Machine learning powered ads are here to stay. I'm not endorsing it in any way, but I see its ultimate power over any other method currently in use. And I fear this is how it is going to stay for at least the near future.
Google uses machine learning to target users and show them the ad they are most likely to be influenced by. Last week, I started seeing cold medicine all over the web. When I open YouTube, every successive ad was for cold medicine. When I read a blog, there is a neat cold medicine ad on the corner. All this without me ever googling or clicking on anything related to it, or even sneezing. But somehow they were showing me the ad. If only I had taken it as a sign that I was about to get a cold.
Machine learning works by taking in an excessive amount of data and finding patterns. In my case, I have an android phone that, as far as I know, regularly pings my gps location and track some of my activities. As a matter of fact, all android phones do the same thing, collecting a massive amount of information. This is how Google conveniently determines foot traffic in a restaurant (Popular times).
To determine popular times, wait times, and visit duration, Google uses aggregated and anonymized data from users who have opted in to Google Location History.
Or traffic on the freeway
Google Traffic works by analyzing the GPS-determined locations transmitted to Google by a large number of mobile phone users. By calculating the speed of users along a length of road, Google is able to generate a live traffic map
But how can they determine I will have a cold before I even know it? All they have to do is look at users with similar activities as me.
Let's say that last week, 5 users who live in the vicinity of LAX (Location A) went to a particular starbucks (Location B), went to a park (Location C), drove for one hour on the freeway (Activity A), went for a run (Activity B), went to a Ralphs grocery store (Location D), then headed home by LAX (Location E). The next day, they googled something in the context of Cold medicine (Activity C) and clicked on the first results which was an ad.
Let's put that on a table.
Collected User Activty
|User||Location A||Location B||Location C||Location D||Location E||Activity A||Activity B||Activity C|
I have no idea who User 1 to 5 are, but looking at this data, I am sure we have something in common. And we also happen to have been hanging out in the same areas. Chances are, if all of them have been googling cold medicine, or bought it, I may be in a susceptible condition to also do it. Maybe we ate the same thing, or shook the same persons hand that was sick, or did something that exposed us to the cold virus. All the machine learning algorithm cares about is that 5 people in this group got the common cold, so even if I didn't know it yet, it was coming for me.
Of course, I got the cold too.
This is a very efficient way of advertising with a very high success rate.
Google Ad revenue & forecast - Source: Recode
Granted that my example is a very, very tiny in scale. Google runs this sort of program with many more parameters, and thousands if not millions of users at a time. Traditional ads have relevancy and click through rate usually in the sub 2%, but machine learning creates a huge improvement.
Many online ads are only paid for when someone clicks on them, so showing you the right ones translates very directly into revenue. A recent research paper from Microsoft’s Bing search unit notes that “even a 0.1 percent accuracy improvement in our production would yield hundreds of millions of dollars in additional earnings.”
More data means a better tuned machine learning algorithm, which means more revenue. Every quarter a company has to show an increase in revenue to satisfy shareholders, as a result more of our data is needed to feed the machine to produce more money. There are advertising networks that promises privacy, but that sounds crazy. The whole point of advertising is to target people who will buy your product. That's why I think the machine learning ads will continue to grow. They generate more revenue.
It sucks that the greatest minds of our generation continue to use their talent only to get us to click on ads.