Hi Friend, what do we really know about AI?

The AI world is abuzz with news of GPT-4, which generates credible-sounding, but sometimes unreliable, information incredibly fast. Meanwhile, GPT-4's creator OpenAI hasn't provided transparency into the model's size, architecture, training methods, or potential for bias. Transparency in AI development and deployment is crucial if we want real accountability.

A lack of transparency in AI could have serious implications for us all, especially as AI tools are rapidly being deployed across multiple sectors, including government, education, health, media and business.

We must demand transparency and accountability to mitigate any negative impacts of these powerful tools. In this context, our recent study, “AI Transparency in Practice - Builders on what works - and what doesn’t,” conducted by Mozilla’s Insights team, which involved several dozen developers, software engineers, product designers and machine learning engineers from seven countries reveals:

• Ethical and legal compliance are not a top priority for AI builders in designing and deploying AI systems. Over 90% of respondents ranked ethical guidelines 11th on a 12-point scale despite the upcoming EU AI Act and Digital Services Act.
• Builders prioritize accuracy and goal fulfillment over legal and ethical compliance.
• There is little incentive to communicate unintended consequences or product limitations to end-users or impacted people during AI development, which can raise ethical concerns.
• Lack of transparency often leaves users struggling to understand how algorithmic decisions affect their lives.

Besides highlighting the problems, our report also provides practical solutions to promote transparency and accountability in AI development. We suggest regular audits, algorithmic impact assessments, and the use of well-understood, interpretable-by-design models in the absence of adequate explainability tools.

At Mozilla, we believe that AI transparency is critical to a fair and just society. We encourage builders of AI systems to prioritize ethical and legal compliance when designing and deploying AI systems.

Join us in fighting for more meaningful transparency in AI.

Mozilla Insights