My answer may be controversial.
Google was once the go-to search engine for the internet. “Google it” even became a generic term for looking up something online. But there’s been a big shift to a different type of answer engine: AI chatbots like ChatGPT and others.
Let’s review the pros, the cons, the risks, and the impact of these choices.
First, we need to review the differences between the two.
AI for search
AI and search engines work differently, and each has strengths and weaknesses. Search is good for fresh, direct lookups, while AI is better for questions, summaries, or creative tasks. Both can be wrong or biased, so check sources. The best answer? Use both and always stay skeptical.
Different in concept
Search engines like Google, Bing, and others work by amassing huge indexes of information from having crawled (AKA read) all the webpages on all the websites they can access. These indexes, which consider hundreds of different factors, are used to answer the question, “Which pages on the internet best represent the terms being searched for?”
AI services, on the other hand, are “trained” on massive datasets that range from specific sets of training data to the same “all the webpages on all the websites they can access”. The goal of all this training is to answer the question, “What words would best follow the words that have been typed in by a user?”
Put simply, a search engine is basically a lookup — get pages that relate to a word or phrase — while AI (specifically large language models) is glorified auto-complete — return the best next word, and the next, and the next, and so on.
That AI works in place of a search engine is because generally, the “best next words” to follow any question or term are often the very answer we’re looking for.
Now that we have an idea about how each works, we can compare the accuracy, citation, timeliness, bias, and environmental impacts of both.
Help keep it going by becoming a Patron.
Related
Where Do You Get Your Answers?
There are many ways to find information on the internet. Here are a few tricks of mine that can work for you.#2192
AI can hallucinate. Search can miss.
I’m sure you’ve heard of AI “hallucinations” — an AI response that is completely wrong. That whole “best next word” model has nothing to do with accuracy. If the most appropriate response to a question doesn’t exist, AI can make it up. AI doesn’t know how to say “I don’t know”.
This is all a function of the training data. The more data AI gets trained on, the more likely it is to have an appropriate and correct answer. Yet, if the data the AI has been trained on is wrong or misleading, the AI wouldn’t know.
Search isn’t perfect either, though. It relies on the way a search is phrased, so ambiguous and incomplete search terms can lead to wildly off-topic results. In addition, search engines are constantly being “gamed” by individuals trying to get their pages (often poor quality, irrelevant, or even malicious) to rank higher.
Both require us to maintain a healthy level of skepticism and independent confirmation. Unfortunately, many people skip this step for either search or AI.
An answer versus a reference
AI prioritizes giving you an answer to your question or search terms. It may or may not include references to where the answers came from.
A search engine prioritizes giving you references to webpages that presumably contain what you’re looking for. Many search engines also include an AI-generated summary that may or may not suffice as the answer you’re looking for.
Freshness
Most search engines constantly scour the web for up-to-date information. Websites that change often or produce timely information, such as news sites, are crawled more frequently to include their latest information in search results.
AIs typically train on a snapshot of data, meaning that what they ‘know’ is only as current as when that snapshot was taken. To overcome this limitation, some AIs augment their responses with searches that are then factored into their responses.
Bias
Search results are heavily influenced by ads (particularly when those ads are difficult to distinguish from organic search results), SEO tactics used by people trying to game the system, and general website and page popularity.1
More than anything, AI is a product of its training data. If that data is limited in scope or biased in some way, then the AI will be similarly limited and biased. AI is also subject to “guardrails” implemented by each provider attempting to prevent it from generating responses that would be considered inappropriate.
Impact
Perhaps the most controversial topic when it comes to AI (well, next to whether or not it’s making us dumber or will lead to our extinction) is its impact on the environment. AI uses a lot of energy. Sure, the Google data center powering Google Search uses a lot of energy as well, but depending on what you measure (electricity use, CO2 generation, or something else), AI is somewhere between ten and hundreds of times more energy-hungry.2
Why? Because a search is a lookup: take the search terms and look up the most appropriate pages from the index of pages crawled. An AI response is a computation, and that takes more work. It’s the difference between looking up a word in a book’s index versus having to write a paragraph about the word from scratch.
Related
My take on it all
Here’s what I do.
- If I have something that feels like a traditional search — maybe I want to find the website for a company, look up the definition of a word, or something that needs to be as up-to-date as possible — I’ll use a traditional search engine. These days, for me, that’s Kagi3.
