Facebook AI: An Honest Assessment
Facebook is all-in on artificial intelligence (AI).
The social network media company has hundreds of people working on AI.
Fast Company reports that CEO Mark Zuckerberg has "tripled [the company's] investments in processing power for AI and machine learning research" in recent years.
This isn't always a good thing.
Facebook's work on AI has moved the field forward. But it's also created some terrible outcomes for users—outcomes like the spread of misinformation, division, and hate speech.
What is Facebook AI?
Facebook AI is the name that the social network giant gives to its internal work on artificial intelligence.
Facebook AI heavily features the company's own AI research, including research papers and open source AI tools it has developed. Facebook AI also has robust commentary on the field and information on Facebook's presence at academic and commercial AI events.
The head of Facebook AI is Yann LeCun, a legendary AI researcher who pioneered some of the first commercial applications of machine learning.
There is also a Facebook AI residency program, a year-long training program where individuals work on AI projects within Facebook in tandem with the company's own researchers.
Facebook says its "AI researchers work from our offices around the globe: Menlo Park, New York City, Seattle, Pittsburgh, Montreal, Paris, Tel Aviv and London."
How does Facebook use AI?
Facebook doesn't just research and develop AI. The company's platform relies on the technology to work.
Facebook's AI capabilities start with text. The company's DeepText system is a deep learning engine that understands text on the platform with near-human accuracy.
Composed of several neural networks, DeepText uses these networks to process the written word as it's used on Facebook.
The resulting data can be used in many ways: the company can now understand more than 20 languages to serve a global audience and better understand queries in its Messenger app.
Thanks to these developments, Facebook is testing recommended responses in Messenger, which offer a logical next response to a friend or family member's message.
As Google and Amazon create their own voice and chatbot assistants, AI will likely play a large role in how Facebook develops its own messaging and artificial assistant capabilities.
Facebook's image recognition (a type of AI) auto-classifies what's happening in images without human captions or tags, allowing users to search photos using keywords even if images are unannotated.
Facebook's DeepFace AI system is responsible for image identification, and at launch in 2014 it was 97 percent accurate (beating out an 85 percent accurate system used by the FBI).
Recent developments also allow the company's machine learning algorithms to automatically annotate photos with text, so that they're more accessible to blind users and more easily searchable by all.
In 2016, Facebook open-sourced a number of its image recognition tools in the hopes this would accelerate facial recognition progress even faster.
Facebook already uses algorithms to determine which content appears on your News Feed. This has been a defining feature of the product since the beginning, though the algorithms have changed over time.
But today, says the company, AI is responsible for more than just guessing what content you'll like.
"Machine learning models are part of ranking and personalizing News Feed stories, filtering out offensive content, highlighting trending topics, ranking search results, and much more," according to the company.
Part of that "much more" is Facebook using artificial intelligence to combat a problem hyped during the U.S. presidential election: fake news.
In 2015, TechCrunch reports, Facebook released an update to fight hoax stories, which worked by penalizing stories flagged fake by a large number of users.
In 2016, Facebook experimented with using a machine learning algorithm to identify fake news.
Today, AI affects what you see when you browse Facebook, based on your interests and what your network finds interesting.
But as the company teaches AI more about the context behind content, CEO Mark Zuckerberg has discussed the possibility of serving you content based on what AI not your friends thinks you'll like.
Make no mistake...
Facebook is on the cutting-edge of AI research, development, and commercial application. It employs some of the top experts in the field. And there are undoubtedly plenty of well-meaning professionals at the company who are looking to move AI forward.
No discussion of Facebook AI is complete without honestly addressing the very real problems the company's technology has caused.
Problems with Facebook AI
Facebook is under fire for using algorithms (powered by AI) to profit by creating division and sowing misinformation.
The company's entire business model runs on advertising. The only way to get advertisers to pay up is to get users engaging with the platform.
And, it turns out, a great way to maximize engagement among some people is to surface fake news, disinformation, and hate speech.
This is both a problem with Facebook leadership and an AI problem, according to Facebook whistleblower Francis Haugen.
Facebook's AI is optimizing towards the goal of maximum engagement.
Facebook's leadership appears to have let the AI do that without appropriate safeguards to stop problematic content from being surfaced.
Not to mention, the decision to maximize engagement above everything else is a conscious leadership decision.
This is a problem that likely has to be solved by humans, not machines.
While the company uses AI to moderate content, it's clearly not working as well as it needs to in order to avoid issues raised by whistleblowers like Haugen.
What about altering Facebook's algorithms?
The problem is that Facebook's AI is composed of extremely complex algorithms, and it's quite likely no single human knows how they all work. That makes regulating the technology itself difficult.
The "black box" problem is a huge issue with a lot of AI systems. It's not always easy or even possible to tell why a machine makes the decisions it makes.
(For instance, Facebook researchers had to shutdown an AI experiment because the bots in the experiment invented a language the humans didn't understand.)
Instead, it appears the company needs to alter its operational incentives to avoid serious issues from happening in the first place.
What does Facebook AI mean for business leaders?
Facebook AI impacts all of us on a societal level.
It also has a major impact on businesses that use the platform to engage with and advertise to customers.
AI already has a major impact on how Facebook works and how each Facebook user interacts with the platform.
Its News Feed is governed by AI, and could become more dependent on the tech in the near future.
The company's machine learning systems are getting better at understanding text and images.
And the ads you see could lean heavily on the preferences and insights AI reveals to the company's data scientists.
The use of AI across every aspect of the platform is likely to accelerate.
In 2016, Facebook introduced FBLearner Flow, an internal platform that shares machine learning knowledge and code across the company.
At a high level, FBLearner Flow makes it possible to apply algorithms and models from one aspect of the company's operations to others, speeding up machine learning developments.
This-and the company's rapid AI developments-have some significant implications for businesses.
1. Be careful of your brand.
The problems with Facebook spreading misinformation and hate speech are very real.
Until the company gets a real handle on the problem, businesses that advertise on the platform need to realize their brand may show up next to this content.
That can be hard to explain to customers and stakeholders. It can also severely damage your brand.
2. Keep a close eye on Facebook News Feed changes in the next couple years.
From ads to organic content, businesses rely on the News Feed to get their messages in front of consumers.
If AI begins to exclusively determine what users want to see, advertisers may have less insight into how those decisions are made.
The hope would be that super-effective AI would make reaching the right audiences with paid ads even easier.
However, this may not apply to organic content if what your brand shares doesn't beat out the content selected by algorithms.
On Facebook, businesses may need to rely far more heavily on paid targeting than engagement from organic sharing.
3. Take a hard look at your Facebook content's value.
Consumers and Facebook have more control over platform content than ever. AI developments at Facebook accelerate changes at scale that potentially give users more ability to filter out unwanted messages.
Mediocre posts and updates are not going to cut it. Businesses need to take a brutally honest look at how much value they're creating with their content.
Assume right now it's not enough, and plan how to change that.
4. Dig into AI tools that help you better understand return on investment.
Depending on the AI-fueled changes made by Facebook and other tech giants, you might need to fight fire with fire.
Keep tabs on AI tools that analyze campaigns and recommend profitable actions.
These may provide critical guidance on how to generate an adequate return on Facebook and other platforms. It pays to start demoing tools that catch your attention.
Whether you love Facebook or hate it, you need to pay attention to Facebook AI. How the company uses the technology has a real effect on society and business as we know it.