If I want to know “what’s Google say” it would be impossible then.
Also as far as the universal algorithm: Let’s say that I search for a combo that not many people would search for or ever will. I get a response, do it again a week later. Have I affected the algorithm for that search? Have I made it so my search result is the whole of the algorithm, so that others will have the results that I found?
Google results are highly manipulated based on prior data from the user. An easy demonstration is a seach like ‘tacos near me’ which will show results for your location.
Google tailors its results to what it knows about you, which is not just your search history, but is also anything that google has managed to learn about you. I don’t think google has ever published exactly what data it keeps about you, but we know that they examine tracking cookies and track what you do on other sites, and especially on google-owned sites like youtube.
You can disable google’s personalized search results. Go under your search settings and look for the section labeled “private results”, and select “Do not use private results”.
If you go under your google account and select Manage Your Google Activity, then select Go To Activity Controls, you can enable or disable what data google tracks on your account, including:
Web & App Activity (what you do on any web site or app)
Location History (your physical location, based on IP or GPS)
Device Information (contacts, calendar events, what apps you have installed on your device, etc)
Voice & Audio Activity
YouTube Search History
YouTube Watch History
Google doesn’t need history on you to give you tailored search results. Their AI engine will start making assumptions about you even if all they have is your location, the time of day, and the keywprds you use. They compare the data you give simply by going to the website to the millions of people they do know about and presto, you get tailored results.
So yes, they do track the heck out of everyone, but all that data also makes it easy for them to make educated guess about people they haven’t tracked.
Google considers what type into their search box as nothing more than an indication of what you want. They’ve learned that what you want isn’t always clear from what you type, so they are always trying to figure out what you really want, and they use any and all data available to decide what that is. You can’t stop Google from analyzing every keystroke and giving you tailored results.
Simple example: If I search for “Lakewood library”, I get information on the public library a mile down the road from me. Which is probably what I wanted. But if someone in Alabama, or Alaska, or Colorado, made the same search, it’d be absurd to give them that result (at least, near the top), because Lakewood, OH isn’t the only Lakewood in the country, and I think the one in Colorado is bigger.
Page Rank, while not the only factor in Googles returned results today, is based off Markov chains and not AI.
While the list of what they use is not public, the primary components have more to do with graphs, probabilities and chain absorption. ML models probably just modify those weights based on location, user history etc…
If I understand it, when someone makes a google search and uses it as a cite they are only citing a personalized result which is not reproducible. If the user opts out of cookies or whatever it will make other assumptions and tailorings and still do this to some lesser extent.
But what about my hypothetical where it’s an unusual search. If I have done it 3 or 4 times it will be the only times it has ever been done. Have I skewed the future searches of everyone by this?
It’s weird because people are very sanguine about what a google search is, and that it is that same thing for others, but it is never the same thing twice apparently.
It is fundamentally different, PageRank is linear algebra, which ML/DL may use, but overfitting, and problems with generalizing of training data restrict it as a primary ML user. Here is a link to the original paper.
Linear algebra may be the foundation of machine learning, but not all things that use linear algebra is ML (unless you are talking to angel investors, then everything is AI)
The insanely low comparative computational cost is the amazing thing about PageRank.
What distinction are you making that calls some computer programs “AI” but not others?
And yes, of course you can specify a state in your Google searches, but it’s easier not to, especially when the exact thing you’re looking for comes up as a suggested autocomplete anyway.
Here is an example that I find somewhat creepy in that google obviously knows more about me than what I have searched on google, but based on my browser history.
A real estate agent sent me an email about a property she thought might interest me. It included a link with a number of photos. Her name is Andrea N___. Later I wanted to go back to those photos and started to google Andrea, intending to add her full name and “real estate” figuring that that would pin her down. I got as far as Andrea and it already had her web site as the top suggestion. So google knew my browser history.
If you’re signed into google and haven’t signed out, yes, it knows your google search history. Easy way to tell is go to google.com and look at the top left of the page and see if there’s a “sign in” link there or a symbol (in my case, a “P”) indicating that you’re signed in. And then you can go check out history.google.com to see what websites you’ve visited while you were signed in on a device.
As an example a comparison of two different searches. My default engine, StartPage, passes searches to and gets results from Google’s engine but strips out the data google uses for personalization. For the second I used Google; I do have personalized results turned off and I am signed out of my account. Every piece of history that can be turned off or deleted in Google’s settings is also off/deleted.
The first page of results for Chinese restaurant have only the number one result in common. It’s a restaurant three time zones away from me. Google itself includes a number of local or extended area restaurant pages. StartPage includes some restaurants but none are in my state. It’s two radically different pages of search results.
Google certainly remembers what you searched for, and what results you clicked on, as an example.
Not sure if Google uses a cookie to track your PC uniquely even if you go from starbucks to starbucks - but I read a item once that suggested you basic settings, browser version, and available programs and versions (i.e. adobe PDF reader) and add-ons - which your browser reports automatically - are sufficient to almost uniquely identify your PC after you’ve used it a while.
Geography is easy - even spamly ads know you location by IP address. Most addresses are easy to geolocate. Most advanced routers have the option to geo-block; i.e. don’t allow logins from Romania.