- If I have something that feels more like a question — perhaps even something I consider half-baked — I’ll throw it at an AI-based search engine like Perplexity4. AI seems to do a better job of “understanding” what I’m trying to ask. I’ve gotten great responses by just pasting in an error message as my “question”, for example. I particularly appreciate that Perplexity includes explicit references to the sources of its answers.
- And if I want AI to “do” something like summarize an article, suggest what aspects of a topic I might be missing, or similarly creative tasks (including image generation), I head to ChatGPT5.
In other words, I use both search and AI for what I feel each is best at right now.
More important than anything else, though, is that I remain skeptical of what I’m told, regardless of where it comes from. I try very hard not to let any of these tools (or any others I might try) lull me into complacency. The answers have to pass the sniff test, and I check references if I’m not 100% certain.
Do this
ChatGPT actually put this the most succinctly:
“Instead of” might not be the right question. AI and search engines serve different strengths.
So that’s my suggestion: why not both? Experiment with AI and find out where it works best for you, but don’t hesitate to use a search engine for the things it’s best at.
Above all, remain skeptical of both. Never blindly accept answers without verifying their accuracy.
Subscribe to Confident Computing! Less frustration and more confidence, solutions, answers, and tips in your inbox every week.
Podcast audio
Footnotes & References
1: Normally, when I mention “bias”, I get a lot of commentary about political bias from people who believe that a search engine has been explicitly tailored to exclude results that don’t agree with some assumed political agenda. This amuses and frustrates me because I get it from both sides: the left believes search leans right, and the right believes that the same search engine leans left. I think it’s more likely that the individual doesn’t like or agree with the results that are, on the whole, relatively objective. Nonetheless, people will yell at me, convinced it’s otherwise.
2: I acknowledge this as a current issue. I think it’s an issue that will diminish over time as more and more renewable energy sources come online.
3: Kagi is a paid service. I do not get any compensation from them for mentioning them.
4: I pay for Perplexity as well, and they also do not compensate me for the mention.
5: I pay for ChatGPT. They don’t pay me.
I started using Kagi as my default search engine six months ago. I am firmly in the “Search Only” camp. I don’t do much as far as creating pictures or writing articles or letters and my experiments with AI (CoPilot, Chatgpt) left me with a “OK, that’s mildly interesting” feeling. Microsoft ticked me off with their efforts to push CoPilot so much, I’ve removed it from my computer. Except for Edge, I’ve not figured that one out yet.
I don’t feel that limiting myself to just using a search engine is causing me to miss something. I’ve learned that if the first set of results doesn’t yield what I’m looking for, changing the original search terms usually yields what I’m looking for.
This doesn’t mean I’m totally ignoring what is going on with AI. I just discovered recently that there is a thing called the “AI Darwin Awards”.
As to environmental concerns here in Georgia- Meta, Google, Microsoft and other companies are planning data centers around the state. Georgia Power is keeping two coal-fired plants on line and is requesting to add more gas-turbine plants to meet expected demands. That’s after bringing on line two additional nuclear plants. Little to nothing about renewables.
You might want to look into the election this November for two of the Public Service Commissioner seats in Georgia. Renewables are one of the issues, but off year elections can sometimes slip under everyone’s radar.
Google and other search engines now give an AI summary before the search results, so it’s a matter of ignoring the AI summary if you want to check the search results. This is a disaster for sites like Ask Leo! because people searching for an answer receive an AI summary and don’t have to click on a link to access the article(s). This can wipe out a website’s ad revenue.
It’s the worst of both worlds, as many sites will go down due to a lack of revenue.
What I do to avoid AI responses is to put -AI then the search query in the search bar. I don’t get AI responses then. For example: “-AI How many people have visited the 9/11 site ?” Only “normal” responses come up then.
Good tip. Which search engine(s)?
AI is learning everyday. Humans’ are stopping learning everyday. What could go wrong? Maybe “SkyNet” is closer than we think?
Hello Leo,
I appreciated your thoughts on search engines and AI. One angle I’d add is the effort required when using a traditional search engine. Google may give me a page of links, but that forces me to open and scan through each one, often many times over, before I find what I’m really looking for. With AI, I get not just the facts, but those facts framed in the way I asked the question. And I did ask it in the form of a question that search engines would not understand.
For me, that difference is important. Search gives me abundance, but it leaves the burden of sorting and filtering on my shoulders. AI, on the other hand, gives me clarity right away, which is often what I value most.
I’m curious if others here also feel that extra burden when using search engines, and whether AI has helped lighten it for them.
