If you’ve used Google recently and been lucky (or unlucky) enough to come across an answer to your query rather than a bunch of links, you’ve been exposed to something called AI Overview. It’s a new core feature that Google is rolling out, and a move that’s been widely expected since the company experimented with its LaMDA large-scale language model in 2021 and since OpenAI’s ChatGPT artificial intelligence chatbot garnered sudden attention in 2023.
Screen capture: The Conversation, CC BY-ND
The feature is yet another in a growing number of add-ons and tools that are integrated into search engines like Google. Notable examples include the Knowledge Graph-driven Knowledge Panel, which is used to display relevant factual information in an info box next to a search result, and Featured Snippets, which are short descriptive pieces of text that are excerpted from a search result and displayed before a link to that page.
But what makes AI summaries different is that they’re not simply extracted from relevant sources, but are generated behind the scenes by Google’s generative AI technology. The company’s goal is to provide personalized, on-demand answers, rather than a standard set of documents or answer boxes that match your query.
This seems almost magical and potentially useful in many situations – after all, people use search engines primarily to find answers, not to find lists of documents – but that’s not the whole picture.
My colleague Emily Bender and I have written about what search engine users need, want, and have. We have shown that users want not just information, but the ability to discover, learn, and question what they find. In other words, users have a wide variety of situations and objectives, and compressing them into a series of links or, worse, a single answer is problematic.
Bad advice
These AI capabilities suck up information from the internet and other available sources and spit out answers based on how they’ve been trained to associate words. The main objection to this is that it takes little account of user judgment, agency or opportunity to learn.
This may be fine for many searches. Want an explanation of how inflation has affected grocery prices over the last five years or an overview of what’s included in European Union AI law? AI overviews are a great way to comb through lots of documents to extract specific answers.
But people’s search needs go beyond factual information. They want ideas, opinions, and advice. Wondering how to stop cheese from sliding off your pizza? Google will tell you to add some glue to your sauce. Or want to know if running with scissors has health benefits? Of course, Google will tell you, “It can also improve your pores and make you stronger.”
A rational user would know that such outlandish answers are likely wrong, but that’s hard to detect with factual questions.
For example, when researching the religious beliefs of US presidents, Google’s AI Overviews incorrectly stated that Barack Obama is a Muslim. This misinformation was widely circulated and debunked years ago, but Google repeated it verbatim and users had no good way of knowing it was false.
What about a student who uses Google for homework to ask what country in Africa starts with the letter K? Kenya meets this criteria, yet Google’s AI Overviews incorrectly responds that no such country exists.
Google has acknowledged and said it has addressed the issues with AI Overview, but concerns remain: can you really trust the answers you receive through the service?
How to avoid AI answers
Yes, there is an alternative. You can always go back to classic Google Search with its 10 blue links:[その他]Click[ウェブ]Choose.
You can then do what you’ve probably been doing for decades: sift through some of the top results, visit some of those sites, and make your own judgment. It takes a little effort, but it gives you back the ability to look at multiple sites and evidence to support or disprove something, and more importantly, it leaves open the possibility for learning, discovery, and serendipity.
AI Overviews is like fast food delivered through a drive-thru: it’s quick, hot, and convenient, but it’s not the healthiest option. Looking through Google’s traditional search results is like looking at a menu at a sit-down restaurant and placing an order that takes a while to get to your table. You might ask the server about those items or even request modifications to the restaurant’s menu. It’s prepared with more care, customization, and control, but it can take longer and be more expensive.
However, this is not the only way to find information. There are alternatives to the Google search engine, including specialized search tools.
For your academic needs, Google Scholar, Semantic Scholar, and CORE can help you find research articles and citations. Looking for medical information? Try PubMed, ScienceDirect, and OpenMD. For your legal needs, there are services like Fastcase, Caselaw Access Project, and CourtListener.
Worried about privacy? Check out DuckDuckGo, Startpage, and Swisscows. But if you still want AI-generated answers, alternatives to Google’s AI Overviews and rival Bing’s Copilot include You.com and Komo, which offer greater transparency about the data they collect about you, stronger privacy, and ways to opt out of having your data collected to train AI models.
A balanced information diet
While you probably can’t afford to eat out at fancy restaurants or cook from scratch every time, it’s important to avoid the drive-thru sustenance traps. After all, you are what you eat, and in the same way, you are how you search.
It’s easy to be fooled by sensational headlines and short news stories without context. But it doesn’t have to define you. You can broaden your search. It’s okay to check out the AI brief at the drive-thru every now and then, but it’s also important to find healthier ways to satisfy your needs, like food and information.
