You’ve probably noticed that Google searches are sometimes a little different these days. Depending on what you search for, sometimes Google will attempt to provide an AI-generated answer. A lot of Googlers have noticed that sometimes these answers get a little wacky. What’s going on? Why isn’t Google just taking users to websites like they used to? And what is the impact of all this AI stuff?

Google’s New Way of Trying to Be “Helpful”

For decades, Google has been the world’s most popular search engine, and for good reason. Historically, Google has provided a good user experience and the search engine thrived to provide the best search results. 

Over the last several years, Google has added more and more features to get you your answer faster. Type in a math equation and Google will give you the answer with a working calculator. Ask Google how many calories are in a banana, and the search engine will give you the answer as well as a table to change your serving size. Google even provides the full nutrition details on the side of the page.

That makes sense to me. I don’t really feel like I need to go to a website to get a simple answer like that. Granted, if a website depends on search traffic to come to it to learn about the nutrition facts of a banana to pay the bills, Google is taking that away from them—something that’s ironic because it’s not like Google itself is doing the work to figure out how much Vitamin C is in a banana, they are just pulling that data from the websites that it indexes. We’ll talk about this a little later, but for really simple inquiries, at least from a user perspective, it makes sense that Google just gives you that content up top.

Over the last few months, Google has been trying to answer more complex questions using its new AI. They label this as AI Overviews, but it’s using its new AI product, Gemini. 

The results have sometimes been pretty funny.

Google’s AI is Giving Some Pretty Bad Advice

Users have been poking fun at Google’s AI overview since it first rolled out, and for good reason.

Back in May, if you asked Google how to get cheese to stick to pizza, the AI would suggest using non-toxic glue. 

An AI isn’t supposed to be able to just make something up like that, right? It’s trained on billions and billions of sentences and information on the Internet, after all. It turns out a single satirical post on Reddit, where a commenter suggested using glue in your pizza cheese, was enough to convince the AI it was worth trying.

While Google has been trying to tweak the system, just last week, the AI suggested that humans should try to eat no more than one small rock every day.

It’s funny! It sort of feels like something a robot would say in Douglas Adams’ The Hitchhiker’s Guide to the Galaxy. We have this incredible technology, but we then decided to have it learn from us, and it’s all going haywire. There’s some beautiful, twisted moral to this, and it’s no wonder that a big chunk of the Internet is joking about it right now.

But this can be dangerous, too.

People do Google searches for lots of reasons. People depend on Google to be correct. If your kid just ate a mushroom on a hike, and Google comes back and tells you that mushroom is non-toxic, when in reality it is very toxic, you are going to have a huge problem on your hands. In fact, that’s a test that The Associated Press did, and found out that some of the information was only somewhat accurate and potentially dangerous.

AI can’t distinguish between truth and conspiracy theories, which is extremely problematic in a world where misinformation is running rampant. It can’t distinguish between good advice and satire. No matter how much work Google does to tweak it, users need to take generative AI responses with a grain of salt, in the same way you might want to fact-check any other information source.

AI Responses are Energy Hogs

Google requires massive amounts of infrastructure and resources to operate. However, for many years, Google Search has more-or-less been optimized to be energy efficient. From the way Google builds their data centers, to their redundancy and content delivery systems, everything is designed to keep Google’s costs and energy usage low.

Back in 2011, Google’s entire infrastructure consumed about the same electricity as 200,000 homes. That’s about a quarter of the output of a nuclear power plant. Keep in mind, this accounts for Google Search, Gmail, YouTube, Google Maps, and everything else that Google had their hands in at the time. 

Over the years, that number has climbed. In 2019, Google used more electricity than many countries, and if Google were its own nation, it would be in the top 90 based on electricity usage.

Google searches are still about as efficient as they used to be, but more and more people rely on Google services each day, so this insane exponential growth just shows how much the world depends on Google and its massive range of services.

Now let’s factor in AI. 

A standard Google search consumes an estimate of 0.3Wh of energy. This is an older estimation from about a decade ago, and I also realize that this figure might not mean much to most people. The energy it used to take to do a Google search was about the same as running a very small, low-energy 10 watt bulb for about a minute and a half.

It’s estimated that a result with an AI response consumes a whole lot more juice—up to ten times more. Two Google AI searches is enough electricity to run a 22-inch LED TV. 

Google receives 99,000 search queries each second, or an average of 8.5 billion searches per day. If AI started showing up in the majority of search results, an incredible amount of electricity and other resources would go towards this endeavor. 

AI’s Impact on the Environment

Most AI processing is done within data centers owned by Google, Microsoft, OpenAI, Amazon, and others. The heavy lifting isn’t done on your smartphone or workstation, so you don’t really feel the impact yourself.

However, data centers are quickly outgrowing electrical grids, which means that energy demands are higher than ever, which means that dirty and less sustainable energy sources remain in operation. Not only do these data centers need power, they also need water to cool their servers, air conditioning to preserve the indoor environment, and plenty of other resources. Over the next few years, the water usage of data centers could equal half that of the UK.

We already saw something similar happen over the last few years with cryptocurrency and Bitcoin mining. As this technology continues to become popular, it will consume more and more power. That’s not necessarily a bad thing if it were always useful.

However, I don’t think Google's suggestion to put glue on my pizza is worth the cost.

AI is a great tool and can be implemented in some very powerful ways for businesses and individuals. Data has a lot of power, and AI can automate and extrapolate things quickly, but like any technology, it’s something you need to use carefully and wisely. 

What are your thoughts? Let us know! And, if you want to learn more about how modern technology can help your business, give Data Net a call at (760) 466-1200!