As many of us often do, I did a Google search for something the other day when I noticed a little bubble alerting me to the fact that AI was used to craft the search result. Quite coincidentally, a colleague had shared an article from MIT Technology Review only a day earlier, so my mind was spinning with concerns over the reliability of AI language model outputs.
This week Google announced LearnLM, its new family of models fine-tuned for learning. One goal is to use LearnLM to enable generative AI experiences via Google Classroom. “These features will help teachers discover new ideas and unique activities, find engaging materials, and differentiate their lessons and content to meet each of their students where they are,” wrote Ben Gomes, SVP, Learning & Education, Google.
However, according to a recent Pew Research Center Survey, teachers themselves are uncertain whether AI tools will benefit student learning, with 35% of high school teachers surveyed saying the tools will do more harm than good.
I had also watched a 2023 TED talk with Sal Khan advocating for AI tutors as a way to save (not destroy) education and I started thinking about the ways that a malicious actor could intentionally taint language models and use AI tutors to proselytize youth.
Malicious actors aside, “There are some people in the computer science community who say that hallucinations are an intrinsic part of generative AI that can’t ever be fixed, and that we can never fully trust these systems,” according to MIT Technology Review.
Fortunately, there is a lot of momentum happening from industry luminaries, Dr. Rumman Chowdhury among them, who are “focused on building innovations to stop algorithmic harm.” Chowdhury announced “the first of ten algorithmic bias bounty programs, dedicated to building the sociotechnical community of algorithmic assessors, critical thinkers and problem solvers.”
Now let’s take a look at what else made industry headlines this week.
May. 17: “Norwegian National Cyber Security Centre recommends replacing SSLVPN/WebVPN solutions due to the repeated exploitation of vulnerabilities in edge network devices,” Bleeping Computer reported.
May. 16: Cybercriminals are abusing Microsoft’s Quick Assist application using social engineering attacks that allow them to infect victims with Black Basta Ransomware.
May. 16: The social security numbers of 53,000 Nissan employees were exposed after a data breach.
May. 15: Scammers are blackmailing and stealing from companies through phishing emails mimicking DocuSign.
May. 15: Santander Bank announced they had a data breach impacting some of its customers and employees in Spain, Chile, and Uruguay.
May. 14: CISA warned organizations to be prepared to respond and remain resilient as cyber threat actors continue to move into US critical infrastructure networks.
May. 14: The Hacker News reported, “Cybersecurity researchers have uncovered an ongoing social engineering campaign that bombards enterprises with spam emails.”
May. 13: A defendant residing in Tracy, California, has been convicted of unleashing computer attack on systems controlling Tracy’s water treatment facility.