Few things will surprise and shock us as much as the ability of A.I. to alter reality--really!

Staff Writer

What separates the “real” from the “virtual” is obvious, right? Think again.

The applications of AI are infinitely greater in number and in impact than anything we could possibly begin to imagine, and they are accelerating at a rate that we simply have no precedent for. Which is why the benefits of AI will startle us just as much as its downside.

One of those applications, that I recently came across, merges the physical and the digital world in a way that is both fascinating and deeply disturbing by altering the very definition of reality in a way that blurs, if not entirely eliminates, the line between the virtual and the physical.

Some background first.

The Real You

We’ve become obsessed with video. According to Nielsen we watch between 5 to 10 hours of video each day, depending on which generation you belong to. (By the way, boomers actually top the list!).

Much of that is video of talking heads. Nothing is more authentic or revealing than seeing unedited raw video of a person. As humans we’re wild to be drawn to it and to seek out every inflection and micro-expression. Its’ how we establish trust. As a result we’ve come to accept that video of a person is the truest form of media since it cannot be faked without our noticing. Even extraordinary CGI can still be differentiated from the nuances of a “real” person’s gestures and voice patterns.

However, a group of researchers from the University of Washington have recently shown how AI can counterfeit the reality of a talking head in a way that is indistinguishable from the real thing. In fact I’m challenged even trying to find the right words to phrase that last sentence since the whole notion of “real” and “reality” is suddenly put into question.

Using machine learning they trained AI to mimic someone’s lip and facial movements while also changing what they were saying in a way that was indistinguishable from real-life (yes, there’s that word “real” again). This isn’t your run-of-the-Milli-Vanilli lip synch; it’s downright creepy good.

The demo video below shows how this was done with clips from speeches and interviews given by President Obama. The results will blow you away. In an era where video is king this gives new meaning to the much bantered about term “fake news.”

Objects In The Rearview Mirror Are Closer Than The Appear

The rate at which AI can learn and evolve is something we have no scale for.

Imagine a super-ball–the kind that you’d play with as a kid. Let’s say that I’ve just given you one that has the ability to bounce to an unlimited height (we’ll suspend the laws of physics for this). Each bounce will be twice as high as the previous one. If the first bounce is 10 feet off the ground how high will it go on the 10th bounce? It will have topped a small mountain one-mile high.

By the 13th bounce it will have crested the Summit of Everest and be approaching the ceiling of commercial air traffic.

At 20 bounces our ball is approaching low earth orbit, after 28 we’ve passed by Earth’s moon. On the 44th bounce NASA might pick up its faint signal as it whizzes past the first Voyager spacecraft nearly 17 billion miles into deep space. After 88 bounces you’ll never see the ball again since it would now be outside of the visible universe.

Not impressed, right? I expected you might not be, after all, we’ve become somewhat immune to large numbers when it comes to the trajectory of technology. But I need to make a confession. I’m not trying to impress you with how fast a doubling phenomenon can scale. My objective is in how we cannot intuitively perceive geometric or exponential scale.

So, try just one more question.

If we had started with a discount super-ball that bounced only 1 inch, instead of 10 feet,  (In other words the initial bounce was less than one percent of the prior super-ball.) how many more bounces would it have taken us to bounce out of the visible universe?

Incredibly, just seven more bounces than if we’d started with an initial 10 foot bounce! In fact, if we’d started with a ball that bounced a full mile the first time (528 times as much as our original super-ball) our 10ft-bounce ball would take less than 10 bounces to catch up with the mile-high super-ball. The math works, but it’s far from intuitive.

When anything progresses geometrically (and this isn’t even exponential growth) you very quickly get to a stage where the numbers are so large that it almost doesn’t matter where you start.

The same is true of how quickly AI will evolve by learning at rates that have no precedent in the way humans learn. We can debate whether we’re five years or 10 years or even 20 years from reaching a point where AI metaphorically bounces to an altitude that makes it smarter than a single human or all humans, but we are just splitting hairs.

Like our super-ball, AI is already bouncing out of eyesight and quickly making its way into the mainstream of our lives and the way we build our businesses. We can argue how incipient or immature these technologies are today but it makes no difference in the long run.

So, think again about what separates the “real” from the virtual. It’s really not that obvious, is it?