AI Wins DARPA Fighter Pilot Simulation

Jeff Brown
|
Aug 27, 2020
|
Bleeding Edge
|
8 min read
  • This AI just won against an F-16 fighter pilot
  • We’re getting closer and closer to Tesla’s masterstroke
  • Roombas are about to get a lot smarter…

Dear Reader,

Days ago, research was published in Spain concerning the transmission of COVID-19 among school-aged kids. This is the largest and most meaningful study to date on this subject.

For five weeks, the study tracked more than 1,900 kids and staff in 22 summer camps in Barcelona, Spain. Every week, saliva samples were taken and analyzed for COVID-19.

What they found may surprise you…

The R0 (R naught), or basic reproductive rate of a virus, was a mere 0.3. For comparison, the R0 of COVID-19 in adults is currently thought to be in the 2–3 range. A common cold is in the range of 2–3, and influenza tends to be in the 1.5 range.

So what does this all mean?

It confirms what has already been shown. Children are not effective at spreading COVID-19. Not only that, but this study demonstrates school-aged kids are six times less likely to transmit COVID-19 than the general adult population.

In short, there is not a damn bit of science that justifies forcing kids to wear face masks.

Kids Wearing Masks in a School in Vietnam

Source: Vox

This research tells us the opposite. School-aged kids are more likely to become ill from seasonal influenza than from COVID-19.

And sadly, more kids have died from suicides resulting from the lockdowns and school closures than from COVID-19. Some kids in China have even dropped dead in physical education class while being forced to wear masks while running.

None of this makes sense.

The data and research are clear.

How can we allow school administrators and state governments to permit the closure of schools? Will our tuition payments be refunded? Will our property taxes be reduced?

And how about the health of our children and grandchildren? We thought that was the whole purpose of school.

Any adults who have underlying conditions can work remotely or mask up with a properly fitted N95 and sealed eyewear and keep their distance.

Any parents who aren’t convinced can always “opt-out,” keep their kids at home, and homeschool.

It is time to put the kids’ best interests and health first.

Now let’s turn to our insights…

AIs versus human fighter pilots in the AlphaDogfight challenge…

The Defense Advanced Research Projects Agency (DARPA) just hosted what it called the AlphaDogfight challenge. DARPA is the government agency responsible for the development of emerging technologies. And this recent challenge shows a key area of interest…

For the contest, DARPA invited companies to build their own artificial intelligence (AI) to compete in a fighter pilot simulation.

Last August, DARPA selected eight companies to compete in the final challenge. The list included Lockheed Martin, representing the legacy defense contractor industry, as well as several smaller companies focused on AI systems.

For the challenge, the AIs that these companies built were pitted against each other as well as a top-notch human F-16 fighter pilot in the United States Air Force. And the results were definitive…

AlphaDogfight in Action

Source: Defense One

The winner was a company called Heron Systems. Heron’s AI beat all seven of its AI competitors. And it absolutely destroyed the human fighter pilot, winning five rounds to zero.

A big part of the AI’s success was its ability to adapt on the fly. The human fighter pilot said that by round five, all the normal tactics he employed weren’t working against the AI. It anticipated and countered his every move. So he was forced to shift his tactics, but he couldn’t adapt fast enough.

This ability to anticipate and adapt to events in real time is a testament to the training model Heron Systems used.

Heron trained its AI using what’s called deep reinforcement learning. This method starts from the ground up. It assumes that the AI has no knowledge whatsoever about how to fly. So Heron’s AI was given no guidelines to start with – it was just told to fly the aircraft.

What’s interesting about this approach is that it leads to incredible failure at first.

A human pilot knows that it’s a bad idea to fly the plane into the ground or other objects. But starting from nothing, Heron’s AI didn’t know this.

As a result, it crashed the plane a lot at first. It had to learn that flying into things led to failure. And learn it did…

Heron’s AI ultimately went through four billion simulations, learning from each one. This gave the AI at least 12 years’ worth of flight experience in a condensed time as well as an ability to adapt on the fly. And that’s why it was able to definitively win the AlphaDogfight challenge.

To me, this is a clear lens on the future.

There’s simply no way a human could go through four billion simulations. An AI can be trained to adapt under any circumstances imaginable.

So the trend is clear. In the future, AI will augment humans in the field or replace us – even in complex tasks like flying and fighting in an F-16.

Of course, this also opens up a lot of questions…

We have to wonder, what if this technology fell into the wrong hands? Imagine a rogue actor being able to best the world’s top fighter pilots with a single AI. It sounds like a plot out of a Tom Clancy novel. I wish he were still around… He would have certainly gotten a kick out of developments like this.

