Can AI discriminate if it can’t justify itself? - FT中文网
登录×
电子邮件/用户名
密码
记住我
请输入邮箱和密码进行绑定操作:
请输入手机号码,通过短信验证(目前仅支持中国大陆地区的手机号):
请您阅读我们的用户注册协议隐私权保护政策,点击下方按钮即视为您接受。
FT商学院

Can AI discriminate if it can’t justify itself?

Elon Musk’s lawsuit against Colorado raises a deeper philosophical question about artificial intelligence and democracy
00:00

{"text":[[{"start":4.95,"text":"When xAI sued the state of Colorado over a law against “algorithmic discrimination” this month, it marked a political stand-off brought on by a rapidly changing technological reality. But more profoundly, the suit raises questions about what it is to be a democracy, and what it is to be human."}],[{"start":22.55,"text":"The law targets AI systems used in the provision of certain services, including education, housing, health and finance, some of them from the government. In passing it, Colorado lawmakers have defied much of the AI industry as well as the White House’s push for a single federal set of AI rules. "}],[{"start":40,"text":"Now Elon Musk’s xAI, which developed the large language model Grok, wants the law struck down because, it alleges, it infringes on the company’s free speech rights and forces it to change Grok to incorporate what it calls the controversial viewpoints of legislators. "}],[{"start":56.45,"text":"The company clearly protests too much. The law requires transparency over risks of bias and a duty of care to avoid unlawful discrimination. At most, Grok cannot be used for allocation decisions in the sectors designated by Colorado’s government. "}],[{"start":72.2,"text":"But beyond the narrow legal questions, there is a much more interesting and deeper matter at stake here. What the Colorado law does is to sweep AI-powered decision-making inside a fundamental demand for justification. The basic ground for doing this is that “my AI said so” is not enough to justify allocation decisions of certain goods and services where there is reason to fear unjust discrimination."}],[{"start":98.45,"text":"Why not? One reason is democratic. Giving legitimate reasons for how we are treated is at the core of the rule of law: the exercise of power must follow publicly known rules that apply to all. This is doubly true for electoral democracies where not just the structure of the law, but those writing and wielding it, should be accountable to citizens for the substantive choices they make."}],[{"start":122.15,"text":"The very term “accountable” is revealing. In democracies ruled by law, those who wield power must account for what they do — in other words, give adequate justifications and reasons — and be held to account by those subject to their decisions."}],[{"start":136.1,"text":"This role of justification and reason-giving was central to the political theories of the late German philosopher Jürgen Habermas, as well as the American thinker John Rawls. But justification matters in interpersonal relations too. "}],[{"start":152.15,"text":"When we want to convince someone to agree with us or persuade them to do something we want, we are called on to justify our demand. Forcing agreement or obedience without justification is abusive, a failure of basic respect. The fundamental human intuition behind Habermas’s and Rawls’s theories is that as citizens and simply as people, we owe it to each other to give reasons. "}],[{"start":176.85,"text":"Unfair discrimination itself can be understood in this light: it happens when some groups get less of something valuable than others, for no justifiable reason. So at one level, Colorado is simply demanding that AI systems must provide whatever reasons are satisfactory to justify differential treatment as legal — as in all other cases."}],[{"start":198.54999999999998,"text":"The complaint shows that xAI is particularly incensed at how Colorado’s law excludes biases designed to “increase diversity or redress historical discrimination” from the definition of algorithmic discrimination. This is indeed an odd choice by the legislature, which could have simply deferred to existing anti-discrimination laws on the question of whether such positive discrimination is permissible (for this, too, justifications can and should be given). "}],[{"start":226.7,"text":"Ironically enough, xAI’s brief in itself constitutes an earnest effort (albeit an unsuccessful one) at justification. The company — or at least its lawyers — grasp the need to offer reasons that others might deem legitimate. The more difficult question is whether LLMs such as Grok can themselves fully justify the answers they give and decisions they make. "}],[{"start":249.14999999999998,"text":"Suppose, for example, a Colorado landlord or triage clinician uses Grok to decide who gets the house or the hospital bed — but makes sure to require the model back up its input with reasons. What then? If Grok can adequately justify decisions as not violating discrimination law, and its developers can document this to users, that should fulfil the new law’s demands. The deeper human requirement for justification, however, requires much more."}],[{"start":277,"text":"Even if LLMs can provide a trace of their “reasoning”, that is not enough. When we admit that we owe each other justifications, we recognise each other as equals, with all the moral commitment to reciprocity that this entails. That is a commitment which — so far — only humans can make."}],[{"start":301.65,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1777185563_3130.mp3"}

版权声明:本文版权归FT中文网所有,未经允许任何单位或个人不得转载,复制或以任何其他方式使用本文全部或部分,侵权必究。

酶研究显示量子计算向药物发现迈进一步

科学家已利用这项技术模拟蛋白质分子的行为

欢迎来到“大蛰伏”时代

为什么没有更多人辞职?

“迷因股之王”大胆收购eBay能否成功?

瑞安•科恩正试图促成一笔560亿美元的交易,将视频游戏零售商“游戏驿站”与在线市场eBay合并。

为什么施罗德家族选择出售

在家族掌门人去世与美国巨头基金崛起之后,英国最大的独立资产管理公司被出售。

公司威胁涨价,消费者将面临更多痛苦

高管警告称,若能源冲击持续,企业将面临更大压力,把成本转嫁给客户。

中国收紧对生产商竞争的监管后,太阳能电池板价格上涨

在一场令头部厂商亏损惨重的价格战之后,价格反弹或将宣告“电池价格不断走低”时代的终结。
设置字号×
最小
较小
默认
较大
最大
分享×