AI Ethics Ignored as Explicit Images Found in Training Data
A recent investigation reveals that popular AI image generators train on datasets containing suspected child sexual abuse material, highlighting a dangerous lack of ethical oversight in the industry.
The AI industry faces intense scrutiny this week after a Stanford watchdog group discovers thousands of suspected child sexual abuse images within LAION, a popular dataset used to train image generators like Stable Diffusion. LAION takes down its training data and pledges to remove the illegal material, but this alarming incident highlights a severe lack of oversight in how generative AI models are built.
As no-code AI tools proliferate, the barrier to entry for creating generative models drops significantly, which tempts companies to prioritize speed to market over ethical considerations. Properly combing through massive datasets for harmful content is a difficult and time-consuming process that many tech giants and startups simply bypass in order to appease shareholders and outpace competitors.
This rush to release AI products without adequate ethical guardrails causes real-world harm, as evidenced by Microsoft's chatbot comparing a journalist to Hitler, ChatGPT and Bard dispensing racist medical advice, and DALL-E displaying Anglocentric biases. Until the industry shifts its focus from rapid deployment to responsible development that includes marginalized stakeholders, AI ethics continues to fall by the wayside.