September 23, 2017 02:41 pm CDT
Two Guys Build a Brain for Fun
By Alex Koyfman, Wealth Daily


Last year, I had the rare opportunity to attend a closed awards ceremony at the Pentagon.

One of the awardees was an old friend of mine from high school — a brilliant electrical engineer who had invented a state-of-the art guidance system for military robots operating in hostile GPS-denied environments such as building interiors and tunnels.

tunnelbot

The Department of the Navy gives these awards out once a year to a handful of the most promising engineers working in their research and development departments.

Of 20,000 such personal, only two-dozen or so receive the honor every year.

For me, just being present and seeing my childhood friend getting the respect of his peers was an honor in itself.

During the hour-long ceremony, I heard about all sorts of new devices and technologies that the Navy's $50 billion/year research budget financed.

Yes, that's $50 billion for the Navy's research branch alone — a sum that trumps all but five of the world's top national defense budgets.

Awards were given out for micro-aerial vehicle advancements, long-range drones deployed by missile or torpedo (or torpedo that turns into missile), a plethora of new guidance systems, and quantum computing.

Things Take a Turn for the Surreal

But the most amazing thing I heard about that day didn't come from the list of inventions created by some of the most brilliant computer and electrical engineers alive today.

It came during a conversation I had with my friend on the drive back to my house, where he was staying for the weekend before flying home to San Diego.

“I've been working on this new project with this Russian guy Mike from work. It's really interesting,” he said as we negotiated the GW Parkway from Northern Virginia back to Chevy Chase, Maryland.

I remember sighing because usually when he starts on these rants, I get bored from his overly technical explanations... flattered that he thinks I have the capacity to understand what he's talking about, of course, but bored nonetheless.

But this time was different. This time, I couldn't help but stay engaged.

“We've been training a neuralnet for the last couple weeks, and it's starting to make some pretty interesting connections.”

“What the hell do you mean, 'training'?”

“It's fairly primitive at this point,” he said without a trace of sarcasm. “We basically just expose it to Google images for a continuous cycle and then ask it questions to see if it's formed any artificial neural pathways based on the image/keyword associations.”

“So you're not programming it with a specific task — you're just asking it to make connections?”

“No, not really make connections. The connections should be a natural side effect of exposing it to the information in the first place. We ask it questions to see if those connections occurred. If we were asking it to make any specific connections, we would be interfering with its innate instinct.”

Innate? Natural? Instinct?

These weren't words I was used to hearing when talking about technology, but this was exactly what my friend was saying — about a neuralnet, or an artificial brain, that he and his friend from work had cobbled together in their spare time.

It was artificial intelligence, but it wasn't making decisions based on specific known parameters programmed into it by software engineers. It was making conclusions based on its own internal concept of logic — something it had to come up with on its own over the course of exposure to countless images.

“It's really nothing too out-there,” he said in his usual California Zen-like tone. “By the time most kids start talking, they've been exposed to the equivalent of trillions of images. We're not doing anything much different. The big change here is that our neuralnet has a perfect visual memory and never forgets any conclusion it may have made in the past. Kids need multiple exposures and reinforcing events to solidify a neural pathway.”

“Ok, so what can it do right now? Say mommy and daddy?”

He chuckled. “Well, for example, it knows a picture of a bridge when it sees one, because it's seen a million pictures of bridges that contain the tag 'bridge'... And it also knows a picture of, say, northwest Washington, D.C. when it sees one, because it's seen a million pictures with that tag.

“And likewise, it knows what daytime looks like and what summertime looks like. But it has never seen an image that contains all those descriptors. So when you ask it to show you a picture of a bridge in northwest D.C. during the day in summer, a regular search engine wouldn't be able to just find that based on keywords — because no such image exists.”

“But your neuralnet can.”

“Well, the reason I mentioned bridges was because that's one feature you can see from the sky anywhere in the world. Same goes for parking lots, airports, etc. So when we asked it to show us a bridge in northwest D.C. during a summer day with no more than six cars on it and at least one pedestrian, it gave us a satellite image of the only one in the world that fit that description.”

“Which one was it?”

“Well, you need to remember that it has to draw data from Google Maps, which doesn't update its photos in real time, and there could have been no bridge that fit that exact description in the photoset available the moment we posed the query, but...”

“What was the name of the bridge, for ****'s sake?”

“The Taft Bridge on Connecticut.”

“Sonofabitch...”

 

taft

Click Image to Enlarge

“Yeah, so it can do that for most anything Google Maps can resolve at this point.

We can find a car parked illegally on a street in Dallas or a swimming pool with an algae infestation in Miami... a police stop that ends in a shooting in Atlanta... I didn't want to run that search but Mike — the other guy — insisted we try.” He paused. “Pretty neat stuff, though.”

Pretty Neat, Indeed

Pretty neat to him was on the verge of science fiction to me... a computer that — just from being exposed to words and images — could apply its knowledge to images that have no keyword descriptors at all across a dataset that was literally the size of the Earth.

It was amazing and scary at the same time... Remember, this wasn't the Navy's $50 billion budget at work. These were just a couple guys in their mid-30s tinkering away on some overpowered desktop computer while their kids attacked each other with Super Soakers in the backyard.

And that, to me, is far more impressive than any supercomputer the Pentagon can finance to identify insurgents from a drone 8,000 miles away.

Also bear in mind that this was almost exactly one year ago... Based on Moore's law, which predicts the increase in processor density (and therefore processing speed), computers have gotten as much as 50% smarter since then.

When weekend tinkerers start messing with self-teaching artificial intelligence that can take the intellect of a four year old and apply it — with no gaps in memory or association — to almost unlimited streams of data, the technological revolution has officially begun.

It's a fear some of us have been harboring for decades, but if the expansion of artificial intellect cannot (or more likely will not) be stopped, I think we need to look at the positive side of things.

On an economic level, this revolution is a new Internet just waiting to happen... a 21st century version of the internal combustion engine, printing press... or gunpowder.

Regardless of its long-term implications, the economic ones will be huge and positive.

My colleague Jason Stutman recently published a groundbreaking report on how to most effectively exploit the coming trends in artificial intelligence and robot design from the ground floor.

It's taken years of collective research to get it all together, and now it's available just in time for a brand-new generation of even smarter consumer electronics to hit the market.

To get a firsthand look at the technological miracles coming your way in the next few years, get instant access to this free presentation right here.

Fortune favors the bold,

alex koyfman Signature

Alex Koyfman

Coming to us from an already impressive career as an independent trader and private investor, Alex's specialty is in the often misunderstood but highly profitable development-stage microcap sector. Focusing on young, aggressive, innovative biotech and technology firms from the U.S. and Canada, Alex has built a track record most Wall Street hedge funders would envy. Alex contributes his thoughts and insights regularly to Wealth Daily. To learn more about Alex, click here.

Bookmark and Share