AI is transforming the business world, but with this rapid change comes risks as organizations struggle with responsible deployment of AI to minimize harm. Transparency is a key aspect of responsible AI, ensuring that decisions made by AI systems are fair, unbiased, and ethical. Many businesses are taking steps towards ensuring transparency, but there are cases where AI use has been opaque.
Examples of transparent AI done well include Adobe’s Firefly toolset, which provides information on the data used to train its models, ensuring users can trust the tool hasn’t infringed copyrights. Salesforce includes transparency as a key element for developing trustworthy AI, while Microsoft’s Python SDK for Azure Machine Learning includes model explainability to understand decisions made by AI. Cognizant recommends creating centers of excellence to centralize AI oversight and adopt best practices across organizations.
On the other hand, examples of transparent AI done badly include OpenAI, which has faced accusations of failing to disclose what data is used to train their models, leading to lawsuits from artists and writers. Image generators like Google’s Imagen and Midjourney have been criticized for depicting professionals inaccurately due to lack of transparency. In banking and insurance, lack of transparency in AI systems used for assessing risk and detecting fraud can lead to customers facing issues without understanding why they are being singled out.
Ensuring AI is deployed transparently is crucial for building trust with customers, eliminating problems caused by biased data, and complying with increasing regulations around AI. Legislation such as the upcoming EU AI Act requires transparency and explainability for AI systems in critical use cases, making it essential for businesses to prioritize transparency and accountability in AI systems. Overcoming the challenge of transparency in complex AI models is crucial for AI to realize its potential for positive change and value creation in society.
