#RSAC: Time to Take Action on AI-Enabled Electoral Vote Influencing

Written by

In a talk at the RSA Conference in San Francisco, students and researchers from University of California, Berkeley presented a theoretical method on how voters could be influenced using technical and automated methods.

Talking about “How AI Inference Threats Might Influence the Outcome of 2020 Election,” the three presented their own research, which included aggregating data to show how misinformation can be spread. Karel Baloun, software architect and entrepreneur at UC Berkeley, said these types of attacks can be nefarious as “attacks on democracy” are often not seen and it can be denied that they took place.

Pointing at the 2016 US presidential election, Baloun said that the hacking of the Democrats’ emails by Russia and passing of them to WikiLeaks “set the narrative for the election” and there is proof that this effort was able to “suppress over 100,000 votes.” He said that there are four examples of elections that have been influenced in history:

  • The 2016 Ukraine Election
  • The 2016 UK Brexit vote on EU Membership
  • The 2019 Hong Kong Anti-Extradition Law Protests
  • The 2020 Taiwan Presidential Election

Ken Chang, cybersecurity researcher at University of California, Berkeley, said that when someone registers to vote, that information should be trusted to be held securely, as all information that is collected is “a critical piece of information.”

With voter registration data, Chang said that the potential of a data breach is obvious, so the conversation needs to be centered on how to protect information, and not on how a data broker can collect and distribute information without the person knowing.

Baloun said that with the experiment it was able to build user voter databases and aggregate this into social media data, advertising, and messaging to influence people. Citing the case of Cambridge Analytica, Baloun said that it was able to use Facebook data that was open, and personal information that is freely obtained and available in the form of credit scores and credit card data.

Saying that it is only a matter of time before AI can do the whole process, as currently Machine Learning is used on Big Data sets, and AI can generate texts and emails and write news, Baloun said that the “technology is well advanced.”

“If you suck the firehose you only get what you’re provided,” Baloun said, pointing out that it could be easy for an attacker to impersonate an influential friend or family member.

Looking at steps to take, Baloun encouraged taking more action when friends and family share such information, and think about what you consume. He also called for the Secretary of State with responsibility for voter records to mandate a disclosure requirement. He also called for a ban by the FFC on creating “personal profiles” pretending to be voters.

“Each one can make a big difference, as the system depends on easily available rich voter profiles, and targeting with messaging,” he said. “To protect democracy we need to make things more expensive and less effective and let humans intervene, as they don’t know it is happening.”

What’s hot on Infosecurity Magazine?