The Biden Administration is planning on developing a national AI strategy to address both the benefits of AI, as well as associated risks. The strategy appears in several interconnected documents. The National Science Foundation will also spend $140 million to promote research and development in AI, according to CNN.
Looking into the topic is Mark Shainman, Senior Director of Data Governance Products at Securiti. He explains to Digital Journal why the Biden approach needs to be detailed and comprehensive.
Shainman sets out what needs to be covered: “When it comes to developing a national strategy, it is crucial to extend policies similar to those created for security and privacy regarding personal data. These should encompass the handling of personal and sensitive data in AI models.”
It is also important to ensure evidence is drawn from different sectors. Accoprding to Shainman: “Organizations, as well as government agencies, are now looking to leverage AI to improve and streamline numerous business functions, gain business insight, and identify risks. These processes span numerous functions; however, as exciting as the opportunities AI brings are, there are risks to be mindful of.”
Expanding on the risk concept, Shainman provides the following scenario: “Once AI models receive data for training, there is no way for it to “forget” that data, if improper or sensitive information is fed to the model, there is a continued risk of exposure. There is one aspect not fully addressed, the critical need for robust data governance and controls around sensitive data that could be potentially exposed to AI.”
There are other factors to consider, especially the nature of bias and decision making. Here Shainman draws out: “As the Biden administration moves on responsible AI usage, another harm to highlight is discriminatory decisions. The decision an AI system would make is only based on the data available and on what the model has been trained on.”
Taking this further with a comparison, Shainman puts forward: “Just like a human, it will only use the data it has. The difference is understandable, AI models cannot be reasoned with or have feelings or empathy humans do. In situations where there needs to be more than just raw data to make final decisions, there needs to be a mechanism in place where a circumstance can be evaluated by a person, who can bring more than just reason to the decision-making process.”
There is also a political dimension that cannot be ignored. Shainman turns to the ability of AI to affect the democratic outcome: “When it comes to AI-generated content that can disrupt elections, the only real difference between a person creating fake content and an AI tool doing it is the speed and scale.”
There are policy-led solutions that can be enacted, however, and the government can look to the private sector for ideas. Shainman proposes: “The key to mitigating this challenge is to do what a lot of top companies, like Google, are doing now: leveraging solutions that can identify AI-generated content. Once the government can do this, it can make the public aware that specific content is election spam content generated by AI”.
