As school resumes, data shows that the usage of ChatGPT has recovered, indicating that students mainly use it for cheating on homework. The decline in usage during the summer break supports this theory. However, this limited range of use cases raises concerns about the long-term potential of AI-powered chatbots like ChatGPT. OpenAI has provided a guide for teachers on using ChatGPT in the classroom, but it cannot reliably detect AI-generated content from human-generated content.
A recent study found that several AI-based chatbots, including OpenAI’s ChatGPT-4, Databricks’ Dolly 2.0, and Stability AI’s StableLM-Tuned-Alpha, exhibit significant gender bias. The study examined the responses of 13 chatbots to different prompts and ranked them based on bias in a professional context and storytelling. The results highlight the need for improvements in language models to mitigate biases—link in the story to the research.
Despite the rapid pace of changes brought by AI in the workplace, many leaders are not considering the potential for significant future transformations. Per a large qualitative study, researchers explored how this “present bias” hinders the adoption of new technologies to make the workplace more productive and inclusive. The researchers recommend that to prepare for the next decade of workplace transitions; leaders should leverage technological advances for inclusion, identify areas where jobs can be automated, use AI in hiring and promotion processes, address concerns about AI integration, and recognize the importance of human interaction and inclusivity.
It would be easy to discount that ChatGPT’s market is for students. I’m less sure – that shows a utility that was quickly adopted. Its capabilities can be directed towards more constructive educational purposes, such as assisting with research or helping understand complex topics. That’s valuable – sure, early, but valuable.
The recommendation to use AI in hiring and promotions has its own ethical implications, especially in the context of observed biases in AI models. Be thoughtful here. A measure twice, cut once approach is going to pay off.