And we’ll wrap up the week with some of those Big Ideas.
Let’s start with a piece in the Atlantic called “The Flaw That Could Ruin Generative AI.” A technical problem called “memorization” poses a significant threat to generative AI companies. Large language models can reproduce copyrighted texts, undermining the fair-use argument. Lawsuits filed by Universal Music Group and The New York Times highlight this issue and its potential impact on the generative AI industry.
And speaking of Generative AI, Medical Economics reports on the ability of chatbots to replace doctors. A recent American Journal of Preventive Medicine study evaluated the accuracy of AI models, ChatGPT-4 and Bard, in providing preventive medicine and primary care recommendations. The findings showed that ChatGPT-4 had 28.6% accurate responses, 42.8% accurate with missing information, and 28.6% inaccurate responses. Bard, however, demonstrated higher accuracy rates with 53.6% accurate responses, 28.6% accurate with missing information, and 17.8% inaccurate responses. Both models struggled with immunization-related questions, and ChatGPT-4’s outdated recommendations highlighted the need for continuous updates in AI systems.
Or let’s consider the dual role of SMB Brand Spoofing from Dark Reading. While AI makes it easier for adversaries to impersonate brands and carry out spoofing attacks, it also enables organizations to detect and block such attacks. Small to midsize businesses (SMBs) are particularly vulnerable to brand spoofing, and AI-powered security tools can help them fight back. SMBs face numerous cyberattacks, with brand spoofing being a pernicious threat. AI-generated fake content makes it easier for hackers to impersonate smaller brands. However, security architects are using AI to develop tools that can detect and block impersonation attacks, providing SMBs with better defense capabilities. In addition to AI, implementing solutions like DMARC and maintaining open communication with customers and vendors can also help prevent brand spoofing.
A harrowing look into Deepfake Nudes in Schools in the New York Times. I’ll quote an early paragraph. “In October, some 10th-grade girls at Westfield High School — including Ms. Mani’s 14-year-old daughter, Francesca — alerted administrators that boys in their class had used artificial intelligence software to fabricate sexually explicit images of them and were circulating the faked pictures. Five months later, the Manis and other families say, the district has done little to publicly address the doctored images or update school policies to hinder exploitative A.I. use.” That’s not the only example, and technologists should understand this issue.
Why do we care?
That New York Times article is sensitive enough that the AI editors I work with refused to analyze it. That tells you something. Understanding these challenges is the first key step in delivering those high-value services to customers. Like all technologies, there are balances to strike, and your role is knowing which tool for which job.