和谐英语

人工智能带来公平(下)

2022-12-14来源:和谐英语

Ms Lobel’s call to use more, not less, personal information challenges data-privacy orthodoxy.

洛贝尔呼吁更多地使用个人信息,这挑战了数据隐私的正统观念。

But she insists that “tracking differences is key to detecting disparities.”

但她坚持认为,“追踪差异是发现不公的关键。

She advocates a careful loosening of intellectual-property rules to provide more transparency over algorithmic decisions.

她主张谨慎地放松知识产权规则,为算法决策提供更多的透明度。

And she floats the idea of a sort of affirmative action in AI to support disadvantaged groups.

她还提出了在人工智能领域采取平权行动来支持弱势群体的想法。

For instance, an algorithm that serves adverts for highly paid jobs to men—because they mostly clinched such posts in the past—can be programmed to show them equally to women.

例如,可以使用编程,将为男性提供高薪工作广告的算法,改为同样向女性显示这些职位的算法-因为在过去大多都是男性获得这样的职位。

As Ms Lobel says, AI need not merely reproduce or entrench old biases.

正如洛贝尔所说,人工智能不需要仅仅重现或巩固已有的偏见。

It can help expose them.

它可以帮助揭露这些偏见。

And it is easier to fix an algorithm than it is to change people’s minds.

而且,修复一个算法比改变人们的想法容易。

The problems with algorithmic formulae are tackled in depth in “Escape from Model Land” by Erica Thompson of the London School of Economics.

伦敦政治经济学院的艾瑞卡·汤普森在《逃出模型岛》中深入探讨了算法公式的问题。

These statistical models are the backbone of big data and AI: if data is the input, algorithms are the tool and models are the product.

这些统计模型是大数据和人工智能的支柱:如果数据是输入,那么算法就是工具,模型就是产品。

They are everywhere, from e-commerce tips to economic and climate-change forecasts.

从电子商务小贴士到经济和气候变化预测,它们无处不在。

Yet rather like the full-scale map of an empire imagined by the writer Jorge Luis Borges, a perfect model of the teeming world will always be beyond reach.

然而,就像作家豪尔赫·路易斯·博尔赫斯想象的帝国全尺寸地图一样,这个人口稠密世界的完美模型永远遥不可及。

The task is to ensure that the abstractions correspond to reality as far as is humanly possible.

我们的任务是确保这些抽象概念尽可能与现实相对应。

“All models are wrong,” runs a venerable saying. “Some are useful.”

“所有模型都是错的,”一句睿智的谚语如是说。“但其中有些是有用的。”

Ms Thompson focuses on a challenge she calls the Hawkmoth Effect.

汤普森关注的是一种她称之为“鹰蛾效应”的挑战。

In the better known Butterfly Effect, a serviceable model becomes less reliable over time because of the complexity of what it is simulating, or because of inaccuracies in the original data.

在更广为人知的“蝴蝶效应”中,可用的模型会随着时间的推移变得不那么可靠,原因或是它所模拟的东西很复杂,或是原始数据不准确。

In the case of climate change, say, this might lead to a prediction for rising temperatures being out by a fraction of a degree.

以气候变化为例,这可能会导致气温上升的预测出现零点几度的偏差。

In the Hawkmoth Effect, by contrast, the model itself is flawed; it might fail to take full account of the interplay between humidity, wind and temperature.

相比之下,在鹰蛾效应中,模型本身就是有缺陷的;它可能没有充分考虑湿度、风和温度之间的相互作用。

This sort of mistake can be much more misleading, and much harder to rectify.

这类错误可能更具误导性,也更难纠正。

The author calls on data geeks to improve their solutions to real-world issues, not merely refine their formulae—in other words, to escape from model land.

该作者呼吁数据极客改进他们对现实世界问题的解决方案,而不仅仅是完善他们的公式--换句话说,要逃出模型岛。

“We do not need to have the best possible answer,” she writes, “only a reasonable one.”

“我们不需要最好的答案,”她写道,“只需要一个合理的答案。”

Before there is a statistical model, she notes, there is a mental version.

她指出,在统计模型之前,还有一个心理版本。

Data scientists need self-awareness and empathy as well as mathematical skill.

数据科学家不仅需要数学技能,还需要有自我意识和同理心。

Both these books exhibit a healthy realism about data, algorithms and their limitations.

这两本书都展示了关于数据、算法及其局限性的健康现实主义。

Both recognise that making progress involves accepting constraints, whether in law or coding.

也都认识到,要想取得进展,就要接受法律或编码方面的约束。

Ms Lobel calls on AI practitioners to remedy the technology’s problems; Ms Thompson asks data scientists to be conscious of the choices and values in a model’s design.

洛贝尔呼吁人工智能从业者纠正这项技术的问题;汤普森要求数据科学家意识到模型设计中的选择和价值。

Their reflections offer the basis for a constructive agenda.

二位的意见为建设性议程提供了基础。

As Ms Lobel puts it: “It’s always better to light a candle than to curse the darkness.”

正如洛贝尔所说:“与其诅咒黑暗,不如点燃蜡烛。”