I’ve spent the last eight months turning to Google News on my personal playground. I manipulated the algorithm and published my stories whether they were related to specific topics or not. This is a big problem.
I’m a regular reporter – a writer. I have no programming skills or formal computer education.
Google is arguably the most technologically advanced AI company in Silicon Valley. It also happens to be worth more than two trillion dollars.
Google News reaches nearly 300 million users. And I was able to play its algorithms by changing one word on a web page. Scary isn’t it?
We have “reinforcing learning” (RL) to thank for this particular nightmare.
Stupid in, stupid out
As Thomas Macaulay of Neural recently wrote:
[The reinforcement learning] technique provides feedback in the form of a “reward” – a positive number that tells an algorithm that the action it has just performed will serve its purpose.
Sounds pretty simple. It’s an idea that works with kids (you can go outside and play after you’ve finished your homework) and animals (doggo does a trick, doggo gets a treat).
Let’s use Netflix as an example. If you look The Karate Childthere is a pretty good chance that the algorithm will recommend Cobra Kai. And if 10 million people are watching The Tiger Kingthere’s a pretty good chance you’ll get a recommendation on it whether you’ve watched related titles or not.
Even if you never take one of the suggestions of the algorithm, it will still show results because it has no choice.
The AI is designed to seek rewards, and it can only be rewarded if it makes a recommendation.
And that is something we can take advantage of.
The data that feeds Netflix’s algorithms comes from its users. We are directly responsible for what the algorithm recommends. Thus, hypothetically speaking, it would be trivial to exploit Netflix’s recommendation system.
If, for example, you want to increase the total number of recommendations that specific content has received from the algorithm, all you have to do is sign up for X number of Netflix accounts and watch that content until the algorithm is selected. up along the traffic where X is whatever it takes to move the needle.
Obviously it’s a little more complicated than that. And there are safeguards that Netflix can implement to mitigate these threats, such as weighting higher data for older accounts and limiting the influence of those who do not meet a minimum viewing threshold.
At the end of the day, this is not a major issue for Netflix because all content on the platform must be explicitly approved. Unlike Google News, Netflix does not source content from the Internet.
It’s the same with Spotify. We could sign up for 10 million free accounts, but that will last forever and we would still only raise flows for an artist who has already been taken care of on the people platform.
But the Google News algorithm is different. Not only does it source content from the internet and aggregate it in terms of popularity, it also sources important data points from journalists like me.
How I took advantage of Google’s News Algorithms to showcase my own content
Last June, I wrote about the strange impact my TNW author profile had on the stories that Google News appeared for the search engine “artificial intelligence strange.”
As one of the few curious editors in the world in charge of the AI section at a major news outlet, the intersection of artificial intelligence and diversity issues is a place of great interest to me.
Topics on AI and LGBTQ + were also a popular combination for technical reporters to cover at the time because June is Pride month.
I was shocked to discover that a disproportionate number of articles I wrote appeared in the search results.
It was as if Google News had declared me the weirdest AI journalist on the planet. At first, this felt like a victory. I like it when people read the articles I write.
But few of the stories the algorithm came up with had anything to do with the strange community and many also had nothing to do with AI.
I quickly realized that the algorithm was probably popping up in my stories because of my TNW author profile.
At the time, my author page stated that I was covering “strange things” and “artificial intelligence,” among other topics.
So I changed my author profile to say “Tristan covers human-centered advances in artificial intelligence, quantum computing, STEM, GLBTQ + topics, physics and space. Pronouns: He / she.”
And, a few days later, the algorithm stopped popping up most of my stories when I was looking for “weird artificial intelligence.”
But when I searched for “GLBTQ + artificial intelligence topics”, the proportion of my articles to other journalists was even more distorted in my favor.
We can assume that this is because more journalists put “weird” in their profile than “LGBTQ + topics”.
In practice, this means that TNW may have accidentally capitalized on Google News’s search traffic for questions such as “STEM-fourth”, “quantum weird” and “weird artificial intelligence”, based solely on the strength of my author profile.
There’s a better-than-zero chance that one of my reviews of video games or cover-up opinion pieces about Elon Musk may have appeared on top of pieces by other journalists that actually related to weirdness and artificial intelligence. That’s wrong.
