The future of drones and computers understanding pictures

Watched a couple TED talks today and boy ooh boy is our future gonna put the world upside down. It's kind of exciting really, but at the same time, we're headed for a privacy invasion like you've never imagined. Forget the single off the shelf hobby drone we're hearing about today. It's nothing, kind of funny in retrospect how everyone is getting so upset about them. Get a load of what's already coming.

Cooperative drones. Drones so small they fit in the palm of your hand. They work together in a swarm functioning in unison to build things, do maneuvers, scout and map out the inside of buildings etc. Like a team of ROBOTS which is exactly what they are, doing some amazing things. And some of them without the aid of GPS to assist stabilization, just a whole lot of math...to calculate its positioning.

But then there's another technology in the works. Technology to enable computers to "understand" pictures. Not facial recognition, but the ability to actually look at a picture and know exactly what it's looking at. Based on billions of images and the collaboration of something like 40,000 people (I forget) to get all those images and feed them to the computer and have it use AI to determine what the picture is about.

This is cutting edge stuff here folks. We're going to have flying robots not just filming us, but have the processing ability to KNOW what it's looking at. And perhaps even react accordingly. Like RAT YOU OUT. Oh, said the quiet part loud. haha But seriously, some amazing stuff is coming our way. Whether it's used for good or bad, too soon to tell. But I can clearly see an army of police bees patrolling the cities or hallways. I can see a world where cars are now obsolete and we fly to work in autonomous drones with the ability to sense it's surroundings.

I'm 43, hope to be alive to hail a taxi drone one day. Cool stuff.

I imagine there are drones the size of houseflies and smaller out there. If you see a fly that outsmarts a window you know the NSA is in your home lol

Computer image recognition is not new, the post office invested in it a long time ago with the automated reading of post codes and addresses. The quality of the result was better than the average human operator but not as good as your excellent human operators.

To expand this technology into recognition of all kinds of things does take more time, resources and effort. There is a strong and growing understanding of neural networks in the biological, information and hardware environments.

How augmented reality, mechanical integration and communication networks progress does raise many issues. Personally I would prefer to stick to my bike than get a drone taxi.

Well, it should be an exciting future.

I wonder how long it will take for the Olympics to bring in drone shooting as an event.

No more clay pigeons. 12G fun at its best. Just blast them out of the sky!

Rabbit shot would do nicely.

P

We're going to have flying robots not just filming us, but have the processing ability to KNOW what it's looking at.

A difference engine making rote comparisons against whats stored in memory. They don't "know" anything.

a reply to: intrptr

Depends on what your definition of "know" is. Right now it's just comparisons. Later comes conditional decisions based on those comparisons and execute commands. Terminator comes to mind

a reply to: FlySolo

The current IO model is insufficient for more advanced AIs. You end up with so much iterations that you run out of space very fast not speaking about aviable computing power that you need for such calculations. Generating an integrated environmental response chain is one thing, dynamically float it´s parameters and cross validate the outcome with multiple datasets and then embed new rule-sets in near realtime is like stepping on an molehill compared to the mount everest.

You run into so much problems, space(ram&disk), flops, energy, autark data assesement (simply meaning you cannot trust one sensor, you have to compute both datastreams individually and run the same simulations in parallel just to compare both outcomes in the end that are useless most of the time or contradict themselfes or even lead to corrupted datafields (like learning wrong knowledge and base further decisions on that. You have te re-evaluate every single rule-set agains any new. That´s the reason it´s said our brain still outsmarts io-chips. We do that 24/7 in idle mode not being aware of it.

Quantum computing it is. Everything else is just ninjaing around the problem. I´ve been working on several projects (I mentioned some over time here) and you end up restraining the software so much you could have worked out your own algorythms and functions way faster and with much more reliability.

Object oriented programming isn´t going to do the trick with current aviable hardware designs. Forget 1/0 model it´s to slow, to much iterations, to much short-memory-data to be used. We simply run into infinity(!!)

