【英文】美国电子隐私信息中心报告：生成式人工智能的潜在有害影响与未来之路（86页）

OpenAI’s decision to release ChatGPT, a chatbot built on the Large  Language Model GPT-3, last November thrust AI tools to the forefront of  public consciousness. In the last six months, new AI tools used to generate  text, images, video, and audio based on user prompts exploded in  popularity. Suddenly, phrases like Stable Diffusion, Hallucinations, and Value  Alignment were everywhere. Each day, new stories about the different  capabilities of generative AI—and their potential for harm—emerged without  any clear indication of what would come next or what impacts these tools  would have. While generative AI may be new, its harms are not. AI scholars have been  warning us of the problems that large AI models can cause for years. 

These old problems are exacerbated by the industry’s shift in goals from research  and transparency to profit, opacity, and concentration of power. The  widespread availability and hype of these tools has led to increased harm  both individually and on a massive scale. AI replicates racial, gender, and  disability discrimination, and these harms are weaved inextricably through  every issue highlighted in this report.OpenAI and other companies’ decisions to rapidly integrate generative AI  technology into consumer-facing products and services have undermined  longstanding efforts to make AI development transparent and accountable,  leaving many regulators scrambling to prepare for the repercussions. And it  is clear that generative AI systems can significantly amplify risks to both  individual privacy and to democracy and cybersecurity generally. In the  words of the OpenAI CEO, who indeed had the power not to accelerate the  release of this technology, “I’m especially concerned that these models  could be used for widespread misinformation.