I use AI extensively in my work with AskLeo!, not so much to answer questions, but to get help with phrasing the answers correctly. When I do ask it a question, I sometime get a completely wrong answer. The AI bot then apologizes. All that to say that depending on AI for answers can be hazardous. I can’t tell you which AI bots because I have 5 AI bot shortcuts on my browser bars. Sometimes when I detect a wrong answer, I’ve gone to a different website and asked the same answer. Surprisingly, n some cases, the other bot gave the same answer.
Gemini fixed that. 😉
Relying on AI for answers can be risky. While I use AI extensively in my work with AskLeo!, I mainly use it for help with phrasing my answers correctly, rather than for answering questions directly. When I do ask a question, I sometimes get a completely wrong answer, and the AI bot will then apologize.
Because I have five different AI bot shortcuts on my browser bar, I can’t tell you which bots I’ve used. When I detect a wrong answer, I sometimes go to a different website and ask the same question. Surprisingly, in some cases, the other bot gave the same incorrect answer.
Hello Leo, I’d like to add a bit more to my previous comment.
Something else I’ve noticed with traditional search engines is the challenge of narrowing results. To really focus a search, I feel like I have to speak a kind of foreign language by using quotation marks, minus signs, or other cryptic rules that I’ve never mastered. That’s a friction point for me, because I have to adapt to the machine’s language before I even get to the long list of links.
With AI, I can just ask in my own words and let the system do the heavy lifting. This becomes a shift from me bending to the search engine, to the tool adapting to my language makes a huge difference in how usable the results feel.
Do others here find the “syntax” of search to be a friction point, or have you learned to work around it?
Hi, Leo et al,
I disagree with you that AI doesn’t know how to say “I don’t know”. I have begun using Duck Duck Go’s “Search Assist”, which I really like (and it’s free)! Anyway, the other day I was asking it to find the current positivity rates for Covid in a certain state (since the CDC has removed that page, online), and it said something like, “Sorry, I can’t find any relevant information on that”. (And that’s not because DDG AI Assist isn’t good, but because that info has been all but removed online!)
So that’s AI way of saying, “I don’t know”.
In addition, one can ask DDG’s Search Assist if you want to see it in your searches; and if so, how often. Great feature. So I can turn it off if I like. And it lists both its summary and the search engine hits.
(And btw, if you like what DDG AI came up with you and tell it, it’ll reply, “Quack! I’m glad”. Which is neither here nor there, but it’s cute.)
I am personally ambivalent towards AI usage (hate most of the way it’s used on You Tube, for example), but I tried this Assist out two weeks ago because I was in a hurry– and it’s a keeper for me. I also glance at the search engine hits, though. So I do both, as you do.
Enjoy your articles,
Many blessings
I was trained as a librarian, the person who sits at a desk and answers questions. Librarians are (were) trained to know the difference between reliable and unreliable sources. Google was good in its early days, now it is mostly ads; I prefer Duckduckgo for general searches, and Bing for some other searches. The AI answers that come up in both Google and Duckduckgo are usually pretty concise and provide a couple of sources which can be followed for more in depth research.
I’ve found ChatGPT useful for answering issues with computer software.
For example, today my PDF editor (Kofax) started printing pages with an unwanted pattern superimposed. The ChatGPT dialogue went like this:
I asked: WHY IS KOFAX PRINTING PDF WITH PATTERN SUPERIMPOSED
ChatGPT answered:
/============================
If Kofax is printing PDFs with a pattern superimposed, it might be due to several reasons:
Watermark Settings: Check if there’s a watermark setting enabled in the Kofax software or the printer settings. This could be a predefined pattern or logo.
Driver Issues: Ensure that the printer drivers are up to date. Outdated or corrupted drivers can lead to printing anomalies.
PDF File Issues: The PDF itself might contain a watermark or a background pattern. Open the PDF in another viewer to verify.
Print Settings: Review the print settings in Kofax. Sometimes, settings related to “Print as Image” or other advanced options can cause unexpected results.
Software Glitches: Restarting Kofax or the computer might resolve temporary glitches.
Printer Configuration: Check the printer’s configuration settings for any patterns or overlays that might be set.
If none of these solutions work, consider reaching out to Kofax support or consulting the printer’s user manual.
/========================================
I found this a useful checklist. Mostly obvious but good for completeness.
Hi Leo and all Responders.
Leo, you have addressed many questions I had re AI and searches, and yes, I have a healthy level of scepticism.
