AI and Your Business
Is the AI available to the public in the summer of 2024 poised to transform the world as we know it? Many believe it might, fueling fears that the dystopian future portrayed in James Cameron’s 1984 film Terminator could be just around the corner if we continue on our current trajectory.
Let me both reassure and caution you. Does today’s AI threaten our very existence? Not in the cinematic sense, at least not yet. However, like any tool, if misused, it can indeed cause harm to you and your business if you don’t treat it properly.
Consider the AI of today, particularly large language models (LLMs), as merely the fourth evolution of Internet search engines. These models are trained on a diverse array of content created by countless individuals. They aid in locating relevant information and presenting it in new ways. However, the accuracy and quality of this information can vary greatly. Some information is good, and some information is bad; some is accurate and some is flat out false nonsense that someone wrote as a joke and is now part of an AI hallucination.
Today’s AI models generate responses based on the “prompts” or queries they receive, drawing on the data they were trained on. They don’t create something from nothing; everything they produce is derived from pre-existing human-generated content, including the algorithms they employ and the humans that helped with the modeling. Early LLM material was often annotated and processed by people in developing countries without the same context as the original authors, so bias differences have been introduced as well.
As someone who went to school during the Internet rise of the ’90s, I recall using early search tools like W3Catalog, WebCrawler, and others. These were our gateways to discovering web content before the arrival of giants like Yahoo, Ask Jeeves, MSN, and Google. I learned to navigate these tools thanks to librarians, who also taught me the essentials of research and the importance of proper citation in anything I write.
Here’s a simpler way to put it: written knowledge differs from spoken words. What is said and passed on can be considered hearsay and is not admissible in court due to the lack of verifiable evidence. In contrast, written records that are shared are permanent and can be used as evidence. This distinction is crucial, especially considering the inaccuracies that often plague eyewitness accounts—a point humorously illustrated in the film My Cousin Vinny which also came out in the 90’s. Every kid played the telephone game growing up, and what is derived from AI could be looked at as a more modern version of that.
Therefore, treat AI-generated content with caution. While it may offer what appears to be original ideas and guidance, it cannot replace the reliability of a well-researched book or cited source of material on the web. The origins of the data LLMs are trained on are often untraceable, making proper end-user validation impossible, and you don’t know the bias introduced in the source material or by the person doing the annotating.
Just as you wouldn’t base your business decisions on advice from a stranger on the street, exercise a degree of skepticism toward AI. However, that doesn’t mean it can’t inspire original ideas or provide valuable guidance, and or help speed up some of your more mundane tasks. AI does produce great output when given a specific and targeted prompt, just like a search engine…
Recommendations
- Do Use AI to help you find how to do a function in Excel
- Do Use AI to help you write a SQL Query
- Do Learn how to write good prompts
- Do Use natural language in your prompts
- Don’t put Intellectual Property or Original Material into public Large Language Models
- Don’t expect AI to write a giant report for you from just a few basic prompts unless you want AI hallucinations.
- Ask AI to include its sources with a prompt like this: “Please write a detailed response on [specific topic], and include citations from reputable sources to support the information. Make sure to cite multiple sources wherever possible, and provide valid hyperlinks to these sources.”
- Validate the sources that AI references – read the source material for yourself, don’t just trust the fact a citation was included because they can be AI hallucinations
- Check for Copyright infringement before using AI output commercially