Using advanced technology responsibly will be one of humanity’s most important challenges in the years ahead.

Tesla’s next big upgrade is in the works…

Tesla is working on version 4.0 of its self-driving semiconductor. We can think of this as the “brains” of Tesla’s Autopilot mode.

And that means another big upgrade is on the way. Tesla’s masterstroke is getting closer…

Tesla made headlines last year with its version 3.0 semiconductor, which was the first version based on Tesla’s in-house design. The first two versions were based on NVIDIA’s design, but Tesla wanted to move its tech forward at an even faster rate, so it made the switch.

And the move paid off in a big way. Tesla’s version 3 hardware was able to process information at a whopping 21 times faster than its previous version. That was a big step toward making fully autonomous Teslas possible.

We don’t yet know what kind of performance improvement the V4 hardware will provide, but we’ll probably get a look at that pretty soon. The chip is going into production next quarter, so I suspect testing results will be available by early next year.

From there, V4 is set to go into mass production in the fourth quarter of next year. That’s when it will start going into the new Teslas hitting the market.

Tesla is partnering with semiconductor manufacturer TSMC to produce the V4 chips using 7 nanometer (nm) semiconductor manufacturing processes. This will improve the chip’s power efficiency, which is critical for electric vehicles (EVs).

Why is power efficiency so crucial?

When a Tesla is operating on Autopilot, it is like a supercomputer. Naturally, that requires a lot of power, which uses up electricity that would otherwise be used for the car’s range.

That’s why using a 7 nm chip is so important – it will reduce the power consumed by Autopilot mode.

This is such a big year for Tesla. It is manufacturing cars at its new plant outside of Shanghai, and it has broken ground at the new plant in Germany. It also announced it will build a new plant in Austin.

We’ll also have a new version of Autopilot before the end of the year, and we can expect some exciting news about its battery technology in September.

And the best part? The company is throwing off free cash flow now.

And we are getting closer and closer to Tesla’s masterstroke – a self-driving ride-hailing network. There is a chance that we might see that happen by the end of next year.

I suspect 99% of people will be shocked when they wake up one day and learn that they can now hail a ride in a self-driving Tesla. But Bleeding Edge readers will know better.

Tesla’s “Uber” is getting closer by the day.

The most interesting thing to come from iRobot in a decade…

iRobot just caught my eye for the first time in over 10 years. The company announced that it is migrating to over-the-air software updates for its robotic systems. iRobot is taking a page out of Tesla’s playbook with this.

iRobot is the company behind the Roomba robot vacuum cleaner. This is the disc-shaped vacuum cleaner that will clean the floor by itself.

A Roomba Vacuum

Source: iRobot

I have used a Roomba for about 10 years now. I’m sure many readers have too. It’s a simple design. There’s one button on the top of the Roomba that you press for it start to cleaning. And there are smaller buttons where you can schedule the Roomba to clean at a certain time every day. That’s all there is to it.

The trouble here is that the Roomba isn’t very smart. It gets stuck a lot. And it doesn’t automatically go back to the charging dock when its battery is low. Instead, it will run its battery dry and just die on the spot. Then you have to manually pick it up and put it back on the charger.

And that’s why iRobot’s new announcement is so interesting. By becoming software centered, it can upgrade, improve, and innovate throughout the year. Roombas can become “smart.”

Among the first announced features is the ability to control the Roomba from a smartphone app.

We will be able to instruct it to clean only a certain room or area of a room rather than doing everything. The Roomba will also be able to suggest cleaning schedules based on use patterns, such as an “after dinner” routine around the kitchen table.

What’s more, iRobot will push out upgrades to all Roombas several times each year. This is exactly what Tesla does with its cars anytime it has a software improvement to roll out.

The beauty of this model is that consumers don’t have to take the hardware back to the manufacturer every time there is an upgrade. And our Roomba won’t need cables to connect to our computers for the upgrade to download.

From the consumer’s perspective, the upgrades happen automatically.

So this is a big shift for iRobot. It is moving from a hardware model to a software and AI-driven model. That’s a great move.

And this is another lens on the future…

We are moving into a world where robots will serve useful functions in the average consumer’s home. These robots will have elements of intelligence that will make them far more effective than they have been up to this point. And robots will seamlessly integrate into a smart home environment.

The world of The Jetsons is much closer than many people realize.

Regards,

Jeff Brown
Editor, The Bleeding Edge


Like what you’re reading? Send your thoughts to feedback@brownstoneresearch.com.


Want more stories like this one?

The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.