Tag Archives: neural networks

Augmented evidence. It’s a logical trajectory.

A few weeks ago I gushed about how my students killed it at a recent guerrilla future enactment on a ubiquitous Augmented Reality (AR) future. Shortly after that, Mark Zuckerberg announced the Facebook AR platform. The AR uses the camera on your smartphone, and according to a recent WIRED article, transforms your smartphone into an AR engine.

Unfortunately, as we all know, (and so does Zuck), the smartphone isn’t currently much of an engine. AR requires a lot of processing, and so does the AI that allows it to recognize the real world so it can layer additional information on top of it. That’s why Facebook (and others), are building their own neural network chips so that the platform doesn’t have to run to the Cloud to access the processing required for Artificial Intelligence (AI). That will inevitably happen which will make the smartphone experience more seamless, but that’s just part the challenge for Facebook.

If you add to that the idea that we become even more dependent on looking at our phones while we are walking or worse, driving, (think Pokemon GO), then this latest announcement is, at best, foreshadowing.

As the WIRED article continues, tech writer Brian Barrett talked to Blair MacIntyre, from Georgia Tech who says,

“The phone has generally sucked for AR because holding it up and looking through it is tiring, awkward, inconvenient, and socially unacceptable,” says MacIntyre. Adding more of it doesn’t solve those issues. It exacerbates them. (The exception might be the social acceptability part; as MacIntyre notes, selfies were awkward until they weren’t.)”

That last part is an especially interesting point. I’ll have to come back to that in another post.

My students did considerable research on exactly this kind of early infancy that technologies undergo on their road to ubiquity. In another WIRED article, even Zuckerberg admitted,

“We all know where we want this to get eventually,” said Zuckerberg in his keynote. “We want glasses, or eventually contact lenses, that look and feel normal, but that let us overlay all kinds of information and digital objects on top of the real world.”

So there you have it. Glasses are the end game, but as my students agreed, contact lenses not so much. Think about it. If you didn’t have to stick a contact lens in your eyeball, you wouldn’t and the idea that they could become ubiquitous (even if you solved the problem of computing inside a wafer thin lens and the myriad of problems with heat, and in-eye-time), they are much farther away, if ever.

Student design team from Ohio State’s Collaborative Studio.

This is why I find my student’s solution so much more elegant and a far more logical trajectory. According to Barrett,

“The optimistic timeline for that sort of tech, though, stretches out to five or 10 years. In the meantime, then, an imperfect solution takes the stage.”

My students locked it down to seven years.

Finally, Zuckerberg made this statement:

“Augmented reality is going to help us mix the digital and physical in all new ways,” said Zuckerberg at F8. “And that’s going to make our physical reality better.”

Except that Zuck’s version of better and mine or yours may not be the same. Exactly what is wrong with reality anyway?

If you want to see the full-blown presentation of what my students produced, you can view it at aughumana.net.

Note: Currently the AugHumana experience is superior on Google Chrome.  If you are a Safari or Firefox purest, you may have to wait for the page to load (up to 2 minutes). We’re working on this. So, just use Chrome this time. We hope to have it fixed soon.

Bookmark and Share

The end of code.

 

This week WIRED Magazine released their June issue announcing the end of code. That would mean that the ability to write code, as is so cherished in the job world right now, is on the way out. They attribute this tectonic shift to Artificial Intelligence, machine learning, neural networks and the like. In the future (which is taking place now) we won’t have to write code to tell computers what to do, we will just have to teach them. I have been over this before through a number of previous writings. An example: Facebook uses a form of machine learning by collecting data from millions of pictures that are posted on the social network. When someone loads a group photo and identifies the people in the shot, Facebook’s AI remembers it by logging the prime coordinates on a human face and attributing them to that name (aka facial recognition). If the same coordinates show up again in another post, Facebook identifies it as you. People load the data (on a massive scale), and the machine learns. By naming the person or persons in the photo, you have taught the machine.

The WIRED article makes some interesting connections about the evolution of our thinking concerning the mind, about learning, and how we have taken a circular route in our reasoning. In essence, the mind was once considered a black box; there was no way to figure it out, but you could condition responses, a la Pavlov’s Dog. That logic changes with cognitive science which is the idea that the brain is more like a computer. The computing analogy caught on, and researchers began to see the whole idea of thought, memory, and thinking as stuff you could code, or hack, just like a computer. Indeed, it is this reasoning that has led to the notion that DNA is, in fact, codable, hence splicing through Crispr. If it’s all just code, we can make anything. That was the thinking. Now there is machine learning and neural networks. You still code, but only to set up the structure by which the “thing” learns, but after that, it’s on its own. The result is fractal and not always predictable. You can’t go back in and hack the way it is learning because it has started to generate a private math—and we can’t make sense of it. In other words, it is a black box. We have, in effect, stymied ourselves.

There is an upside. To train a computer you used to have to learn how to code. Now you just teach it by showing or giving it repetitive information, something anyone can do, though, at this point, some do it better than others.

Always the troubleshooter, I wonder what happens when we—mystified at a “conclusion” or decision arrived at by the machine—can’t figure out how to make it stop arriving at that conclusion. You can do the math.

Do we just turn it off?

Bookmark and Share