Artificial intelligence: not there yet

@MikeInNH Years ago, I took a programming team to an ACM programming contest. The teams were given a problem to code and one of the criteria was to minimize running time. Our team finished 14th out of more than 100 colleges and universities. I was really proud of our students on the team and it reflected well on my colleagues who taught the courses these students had taken.

There are autonomous vehicles in use today, and there have been for a long time, but they cost a lot. They aren’t automobiles, but air and space vehicles. Drones are largely autonomous. Even piloted aircraft are autonomous except for takeoff and landing. A college friend was an AF navigator on large cargo aircraft. He said the pilots only controlled the plane during takeoff and landing. Otherwise, it was his plane. He input the flight plan into the flight computer, and it did the rest. He checked location and heading periodically. Airliners do the same thing. Spacecraft are really autonomous. The ground team mostly monitors mission operations and intercedes only when something isn’t as it should be. All that costs a tremendous amount of money, though.

If you’re talking about Amazon delivery drones, then yes, mostly, but those are still very new and only operating in limited test environments. If you’re talking about the little camera drones you can buy at Best Buy, then definitely not. Flight stabilization is automated, but the drone pilot is the one telling the drone where to go. Some of them can do rudimentary path following, but the pilot still has to be there to make sure the drone doesn’t screw up, or something doesn’t get into the path. Some have obstacle avoidance, but even the ones that do can hit stuff because it’s not perfect. And if the GPS signal goes down (which isn’t uncommon on drones), then you’d better hope you know how to fly the thing for real or you’re probably gonna crash.

That’s a pretty common misconception, but it isn’t the case. Many of the tasks are automated, but the pilot still needs to tell the computer what tasks to do and when. Even the commercial airliners with very sophisticated flight management systems still need the pilot to tell them it’s OK to, for instance, climb or descend per the plan. Airplane autopilots are there to reduce the tasks a pilot has to perform manually, but they cannot eliminate the pilot during any phase of flight.

And even spacecraft aren’t autonomous in the way that Tesla wants you to think about when you consider their “autopilot.” Ground controllers still tell the spacecraft what to do and when, and the calculations for how to do it are carried out on the ground as well. There are some automated sequences, but those sequences almost always have to be started by a human pushing buttons. Just because many spacecraft do not have humans riding inside of them does not mean that there aren’t a bevy of humans on the ground 24/7 controlling what it does.

Point being, even the very expensive automation systems in planes and spacecraft don’t do what Musk would have you believe the (relatively) cheap so-called “full self-driving” autopilot in a Tesla can do.

And on top of all that, spacecraft and airplanes don’t have to deal with 60 other spacecraft or airplanes flying in formation with them, all being piloted by unpredictable distracted apes with cell phones. Airplanes and spacecraft never have to worry about a little kid chasing a ball from behind a parked car. For the majority of flight time, airplanes and spacecraft have hundreds of feet if not miles separating them from the next vehicle.

In other words, making a 100% autonomous autopilot for a car is worlds harder than for a plane or spaceship, and we haven’t even really done it in aviation or spaceflight yet.

And that’s why I’m so peeved by this “full self driving” nonsense Tesla keeps foisting on the public. It’s bunk, and it’s encouraging idiots to do idiotic things.

1 Like

I wanted to purchase one of these robotic mowers that use AI so that I won’t have to push a mower. Mrs. Triedaq would o.k.the purchase, but she wants a robotic vacuum cleaner.
We are now at a standoff. She has to run the Eureka vacuum cleaner purchased in 1977 and I have to mow with the Toro push mower purchased in 1988.

Much of space flight is simple ballistics, coasting along based on how hard of an initial push the rockets gave, and how the earth’s, moon’s and other’s gravity affects the flight. People and computers get involved for the ‘mid-flight corrections’ and the ‘insertion into orbit burns’. All of which is MUCH simpler than driving across town to the grocery store.


