answers is immediately noticeable when you compare these two images, as the community provided a longer list of brands and more global brands. These results more accurately represent the coffee drinking habits of a worldwide customer base.

I broke down these results even further, and explained Applause’s own journey with AI, in this webinar linked below. Make sure to tune in to see the full results and learn why the human element is still essential for AI success.

Want to see more like this?
Published: December 5, 2019
Reading Time: 4 min

Beyond Traditional Testing: Advanced Methodologies for Evaluating Modern AI Systems

As AI systems continue to demonstrate ever more complex behaviors and autonomous capabilities, our evaluation methodologies must adapt to match these emergent properties if we are to safely govern these systems without hindering their potential.

Agents and Security: Walking the Line

Common security measures like captchas can prevent AI agents from completing their tasks. To enable agentic AI, organizations must rethink how they protect data.

How Agentic AI Changes Software Development and QA

Agentic AI introduces new ways to develop and test software. To safely and effectively make the most of this new technology, teams must adopt new ways of thinking.

Automation vs. Agentic AI: Key Differences

Explore the core differences between rule-based automation and agentic AI, and their roles in modern software QA.

Usability Testing for Agentic Interactions: Ensuring Intuitive AI-Powered Smart Device Assistants

See why early usability testing is a critical investment in building agentic AI systems that respect user autonomy and enhance collaboration.

Do Your IVR And Chatbot Experiences Empower Your Customers?

A recent webinar offers key points for organizations to consider as they evaluate the effectiveness of their customer-facing IVRs and chatbots.
No results found.