In the news industry, these page views (or lack thereof) can affect what journalists choose to cover or what their editors allow them to do. Page views can also cost people their jobs.
Not to mention the fact that news consumers don’t necessarily get the strongest or most popular articles when they search for the Google News app.
When I see people claiming that the media doesn’t cover the stories they think are important, or that the whole field has misunderstood a story, I can’t help but wonder what effect algorithms have on their perception.
I’m against the algorithm
Google News did personal things when it put an election algorithm into production, which took it upon itself to decide that everything I wrote was weird just because I am. That’s a stupid way to take care of the news.
I couldn’t help but wonder how stupid the AI really was. Could I convince Google News to show my TNW articles whenever someone searched for a term I wanted?
The answer is a very sad “yes”. Despite the fact that almost 300 million people use Google News, I can somehow determine what they see when they search for specific topics by just changing one word in my author profile.
I put the word “Spiderman” in my profile and, to the endless amusement of my inner child, I became synonymous with “quantum computing Spiderman” in Google News.
This was not small considering the amount of a delightful quantum specialty Marvel packs in the MCU.
There is a lot of news discussing Spiderman and quantum physics. None of them were written by me. But that doesn’t stop Google News’s algorithm from giving my unrelated bits top privileges in search results.
I didn’t stop at Spiderman. As a kid, Spiderman was my favorite superhero. But Voltron was my favorite cartoon.
If you go to Google News right now (at the time of the publication of this article) and search for any of the following terms, you will find a significant portion of the results returned by the algorithm I wrote.
- Voltron artificial intelligence
- Quantum Voltron
- STEM Voltron
- Physics Voltron
Remember, I never actually wrote about it Voltron. That’s why the algorithm won’t return results for my work if you just search for “Voltron” on its own.
Google’s News Algorithms act as recommendation models despite being deterministic.
This is significant because TNW only puts a link to my author profile on pieces I have published. The actual text in my profile does not appear on the articles I write.
If I change my author profile, the algorithm changes the Google News search results.
Beyond Voltron, aka the scary part
It is a safe bet that my ability to make one of the most popular news aggregators in the world to display my work at will is a bug and not a function.
But, technically speaking, I’m not doing anything wrong. I did not sign a contract with Google stating that I would not change my TNW author profile. This is a Google issue. And it’s much bigger than me or TNW.
If by chance I can find a way to turn Google News into a sandbox for my journalistic jokes, what could someone with really bad intentions do?
Google will probably fix this problem sometime. Maybe some developer will change a single line of code somewhere in the giant Google haystack, and the algorithm will stop crawling author pages. It will be as if none of this has ever happened.
But what about the next AI model? Can doctors or insurance companies exploit Google’s medical AI? Can governments or corporations take advantage of Google Search algorithms? Can extremists exploit YouTube’s recommendation engine?
These issues are not unique to Google. Almost every major technology company from Meta to Amazon uses RL agents who train on public data to make recommendations and decisions.
Just ask Microsoft about Tay, the RL-powered chat it stupidly allowed to train on interactions with the general public.
“Tay” has gone from “people are great” to a full Nazi in <24 hours and I don't care about the future of AI at all. pic.twitter.com/xuGi1u9S1A
– Gerry (@geraldmellor) March 24, 2016
RL models give large technology companies the ability to serve customers, clients, and users on scales that would otherwise be unsustainable. Without them, none of the big tech companies would have the reach (and value) they currently have.
But they don’t necessarily make things better. It should scare everyone to know that streaming entertainment, social media, and news aggregating surface algorithms all work according to the same general principles.
If it’s easier for a journalist to decide which articles you read by playing the algorithm than producing quality work, it will be exponentially harder for people who write quality work to reach human audiences.
The bottom line is: no matter how you feel about the current state of journalism, it can certainly get worse. And, with algorithms like Google being responsible, it will almost certainly do.
I’m probably not the first reporter, SEO engineer or influencer to hit a simple yet effective exploit like this. I certainly won’t be the last.
Fix your shit Google.
Source
Tristan Greene