To give you a real world example, imagine you´d have to build a flying bird.
That learns to fly on it´s own and optimize flight patterns to save energy or time (while considering and presimulating environmental parameters). We did it and you end up ordering unique controllers at semiconductor writing your own protocol bus and all that comes entailed with it. You end up writing your own physics engine...

And one day you come back to work, someone is waiting for you at the entrance to give you new access codes and a new project. Your personal belonings from that labor your left 8 hours before unknowingly nicely stacked in two boxes and not a screw left to be seen through the windows. They waited until the last idiot left the building and seized everything overnight.

Last hint: go analog not digital.


Really last hint: You basically simulate every thing and execute the hardware functions and keep track/record of sensor values and compare them to your simulations. I remember every second in freeflight in windchannel needs around two to three minutes of afterprocessing in our cluster and if that finished, another couple of hours re-evaluating every single damn ruleset. The you run simulations with float parameters that can take weeks if give it too much room. A room full of humans will then resume code optimizations or leave it running until you get a half-way stable castrated AI that will give reliable results over a long period of time. Not like flying 10 minutes like a charm only to have one sensor # up and decide do tilt the Z axis and kiss the ground.

a reply to: FlySolo

Depends on what your definition of "know" is.

Computers will never know that they know.

Like you know your name. How do you know your name? You just know.

Computers run programs. The best they can do is execute the next instruction. These are written by programmers. If this, then that… or that.

a reply to: intrptr

I used to be a programmer. IF this then that is called conditional logic. We could be arguing semantics on the definition of "know" My computer knows to changes functions when I do something, regardless if it doesn't know what I want it to do. Future computers may "know" a crime is taking place based on the executable instructions given to it. That's all I'm saying

originally posted by: FlySolo
a reply to: intrptr

I used to be a programmer. IF this then that is called conditional logic. We could be arguing semantics on the definition of "know" My computer knows to changes functions when I do something, regardless if it doesn't know what I want it to do. Future computers may "know" a crime is taking place based on the executable instructions given to it. That's all I'm saying

Your computer doesn't "know" anything.

Then use different terminology. I turn my lights off and on, the light doesn't "know" this.

This is cutting edge stuff here folks. We're going to have flying robots not just filming us, but have the processing ability to KNOW what it's looking at.

Just illuminating the terrain…

a reply to: intrptr

Then use different terminology. I turn my lights off and on, the light doesn't "know" this.

Someone's being nit picky. I can use whatever terminology I want and if that's your only gripe regarding my OP, one little word, then the rant forum is the other direction. My computer knows to shut off because it has a processor. Unless your light is part of smart home technology then you haven't even drawn a comparable comparison. And if it is a smart home light switch, then it KNOWS to shut off when you tell it.

a reply to: FlySolo

The future of drones and computers understanding pictures

Its not nitpicking to call you on proposing that computers "understand" anything.

You "used to be a programmer" and you don't know this?

a reply to: intrptr

You are nitpicking. Clearly, for all intended purposes I use the word "know" to describe the internal workings of a micro processor in terms of functions calling functions. What else would you call it? If you're suggesting I think computers have cognitive abilIties, and I should know better, then you've missed the point and like air, it's gone right over your head

Its not nitpicking to call you on proposing that computers "understand"

And to correct you again. Computers do understand. They understand Java, C++, VB, Cold Fusion, PHP, Pearl. Which are all languages unique to one another. Computers understand instructions. They understand what you tell it to understand. Understand?

a reply to: FlySolo

Well 'computers executing programs' doesn't catch as much "air" as 'computers understanding' things, now does it.

Facial Recognition Software is already out there. You added the computer "knowing" twist to sell a thread.

You may dismiss that as nitpicking, but your thread title was misleading…

I know you know that…

a reply to: intrptr

If my title was misleading, then you better write to TED talks and tell them their title was misleading too. Did you not even watch the video? I think not seeing I'm wasting my time here.