Simple, replace the brushes on the robotic vacuum with blades, you will both be satisfied.

Believe it or not…autonomous air vehicles are a lot less complicated then autonomous ground vehicles.

I was thinking more about military drones.

It’s all determined well before launch and rehearsed many times. Ground control intercedes when perturbations to the orbit require corrections, or telemetry shows a different anomaly. Most of these anomalies are planned for, and ground control tells the satellite to run a program to correct the issue. Sometimes it’s something so unusual that it isn’t planned for, and ground operations has to determine what is wrong, write a program to correct it, send the code to the spacecraft, then execute it.

I do believe it. There aren’t many things that a drone can run into once they get high enough, just other aircraft and birds. AVs also have to recognize the surrounding environment, a very difficult task. Facial recognition is an example of problematic AI. People with lighter complexions are more easily recognized, but dark completed people are often not recognized. It seems to me that shadows are closer to the skin color in dark skin tones, and and this loss of contrast makes facial recognition more difficult.

+The MoBot lawnmower was made in a small plant in Tonawanda NY on Hackett Drive starting in 1969. Some sources say the company was from North Tonawanda NY but I know the plant was in Tonawanda because I used to deliver there, The two cities are in different counties and are divided by The Erie Canal’s present day Western terminus at the Niagara River.

Quite, and that’s why it’s not up to the imaginary Tesla autopilot level. Tesla would have you believe that if it weren’t for the pesky government and their stupid grownup rules, you could get in the car, tell it “go to Blake’s on Menaul Blvd in Albuquerque,” and then crawl in the back seat and take a nap while it drives you to the burger stand. Not only are their cars not actually capable of doing that, but neither is any other vehicle, even the ones that operate in an environment where it would be much easier to do so.

Actually that has been done. HOWEVER…In very very limited testing. It’s called a positive test with very few obstacles in it’s way. If they ran the test to another address it may not work. One positive test means NOTHING. There’s still a lot more work that needs to be done. But there’s progress in the right direction. I still think we’re about 10 years or less away.

Yeah, I was aware of the testing, but I don’t consider that to be “it’s been done” from a practical standpoint. Once those tests turn into routine operations… Well then Teslas still won’t be capable of it because doing it in a ground vehicle navigating amongst hundreds of unpredictable apes is much harder than doing it in a drone that flies 200 feet up or even an airplane that has to deal with other aircraft.

I am not convinced regular intelligence is there.


That reminds me of an old joke. A man wanted his wife to mend his socks and she wanted a pet owl. She wouldn’t give a darn if he didn’t give a hoot.

Yes, the majority of spaceflight requires about as much “steering” as a bullet. The trajectory is determined well in advance.

I got curious and did some quick research. It turns out traffic sign recognition has been around since the 2009 model year. Traffic-sign recognition - Wikipedia
I’d think 12 years would be enough time that an obvious, preventable bug like this one wouldn’t surface but apparently not.

It’s been around, but it hasn’t been foolproof. Sometimes it misses the sign entirely. Other times it misreads it. Not a big deal when all it’s doing is putting up a display that tells you what it thinks the speed limit is, but a much bigger deal when that data is being used by the AI to make driving decisions.

1 Like

Which makes me wonder why the supposedly smart human engineers at Acura/Honda didn’t program the car to compare the sign with GPS data and toggle the display accordingly.

That wouldn’t be foolproof either. My car does the GPS speed thing, but doesn’t read speed limit signs. It still gets the speed limits wrong sometimes. If the local government changes the speed limit after the navigation database is distributed, the car has no way of knowing what the new limit is.

There should be a way for easy updates. This isn’t rocket science. There’s no technical reason the change can’t take effect within minutes of the local speed limit change.

There is a way for easy updates. A few I think are doing it, or at least have Android Auto/etc capabilities to allow your smartphone to be the navigation computer, but most are not. If they switched to automatic near-real-time updates, they couldn’t charge you $100-200 every year for a nav database update.