Wish there was a benchmark for ML safety? Allow us to AILuminate you...
Very much a 1.0 – but it's a solid start
AI + ML
5 Dec 2024 | 8
Nowhere to run to, nowhere to hide, muaha... Boffins build laser-eyed intelligent cam that sorta sees around corners
Artificial intelligence with frickin' lasers beams attached can see objects hidden around corners, according to a study published in the journal Optica on Thursday. Academics, led by boffins at the Stanford University, hope the technology will allow self-driving cars to spot potential hazards, such as pedestrians crossing at …
-
Friday 17th January 2020 07:08 GMT Pascal Monett
"the researchers believe their technique is promising"
Of course they do, they need the funding. I, however, am very doubtful that this can, one day, be used to help self-driving cars navigate the crowded streets of New York. I doubt very much that their laser will even be able to hit the wall and, if it does, it certainly won't be able to reflect anything back to the vehicle's detectors.
Besides, given that you have to get the angle just right, I don't see this working at all outside laboratory conditions.
For this tech to actually work, statistical analysis machines need to progress and miniaturize to the point where each vehicle has its own. I don't see that happening any time soon.
-
Saturday 18th January 2020 08:57 GMT Muscleguy
Re: "the researchers believe their technique is promising"
Since their model assumes the wave nature of light to work then it can be used for any other wavelength such as microwaves or radar or lidar or whatever similar sort of system vehicles are equipped with.
This is a proof of principle which means the deep learning system can be used with non laser wavelengths.
I’m a mere ‘stamp collector’ but as a scientist I can also see this from designing and analysis other experiments. Us Biologists use model systems from slime molds to mammals to gain insights into how our cells, tissues, organs and bodies function using the same proof of principle.
You have to be careful of it though. For eg just because you can get cells in tissue culture (immortalised cell lines are in effect cancer cells) to do something is not a priori evidence that normal cells in bodies do it. So whenever I see such papers I reserve judgement unless and until it is proven to happen in vivo (I’m an organismal biologist so to me in vivo means in an animal, biochemists and molecular biologists have different definitions).
-
Friday 17th January 2020 09:29 GMT Anonymous Coward
How powerful?
Given the light has to reflect off 3 not-intended-to-be-reflective surfaces, how powerful does the laser have to be? I get that the detector could be very sensitive but even that seems tricky. i.e. very sensitive to the laser return but not at all bothered by e.g. sunlight. But there is still going to be several orders of magnitude of loss which suggests that the laser output could well need to be eyesight threatening ...
-
Friday 17th January 2020 11:18 GMT Evil Auditor
Re: How powerful?
Since they hope for use in self-driving cars to detect hazards, I'd suggest to spec up the laser's power to weapons-grade: nothing blocking your way and you can easily get rid of the feckin' [insert hate-car brand of your own disliking] in front of you.
Speaking of weapons-grade, you might need nuclear fission to power the bloody thing. Where's my Ford Nucleon?
-
Friday 17th January 2020 23:45 GMT John Brown (no body)
Re: How powerful?
"Given the light has to reflect off 3 not-intended-to-be-reflective surfaces, how powerful does the laser have to be? I get that the detector could be very sensitive but even that seems tricky. i.e. very sensitive to the laser return but not at all bothered by e.g. sunlight. But there is still going to be several orders of magnitude of loss which suggests that the laser output could well need to be eyesight threatening ..."
Yes, but it's an interesting first step. I remember watching Time Team and the various tools they used such as soil resistivity and ground penetrating radar evolve over the years. In those early stages the results were pretty crap, but useful. Later, better processing with faster CPUs and better algorithms massively improved the speed and quality of the results.
I suspect the hype about AI cars seeing around corners is just that. Hype. They have something new and clever and AI based self-driving cars is a good and current hook to hang it on for the required funding.
-
-
Friday 17th January 2020 15:25 GMT ThatOne
Cars, really?
Of all the possible use cases, they chose cars?
So, to see if a pedestrian (or car) is about to jump out from behind an obstacle, you need your car to be always preceded by a mobile wall (because the necessary reflecting surface will be missing 99.9% of the time), and your hidden pedestrians or cars have to never exceed a centimeter of size, else the system won't be able to identify them (could be a discarded soda can lying there).
Except those minor caveats, it's all good.
-
Friday 17th January 2020 16:31 GMT badgeror
Use for self driving car? yeah sure...
Just a reminder for these researchers : you're never going to get the SNR needed in outdoor, daylight scenarios, for self driving car unless you want to blind any human/animal/camera in sight
But I guess there are some military out there who whould like this tech a lot for counter-terrorism units and special ops
-
Friday 17th January 2020 19:07 GMT Il'Geller
There is a clear and unambiguous definition of Artificial Intelligence:
AI searches for, finds, and uses answers in the form of structured texts; based on their
- explicit contexts that can be read,
- and implicit subtexts that are implied as dictionary definitions and allusions to other texts; where an allusion is an expression designed to call something to mind without mentioning it explicitly; an indirect or passing reference.
Thus, everything is reduced to finding answers to the questions asked by computer.
-
Biting the hand that feeds IT
