(点击右边三个点，可调整速度，电脑上可下载)

听力参考原文 ↓↓↓

[00:00.04]European Union lawmakers

[00:02.20]gave final approval to the 27-nation group's

[00:06.32]artificial intelligence law Wednesday.

[00:09.32]The rules are expected

[00:11.60]to take effect later this year.

[00:15.36]Lawmakers in the European Parliament

[00:18.20]voted in favor of the Artificial Intelligence Act,

[00:22.00]five years after regulations were first proposed.

[00:26.48]Major technology companies

[00:28.72]have generally supported the idea.

[00:31.28]But they want to make sure new AI requirements

[00:35.16]work in their favor.

[00:37.32]OpenAI chief Sam Altman suggested

[00:40.92]the maker of ChatGPT might pull out of Europe

[00:44.40]if it cannot comply with the AI Act.

[00:47.56]He later said his company had no plans to leave.

[00:51.92]Here are some details about Europe's new AI rules:

[00:57.40]Like many EU regulations, the AI Act

[01:00.56]started as consumer safety legislation.

[01:03.76]The EU took a "risk-based approach"

[01:06.48]to products or services

[01:08.32]that use artificial intelligence (AI).

[01:13.08]If an AI application is risky,

[01:15.84]then more rules cover it.

[01:18.16]Most AI systems are expected to be low risk,

[01:21.80]like content recommendation systems

[01:24.48]or filters that block spam, or unwanted email.

[01:28.56]Companies can choose to follow

[01:31.24]voluntary requirements and codes of conduct.

[01:35.88]High-risk uses of AI include

[01:38.48]tools used in medical devices

[01:40.96]or important infrastructure

[01:43.00]like water or electrical networks.

[01:45.68]Those face additional requirements

[01:48.56]like using what the legislation calls high-quality data

[01:52.60]and providing clear information to users.

[01:56.88]Some AI uses are banned

[01:59.08]because they are considered

[02:00.68]to present an unacceptable risk.

[02:04.08]Those include things like social scoring systems

[02:07.56]that are meant to govern how people behave.

[02:10.16]Some sorts of predictive policing

[02:13.44]and emotion recognition systems

[02:16.00]also are reportedly banned

[02:18.20]in schools and workplaces.

[02:21.40]Other banned uses include ones

[02:23.92]that police use to scan faces in public places

[02:27.60]using AI-powered remote

[02:29.84]"biometric identification" systems.

[02:32.68]There is an exception for use

[02:34.96]in serious crimes like kidnapping or terrorism.

[02:40.28]The law's early versions centered

[02:42.68]on AI systems that carry out limited tasks,

[02:46.28]like reviewing employment information

[02:49.36]and job applications.

[02:51.48]But general AI models, like OpenAI's ChatGPT,

[02:56.96]forced EU officials to add rules for generative AI models.

[03:02.00]AI chatbot systems

[03:04.12]that can produce lifelike responses,

[03:06.76]images and more

[03:08.88]are examples of generative AI models.

[03:13.16]Developers of general purpose AI models

[03:16.48]will have to provide detailed descriptions

[03:19.44]of the writings, pictures, video

[03:22.52]and other data on the internet

[03:24.52]that was used to train the systems.

[03:27.20]They must also follow EU copyright law.

[03:31.96]AI-generated pictures, video or audio

[03:36.12]of existing people, places or events

[03:39.24]must be labeled as artificially produced.

[03:42.44]These sorts of media are known as "deepfakes"

[03:46.76]because they appear to show real people doing

[03:49.76]or saying things that are not real.

[03:54.24]There are reportedly extra rules

[03:56.72]for the biggest and most powerful AI models

[04:00.08]that carry "systemic risks."

[04:02.68]Those include OpenAI's GPT4

[04:06.36]and Google's Gemini.

[04:10.08]The EU first suggested AI regulations in 2019.

[04:15.24]Europe was quick to propose rules

[04:18.12]for the new and developing industry.

[04:22.32]In the U.S., President Joe Biden

[04:25.00]signed an executive order on AI in October.

[04:28.72]The U.S. Congress is likely to propose legislation.

[04:33.72]Lawmakers in at least seven U.S. states

[04:37.24]are working on their own AI legislation.

[04:40.92]And international agreements are possible too.

[04:44.36]Chinese President Xi Jinping

[04:48.20]has proposed his Global AI Governance Initiative

[04:51.80]for fair and safe use of AI.

[04:55.16]Other major countries, including Brazil and Japan,

[04:59.76]are developing rules, as well as the United Nations

[05:03.76]and Group of Seven industrialized nations.

[05:09.12]The AI Act is expected to officially

[05:12.08]become law by May or June,

[05:14.52]after approval from EU member countries.

[05:17.48]Rules will start taking effect slowly.

[05:21.04]Countries will be required

[05:23.48]to ban unapproved AI systems

[05:26.08]six months after the law takes effect.

[05:30.16]Rules for general purpose AI systems like chatbots

[05:34.32]will start going into effect in one year.

[05:37.48]By the middle of 2026,

[05:40.28]the complete set of regulations,

[05:42.44]including requirements for high-risk systems,

[05:45.36]will be in effect.

[05:47.96]Each EU country

[05:49.44]will set up their own AI enforcement agency.

[05:52.80]Citizens can make a complaint if they think

[05:56.44]they have been the victim of a violation of the rules.

[06:00.00]And the EU will create an AI Office

[06:04.20]that will oversee the law

[06:06.12]for general purpose AI systems.

[06:09.68]Violations of the AI Act

[06:11.84]could be punished with a fine of up to $38 million,

[06:16.32]or seven percent of a company's worldwide revenue.

[06:20.44]I'm Dan Novak.

_____________________

Words in This Story

regulation — n. a rule or a group of rules that control how an industry can carry out business

comply — v. to do what is requested to be done or what is required by rules or law

consumer — n. a person or group that buys goods and services that are not for industrial purposes

filter — n. something that permits only what is wanted through while blocking unwanted things

codes of conduct –n. a set of rules that a group of people or businesses agrees to usually voluntarily

infrastructure — n. the structures and systems that are needed for modern life like roads, electricity lines, dams and many other things

scan –v. to use a camera to take a picture of a group of people to find a certain person

biometric –adj. related to taking measurements of the human body to confirm someone's identity

artificial — adj. made by people, not happening naturally; something that is not real

complaint –n. an official statement of dissatisfaction that is presented to a public official with the expectation that the problem will be dealt with

revenue — n. money collected by a business through sales, investment and other operations

本文来自公众微信号：练英语听力