Using AI to predict the Oscars (and maybe even save humanity)

A decade ago, I wrote a graphic novel about a researcher who used AI technologies to connect vast populations of humans into a unified superintelligence that could solve all the world’s problems. I titled the book Monkey Room because it plays on the old concept that if you put a million monkeys into a room with a million typewriters and have them pound the keyboards for millions upon millions of years, they’ll eventually produce the complete works of Shakespeare by dumb luck.

In the book, we humans are the monkeys. And the AI that watches all of humanity’s actions and reactions is the room that we foolishly traps ourselves within. 

I wrote the book as a cautionary tale, warning that the human race could be turned into a mindless factory for generating random thoughts (content) that is vacuumed up by a super-intelligent AI to emulate human thinking while lacking any human values, morals, emotions or sensibilities. 

Keeping human qualities “in the loop”

Now, more than a decade later, I can’t help but wonder if ChatGPT, LaMDA and other Large Language Models (LLMs) are the first dystopian steps towards building a real “monkey room” that will reduce humanity to a source of fluctuating data points that an amoral super-intelligence uses to do “the real thinking” for us.    

Is ChatGPT a step toward a dystopian Monkey Room scenario?

Maybe I’m overstating the danger, but these risks seemed so clear to me back in 2014 that I founded a company called Unanimous AI that has pursued the opposite mission: To use AI to connect people in ways that amplify and elevate our collective intelligence while preserving our human values, morals and sensibilities. This mission stands in stark contrast to so many current AI efforts that push to automate decision-making in ways that treat us humans as mere data points, taking our most human qualities out of the loop.  

How AI can be used to amplify human intelligence rather than replace it

Like many researchers, I looked to mother nature for inspiration and began studying how biological systems enable large populations to amplify their intelligence. 

It turns out that evolution has been wrestling with these problems for hundreds of millions of years and has solved it many times, enabling a wide range of organisms (from schools of fish to swarms of bees) to “think together” in ways that make the population significantly smarter than the individual members. 

Biologists call this swarm intelligence, and it works very differently from how we humans usually make group decisions.

Instead of taking polls or conducting votes or building a hierarchy with a “decision-maker” at the top, mother nature creates real-time systems in which all members can push and pull on the group in a giant multi-dimensional tug of war. This allows them to converge together on solutions that are almost always smarter than individuals would have come up with on their own. 

Bees, for example, can make decisions by vibrating their bodies in unison, reacting to each other in a process called “waggle dancing,” and it’s been shown to converge on optimal solutions to complex multivariable problems. 

“Hive mind” is not a pejorative

This is where the phrase “hive mind” comes from, but the pejorative context is totally misplaced.  In reality, we have a lot to learn from flocking birds, schooling fish and swarming bees because they can make remarkably skilled decisions without forming a “herd mentality” where one individual gets spooked and runs off a cliff and everyone else follows. 

Herds are asynchronous structures where the impulses of a few initial actors attract the many to follow suit. Swarms are synchronous structures where all members interact in real-time, pushing and pulling off each other in a system that deliberates and efficiently finds optimal solutions. 

Now think about social media, where a single tweet can kick off a single “like,” which in turn can kick off a cascade of “likes” — talk about a herd mentality.  The process is called “social influence bias,” and it’s part of the reason we humans have collectively been making such bad decisions over the last decade.  We’ve built a technological infrastructure that amplifies noise the same way a single sheep who sees a shadow and gets spooked for no reason can lead hundreds of others in a stampede to nowhere. 

The “snowballing” effect

For example, a 2013 research study by Hebrew University and MIT showed that a single upvote on a piece of content can increase the likelihood of the next upvote by 32% , and increases the chances that the content is positively rated overall — after thousands and thousands of votes — by 25%. 

This is called “snowballing,” and it’s basically us humans jumping off a cliff. And guess what? Now we’re feeding the most liked and shared content into AI systems that use them to represent humanity.  Sound like a good idea to you? Not to me — that’s why I believe we humans need to learn from mother nature and shift our online interaction model from herding to swarming. It makes groups smarter. 

Of course, back in 2014 when I began working to build a system, I hit up against a very serious problem: We humans didn’t evolve the ability to form real-time synchronous systems the way birds and bees and fish do.

So, I began developing a technology called artificial swarm intelligence (ASI) that I thought might allow networked human groups to think together in intelligent swarms. As I assembled a team of engineers and researchers, we had no idea if it would work, but we took comfort in the fact that mother nature usually points us in the right direction.  And guess what? She did.  

Combining thoughts and insights in real time

It turns out that artificial swarms really work, enabling networked human groups to combine their thoughts and insights in real-time, producing better decisions and smarter forecasts and more accurate medical diagnoses and business evaluations. It has even been shown to boost IQs. (For more details, check out the TEDx talk I gave in 2017 to explain the underlying science while providing examples validated in university studies.)

Of course, saying a new technology works or pointing to academic papers that prove it works is not as much fun as testing the concept on high-profile events where anything can go wrong. At Unanimous, we’ve done this many times in the past, using human groups and Swarm AI to predict a wide range of events from the Kentucky Derby and the Super Bowl to the 2020 election — and with great success.

Which brings me to the 2023 Academy Awards airing live this Sunday. 

For the seventh year in a row, our researchers at Unanimous AI have invited a group of randomly selected “movie enthusiasts” to participate online as real-time swarm intelligence and predict all major categories of the Oscars. If things go the way they have in the past, this group of just 20 amateurs will match or outperform most professional movie critics. 

Again, this is not a vote or a poll. These 20 individuals formed a real-time system mediated by swarm intelligence algorithms that helped them converge on the best combination of their individual insights and intuitions.  Each forecast is performed in about 60 seconds and looks something like this: 

The process of predicting the Oscars took about 30 minutes and was conducted entirely online. It produced the set of results shown in the table below. As you can see, the swarming method outputs not just a prediction for each award but a probabilistic confidence. 

As listed, the most likely movies to win Oscars include All Quiet on the Western Front, which is predicted to win Best International Film, and Guillermo De Toro’s Pinocchio, which is predicted to win Best Animated Feature Film. And finally, Everything Everywhere All At Once is predicted to be the big winner overall on Sunday night. 

Will all the predictions above be correct? Probably not, but if the 2023 results are similar to previous years, we can expect artificial swarm intelligence to produce a set of forecasts that land between 81% and 93% accurate when results are announced. 

Of course, using artificial swarms to amplify the intelligence of human groups is useful for far more important things than predicting the Oscars. 

For example, the United Nations has used artificial swarm intelligence to help forecast famines in hot spots around the globe, while other groups are exploring the use of swarms to facilitate negotiations among entrenched parties with adverse interests.

Personally, my hope is that all researchers working in AI push harder to keep humans in the loop, amplifying our wisdom and insights rather than reducing us to data points or replacing us with algorithms. 

Louis Rosenberg is a pioneering researcher, inventor and entrepreneur in the fields of virtual reality, augmented reality, and artificial intelligence. He is the founder of Immersion Corporation (IMMR: Nasdaq), Microscribe 3D, Outland Research, and Unanimous AI.