Even before the Web was spun, search engines were librarians. When you ask a question, they will find material that can help you answer the question. I put that into my stories as footnotes and later, thanks to the Web, they became links.
Generative AI, when applied to search, changes the equation. The background that used to be your answer becomes a set of footnotes. The engine will give you its summary, an answer to your question.
This leads to big problems and liabilities, on both sides of the screen.
First, most people are going to simply rely on the answers they get and continue with their work. That’s the point of the productivity improvements AI promises.
But what if you, or the search engine, misinterpreted the answer? What if the question were badly formed? What if it turns out you’re wrong?
At the very least you’ll want to go back to those footnotes. What if they’re not there? They’re wiped when you finish a search now. Only those links you saved and put in your own story remain. What if there were lies in others that the engine relied upon to give you the answer?
You can throw all the caveats and boilerplate you want at this. It won’t provide protection when big mistakes are made. They will be made.
The point is that generative AI is going through a human process, for which humans commonly take responsibility. But there’s no way the business model of Google can handle that risk at scale.
Now. What if the search engine had been gamed? What if someone used Search Engine Optimization (SEO) to get false stuff into the answer, or what if someone used AI to create a whole bunch of slightly phony answers to tilt the playing field?
One way to deal with that is for the search engine to only use “reputable” sources in creating answers. Define reputable. Is Fox News reputable? Am I? Someone must decide, before a source’s data is dumped into a Large Language Model (LLM) and turned into answers for some college sophomore. Those that don’t come up to snuff will find their stuff can’t be found. Republican politicians are already telling public servants they can’t talk to the engines about this, and telling the engines to ignore what other politicians might say.
This is going to slow the public-facing, mass-market uses of AI. By years. We will get productivity. Companies can manage the data sets used in their own LLMs, keeping everything proprietary and separate. Many services will improve. Management will get more clarity, faster, on sales channels and supply chains.
Just don’t expect to ask Google about Donald Trump and get a straight answer any time soon.
