人工智能的恶意使用：预测，预防和规避

今天看到了一片不错的paper，和大家分享一下。主要讲AI大规模应用在健康，运输，文化，科技等领域带来积极改变的同时，也会越来越多的带来信息安全、政策安全方面的风险。因为当今电脑系统在根本上是不安全的，在一个本不安全的系统上开发至关重要的机器学习系统将会自然而然导致AI的不安全性。文章对AI安全风险进行了预测，并探讨了预防和规避风险的策略。下面是文章摘要。

In the coming decades, artificial intelligence (AI) and machine learning technologies are going to transform many aspects of our world. Much of this change will be positive; the potential for benefits in areas as diverse as health, transportation and urban planning, art, science, and cross-cultural understanding are enormous. We’ve already seen things go horribly wrong with simple machine learning systems; but increasingly sophisticated AI will usher in a world that is strange and different from the one we’re used to, and there are serious risks if this technology is used for the wrong ends.

Today EFF is co-releasing a report with a number of academic and civil society organizations1 on the risks from malicious uses of AI and the steps that should be taken to mitigate them in advance.

At EFF, one area of particular concern has been the potential interactions between computer insecurity and AI. At present, computers are inherently insecure, and this makes them a poor platform for deploying important, high-stakes machine learning systems. It’s also the case that AI might have implications for computer [in]security that we need to think about carefully in advance. The report looks closely at these questions, as well as the implications of AI for physical and political security. You can read the full document here.

原文地址
https://www.eff.org/files/2018/02/20/malicious_ai_report_final.pdf