Can big genAI providers FactCheck each other?
by Stephan Steiner
Generative AI holds tremendous potential, yet its pitfalls become apparent when faced with instances like Google’s recent introduction of Gemini (formerly Bard). Users reported significant issues with the accuracy of AI-generated images, revealing a concerning lack of precision in depicting historical scenarios. For instance, when tasked with creating an image of 1800s United States Senate members, Gemini produced a collection featuring individuals from diverse backgrounds, including Black, Asian, and Native American women. This was despite the historical context, as the first female Senator, Rebecca Felton, Georgia, did not assume office until 1922 [1]. Similarly, when prompted to visualize Nazi-era soldiers in 1943, the AI again presented a diverse group, including Asian and Black women. More details on these incidents can be found on TheVerge [2].
In response to user complaints, Google’s Senior Vice President of Knowledge and Information, Prabhakar Raghavan, acknowledged the shortcomings, stating, “Some of the images generated are inaccurate or even offensive. We’re grateful for users’ feedback and are sorry the feature didn’t work well” [3].
The incident has sparked discussions on bias within AI analytics models, raising questions about the control and trustworthiness of AI systems. We won’t get into political debate here, it underscores the power of companies, with substantial resources, to influence opinions and shape perspectives. This leads to crucial inquiries: Who governs AI models, and how can their trustworthiness be ensured?
While the aforementioned cases are at best “inaccurate”, they prompt reflection on the potential for more malicious attempts to manipulate public information in the future. Detecting and combating falsely generated AI content becomes a critical concern, and determining liability for any resulting harm remains an open question.
Considering that major tech providers likely employ unique algorithms, there arises the possibility of establishing a “trust score” by comparing the outcomes of different companies, I.e. use multiple models to FactCheck each other?
Trust in generative AI’s capabilities holds significant value for humanity, but it necessitates a reliable and transparent source. The current incident serves as a reminder of the importance of addressing these challenges to harness the full potential of generative AI responsibly.
What are your thoughts on this matter?
#StephanSteiner #AITrust #AIfactcheck
Sources:
[1] https://www.senate.gov/artandhistory/senate-stories/rebecca-felton-and-one-hundred-years-of-women-senators.htm
[2] https://www.theverge.com/2024/2/21/24079371/google-ai-gemini-generative-inaccurate-historical
[3] https://blog.google/products/gemini/gemini-image-generation-issue/