What I really appreciate is the intelligent and constructive criticism from other responders. You attract a very helpful audience, I’ve learnt a lot, thank you all very much.
My initial ” deep dive” into AI.
https://spoonbillcourier.com/communicating-with-machines-a-personal-look-into-our-ai-future/
AI is great at answering specific questions. And the more detailed and explicit is the request, the better is the answer. For example, “How do I write a Python script that is queued from a shell in the RHEL OS and counts the number of commas in an ASCII text file?” But if you ask questions like “Does God exist” or “What is the best hot sauce?” or “Why are the so vile” then be prepared for silly, wrong, or biased answers. And realize that if you ask questions like “What is the {removed, but you get the point} then don’t be surprised if you have a visit from the authorities.
Today, I asked Copilot what’s are best rated submarine sandwich shops near the airport. That’s specific and I got a list of shops with their customer ratings. You can use that kind of phrasing to get an objective answer. Still, AI has been known to get things wrong, AKA hallucinating.
Unrelated question
Do you use “” around words as scare quotes (meaning so called” OR just for emphasis of that word meant to intepret as its literal meaning ?
I belong to India. Here “” can be used for emphasis and not as scare quotes.
I want to know what you mean when you use “” around single words?
I use quotes when actually quoting someone else’s words, and as you indicate, to mean “so called” or to otherwise cast doubt on the meaning of the word or phrase being quoted.
If I want to emphasize I’ll use italics or bold.
Who is this guy AI you keep referring too? Seriously though, A.I. has taught me to have a deeper hate for crappy fonts because WAY too many make the lower case “L” and the upper case “i” look exactly the same. You’d think after the old “expIorer.exe” virus of yester-decade, we would know better (look again, that’s not a lower case “L”).
It even confused the Secretary of Education who talked about A one in school.
“A school system that’s going to start making sure that first graders, or even pre-Ks, have A-One teaching in every year. That’s a wonderful thing!”
I use Firefox as my default web browser, and have set it as my default on both Windows and in Garuda KDE-Lite GNU/Linux. I have kept the majority of Firefox’s settings as set when I installed it, with the exception of the Appearance and Home page content settings. Firefox defaults to the Google Search engine, and when I do an Internet search in Firefox, I read the AI generated response at the top of the results, then I go through the first several results in the search listing further down the page (beyond the Ads). I find that the summary at the top usually provides the correct information I’m looking for, based on the content of the first few web sites listed.
When I want to search for technical answers, one of the first places I look is on Ask Leo. Nine times out of ten times, I find exactly what I’m looking for, and in clear, concise, and understandable terms. Thank you Leo for the collection of past items on your web site 🙂
Ernie
The impression I have gotten about the AI on web pages is that it’s using the search results to kind of summarize an understanding of what you would have to otherwise sift through. Very VERY useful and time saving.
Not to mention cutting out the advertising until the AI companies start to stick the advertising into the AI search results. Which we all know is going to happen.
Whether the answers are correct / it utilized good existing answers.. that’s another question. Also as pointed out the whole hallucination effect.
AI’s aren’t really intelligent. They’re overglorified databases. The answers they come up with are based on the intelligent (or not so) writing of a human beings.
You can’t duplicate the mind of a brilliant scientist who comes up with stuff no one else has thought of. The best you can do is feed off of the advancements that they presented during their life.
The way I look at IQ is not that it represents superior knowledge, but rather that it is able to take knowledge from different fields and endeavors and create something different or new from it. Even advances from brilliant individuals in specific fields or mathematical concepts likely pulled ideas from other places and stuck them in either deliberately or otherwise.
Think about it like this — You can feel (touch) things around you and you can see light and so forth and get a perspective on all of these things. We also live in concern for our mortal coil. We love, hate, and every other emotion. AI’s do not exist at all in our reality. No amount of physics or other knowledge can really overcome that.
There are an endless variety of things they just flat out can’t experience.
We’re always going to be a superior intelligence for our own reality.
More than overglorified databases, I’d call them overglorified dataminers. The database is already out there. It’s the World Wide Web.
As for the environmental consequences of searching: a relative of mine pointed me to the Ecosia browser (ecosia.org). It is specifically targeted at protecting the environment and even use part of its revenue to plant new forests. Ecosia can be used as search engine in your browser of choice, but there is also a separate Ecosia browser. Wikipedia has a page about Ecosia too.
I use the search engine now as my default in Chrome on my laptop, but also installed the browser as default on my phone.