AI Is Only as Smart as the Data You Give It
Artificial intelligence can analyze large amounts of information, identify patterns and generate recommendations in seconds. However, speed and confidence do not guarantee accuracy. AI is only as reliable as the data, assumptions and instructions shaping its output.
Our CEO, Katrina Starnes, connects this idea to the Johari Window, a model developed by psychologists Joseph Luft and Harrington Ingham in 1955. Its four panes represent what is known and unknown to ourselves and others, helping explain self-awareness, blind spots and communication. (Source: https://www.communicationtheory.org/the-johari-window-model/)
Katrina first encountered the model in college and was fascinated by the idea that every person views the world through a unique window. Relationships, emotions, culture, trauma and life experiences can expand, shrink or cloud that window.
The same principle applies to research.
Every survey participant, interviewee and focus group member responds through their own window. Researchers and organizations also bring assumptions into the process. Internal teams naturally absorb their organization’s language, priorities and culture. That context is valuable, but it can also make certain issues harder to see.
This is not due to a lack of skill. It is the invisible influence of being “inside.” External research can widen the window by challenging assumptions and uncovering perspectives that might otherwise remain hidden.
AI Has a Window, Too
AI tools do not interpret information neutrally. Their answers are influenced by their training data, system design, prompts and decisions made by their creators. Incomplete, outdated or unrepresentative information can therefore produce incomplete or biased conclusions.
Those conclusions may still sound polished and authoritative. OpenAI explains that language models can confidently produce incorrect answers, commonly called hallucinations, and recommends verifying important information. (Source: https://openai.com/index/why-language-models-hallucinate/)
The risk is already affecting organizations. McKinsey’s 2025 global survey found that 51% of respondents from organizations using AI had experienced at least one negative consequence. Nearly one-third of all respondents reported consequences caused by AI inaccuracy. (Source: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai)
Bias can also have serious real-world effects. A National Institute of Standards and Technology evaluation found that, for some facial-recognition algorithms with one-to-one verification tests, false-positive rates for Asian and African American faces were 10 to 100 times higher than for Caucasian faces. (Source: https://www.nist.gov/news-events/news/2019/12/nist-study-evaluates-effects-race-age-sex-face-recognition-software)
NIST’s AI Risk Management Framework emphasizes that trustworthy AI should be valid, reliable, transparent and fair, with harmful bias identified and managed. (Source: https://www.nist.gov/itl/ai-risk-management-framework)
Use AI to Support Thinking, Not Replace It
AI can help researchers analyze feedback, identify themes, explore alternative interpretations and challenge groupthink. However, it should not replace human judgment, context or properly designed research.
Before acting on an AI-generated insight, organizations should ask:
- Is the original research reliable?
- Who is represented in the data?
- Whose perspective might be missing?
- Can the conclusion be independently verified?
- What happens if the AI is wrong?
As Katrina explains, people, organizations, researchers and AI systems all interact with the world through different windows. The goal is not to assume any window is perfectly clear. It is to identify blind spots, include more perspectives and use strong research to see the larger picture.
AI may help us see farther, but thoughtful research helps ensure we are looking in the right direction.
