Generative AI Dominates 2022 as Powerful Public Models Spark Debate
Stanford's HAI reviews a transformative year where generative models like ChatGPT and DALL-E 2 go mainstream, sparking both excitement and serious ethical concerns.
Generative AI dominates the biggest headlines of 2022 as major companies release powerful public models. OpenAI launches ChatGPT and makes DALL-E 2 broadly available, while competitors like Google's Imagen, Midjourney, and Stability.AI's Stable Diffusion enter the market. These tools demonstrate impressive capabilities in generating both text and images from simple prompts, quickly attracting the attention of businesses and consumers alike.
Despite their popularity, these generative models reveal significant ethical issues and biases inherent in their internet-based training data. Researchers discover that image generators produce biased outputs, such as depicting firefighters exclusively as white men and janitors as people of color. Additionally, the technology prompts copyright concerns from artists who recognize their original work in generated images, and misuse leads to bans on platforms like Reddit.
Beyond generative tools, the AI community grapples with a highly publicized panic over sentient AI after a Google engineer claims the LaMDA language model possesses a soul. While experts widely dismiss these sentience claims, the situation highlights how easily advanced models trick people into believing they interact with a real intelligence. As the year closes, the industry anticipates the next major leap with rumors of GPT-4 circulating at the NeurIPS conference.