7 . Recently, as a journalist, I have published many stories about social media, privacy and artificial intelligence (AI), among other things. So when ChatGPT told me that my output may have influenced its responses to other people’s prompts (提示), I rushed to wipe my data from its memory. As I quickly discovered, however, there is no delete button. AI-powered chatbots never forget what they have learned, because they are trained on data sets including vast numbers of websites and online articles. As long as they exist on the Internet, they get to be remembered.
That means the likes of ChatGPT are possible to let out sensitive personal information, if it has appeared online, and that the companies behind these AIs will struggle to make good on “right-to-be-forgotten” regulations, which force organizations to remove personal data on request. It also means we are powerless to stop hackers (黑客) controlling AI outputs by planting misinformation or ill instructions in training data. All of these explain why many computer scientists are busy to teach AIs to forget. While they are finding that it is extremely difficult, “machine unlearning” solutions are beginning to appear. And the work could prove vital beyond addressing concerns over privacy and misinformation.
The new generation of AI-powered chatbots like ChatGPT, which produce text in response to our prompts, are underpinned by large language models. These are trained on mountains of data, most of which is scraped from the Internet. From this, they learn to spot statistical patterns, which means they can predict the likeliest next word in a sentence, producing fluent answers to our every question.
The trouble is that the way AI chatbots work means that when they learn something, it can’t be unlearned. This creates a significant problem when it comes to privacy, as Dr. Zhang made clear in recent research. He highlighted how difficult it will be for AI companies to obey the “right to be forgotten”, which the European Union declared a human right back in 2014.
1. Why can chatbots store what it has learned forever?
A.Lack of a delete button. | B.Manual code entry. |
C.Strong review ability. | D.The way they are trained. |
2. Why do scientists try to teach AI to forget?
A.Because AI has many security risks. | B.Because AI stores limited information. |
C.Because AI needs to be updated regularly. | D.Because AI imitates humans completely. |
3. According to Paragraph 3, what can we learn about chatbots?
A.Chatbots can answer 90% of the questions. |
B.Chatbots’ data comes from the designer’s software. |
C.Chatbots produce the answers based on the prompts. |
D.Chatbots can predict your problems in advance. |
4. What does the underlined word “underpinned” in Paragraph 3 mean?
A.Controlled. | B.Adjusted. | C.Boosted. | D.Supported. |