In one sense of the word, we all use "AI" in some way, whether we know it or not. We use it every time we use any form of online search engine, whether Google or Bing or Yahoo or something else. The user needs to be careful of using whatever he or she turns up in a search uncritically, but using this form of "search and find" software is not what has landed some unfortunate sorts in trouble.
The Author's Stance on Artificial Intelligence
The author, Bob Arrington, clarifies that while he's a skeptic of uncritically believing AI has human intelligence or personality, he is not against using AI applications. He emphasizes the growing necessity of understanding how to use them, especially distinguishing between basic search engine AI and more complex "generative AI" for content creation.
Generative AI and Its Associated Risks
Generative AI applications use large language models to rapidly search vast amounts of data, make statistical correlations, and produce documents, spreadsheets, charts, or images based on user specifications. While highly useful, this technology carries inherent risks that can lead to ethical or legal issues for unwary users, prompting the author to share insights from his courses on the subject.
Overview of Generative AI Providers and Core Risks
Prominent generative AI providers include OpenAI (ChatGPT) and Anthropic (Claude), with some applications tailored for specific professional fields such as law, accounting, and medicine. The author highlights three major risks associated with using generative AI: security vulnerabilities, inherent biases in its output, and "hallucination"—the fabrication of information.
Understanding and Mitigating Security Risks
Basic generative AI applications are inherently insecure as they exchange data in cyberspace, making them unsuitable for private or confidential information. The common misconception that paid subscriptions offer enhanced security is addressed; they usually provide more features but not data protection. True security typically requires "executive" or "enterprise" closed-loop versions, though caution is always advised as no system is unhackable.
Combating Bias and Hallucination in AI Output
Generative AI can frequently exhibit bias because its underlying large language models are trained on human-inputted data, which may contain biases. The solution is for users to specifically request unbiased or balanced analyses. "Hallucination," or the AI fabricating information, is another critical risk, for which the only reliable defense is for users to personally verify every source citation and piece of information provided by the AI.
