Every time you pick up your smartphone, you’re summoning algorithms. They’re used for everything from unlocking your phone with your face to deciding what videos you see on TikTok to updating your Google Maps route to avoid a freeway accident on your way to work.
An algorithm is a set of rules or steps followed, often by a computer, to produce an outcome. And algorithms aren’t just on our phones: they’re used in all kinds of processes, on and offline, from helping value your home to teaching your robot vacuum to steer clear of your dog’s poop. Over the years they’ve increasingly been entrusted with life-altering decisions, such as helping decide who to arrest, who should be released from jail before a court date, and who’s approved for a home loan.
In recent weeks, there has been renewed scrutiny of algorithms, including how tech companies should shift the ways they use them. This stems both from concerns raised in hearings featuring Facebook whistleblower Frances Haugen and from bipartisan legislation introduced in the House (a companion bill had previously been reintroduced in the Senate). The legislation would force large tech companies to allow users to access a version of their platforms where what they see isn’t shaped by algorithms. These developments highlight mounting awareness about the central role algorithms play in our society.
“At this point, they are responsible for making decisions about pretty much every aspect of our lives,” said Chris Gilliard, a visiting research fellow at Harvard Kennedy School’s Shorenstein Center on Media, Politics and Public Policy.
Yet the ways in which algorithms work, and the conclusions they reach, can be mysterious, particularly as the use of artificial intelligence techniques make them ever more complex. Their outcomes aren’t always understood, or accurate — and the consequences can be disastrous. And the impact of potential new legislation to limit the influence of algorithms on our lives remains unclear.
At its most basic, an algorithm is a series of instructions. As Sasha Luccioni, a research scientist on the ethical AI team at AI model builder Hugging Face, pointed out, it can be hard coded, with fixed directions for a computer to follow, such as to put a list of names in alphabetical order. Simple algorithms have been used for computer-based decision making for decades.
Today, algorithms help ease otherwise-complicated processes all the time, whether we know it or not. When you direct a clothing website to filter pajamas to see the most popular or least expensive options, you’re using an algorithm essentially to say, “Hey, Old Navy, go through the steps to show me the cheapest jammies.”
All kinds of things can be algorithms, and they’re not confined to computers: A recipe, for instance, is a sort of algorithm, as is the weekday morning routine you sleepily shuffle through before leaving the house.
“We run on our own personal algorithms every day,” said Jevan Hutson, a data privacy and security lawyer at Seattle-based Hintze Law who has studied AI and surveillance.
But while we can interrogate our own decisions, those made by machines have become increasingly enigmatic. That’s because of the rise of a form of AI known as deep learning, which is modeled after the way neurons work in the brain and gained prominence about a decade ago.
A deep-learning algorithm might task a computer with looking at thousands of videos of cats, for instance, to learn to identify what a cat looks like. (It was a big deal when Google figured out how to do this reliably in 2012.) The result of this process of binging on data and improving over time would be, in essence, a computer-generated procedure for how the computer will identify whether there’s a cat in any new pictures it sees. This is often known as a model (though it is also at times referred to as an algorithm itself).
These models can be incredibly complex. Facebook, Instagram, and Twitter use them to help personalize users’ feeds based on each person’s interests and prior activity. The models can also be based on mounds of data collected over many years that no human could possibly sort through. Zillow, for instance, has been using its trademarked, machine-learning assisted “Zestimate” to estimate the value of homes since 2006, taking into consideration tax and property records, homeowner-submitted details such as the addition of a bathroom, and pictures a house.
The risks of relying on algorithms
As Zillow’s case shows, however, offloading decision-making to algorithmic systems can also go awry in excruciating ways, and it’s not always clear why.
Zillow recently decided to shutter its home-flipping business, Zillow Offers, showing how hard it is to use AI to value real estate. In February, the company had said its “Zestimate” would represent an initial cash offer from the company to purchase the property through its house flipping business; in November, the company took a $304 million inventory writedown, which it blamed on having recently purchased homes for prices that are higher than it thinks it can sell them.
Elsewhere online, Meta, the company formerly known as Facebook, has come under scrutiny for tweaking its algorithms in a way that helped incentivize more negative content on the world’s largest social network.
There have been life-changing consequences of algorithms, too, particularly in the hands of police. We know, for instance, that several Black men, at least, have been wrongfully arrested due to the use of facial-recognition systems.
There’s often little more than a basic explanation from tech companies on how their algorithmic systems work and what they’re used for. Beyond that, experts in technology and tech law told CNN Business that even those who build these systems don’t always know why they reach their conclusions — which is a reason why they’re often referred to as “black boxes.”
“Computer scientists, data scientists, at this current stage they seem like wizards to a lot of people because we don’t understand what it is they do,” Gilliard said. “And we think they always do, and that’s not always the case.”
Popping filter bubbles
The United States doesn’t have federal rules for how companies can or can’t use algorithms in general, or those that harness AI in particular. (Some states and cities have passed their own rules, which tend to address facial-recognition software or biometrics more generally.)
But Congress is currently considering legislation dubbed the Filter Bubble Transparency Act, which, if passed, would force large Internet companies such as Google, Meta, TikTok and others to “give users the option to engage with a platform without being manipulated by algorithms driven by user-specific data”.
In a recent CNN Opinion piece, Republican Sen. John Thune described the legislation he cosponsored as “a bill that would essentially create a light switch for big tech’s secret algorithms — artificial intelligence (AI) that’s designed to shape and manipulate users’ experiences — and give consumers the choice to flip it on or off.”
Facebook, for example, does already have this, though users are effectively discouraged from flipping the so-called switch permanently. A fairly well-hidden “Most Recent” button will show you posts in a reverse chronological order, but your Facebook News Feed will go back to its original, heavily moderated state once you leave the website or shut the app. Meta stopped offering such an option on Instagram, which it also owns, in 2016.
Hutson noted that while the Filter Bubble Transparency Act clearly focuses on large social platforms, it will inevitably affect others such as Spotify and Netflix that depend deeply on algorithmically-driven curation. If it passes, he said, it will “fundamentally change” the business model of companies that are built entirely around algorithmic curation — a feature he suspects many users appreciate in certain contexts.
“This is going to impact organizations far beyond those that are in the spotlight,” he said.
AI experts argue the need for more transparency is crucial from companies making and using algorithms. Luccioni believes laws for algorithmic transparency are necessary before specific usages and applications of AI can be regulated.
“I see things changing, definitely, but there is a really frustrating lag between what AI is capable of and what it’s legislated for,” Luccioni said.