Will 'Vision AI' Be The Next Frontier for Developers? (venturebeat.com) 44
A partner at an early-stage investment firm argues that "in the 2000s everyone was learning HTML and making a website. In the 2010s everyone was learning to develop mobile apps. In the 2020s all the developers are going to build Vision AI."
Where the web had its impact was by digitizing manual paper-based processes... I believe the next big wave is Vision AI, and for the same reason: It offers the opportunity to digitize the next massive trove of information in the world, that which is not on paper but which can be seen through a camera... Why use a temperature sensor when a camera can see reflected light frequencies and determine the temperature? The latest cellphones are integrating LIDAR sensors into their cameras, and I believe the camera sensing suite will become even more sophisticated. Combine this with emerging computer vision technology powered by AI, and together you have Vision AI.
Vision AI has the power to unlock the future of automation in a way not seen since the Web Revolution where every form and phone call was turned into a site, and we unlocked all the resulting searches, analytics, and automated processing that is now commonplace. Just like there are web boot camps, there will soon be computer vision boot camps to enlarge the circle of access to this new technology. Anything you want to count, record, analyze, or store can be obtained by teaching Vision AI to look for it. And that's just capturing the data, the way web forms did. After that unfolds everything we can do with that data. Provide reports, comparisons, and analysis. Make predictions. Profile and advertise. Learn and educate...
The real changes come when computers start measuring and counting things that are either too vast for humans to count — every dead oak tree in California — or too expensive for humans to count — every yeast cell in a culture — or too difficult for humans to perceive — the change in gait that suggests a medical condition.
During this decade we will see boot camps teaching hundreds of thousands of developers to utilize Vision AI tools, just the way we taught millions to code the web. After that, we will see our world for the next level of data that it presents and be able to act on that.
A disclaimer at the end of the article acknowledges that "I currently have a vested interest in eight Vision AI companies."
Vision AI has the power to unlock the future of automation in a way not seen since the Web Revolution where every form and phone call was turned into a site, and we unlocked all the resulting searches, analytics, and automated processing that is now commonplace. Just like there are web boot camps, there will soon be computer vision boot camps to enlarge the circle of access to this new technology. Anything you want to count, record, analyze, or store can be obtained by teaching Vision AI to look for it. And that's just capturing the data, the way web forms did. After that unfolds everything we can do with that data. Provide reports, comparisons, and analysis. Make predictions. Profile and advertise. Learn and educate...
The real changes come when computers start measuring and counting things that are either too vast for humans to count — every dead oak tree in California — or too expensive for humans to count — every yeast cell in a culture — or too difficult for humans to perceive — the change in gait that suggests a medical condition.
During this decade we will see boot camps teaching hundreds of thousands of developers to utilize Vision AI tools, just the way we taught millions to code the web. After that, we will see our world for the next level of data that it presents and be able to act on that.
A disclaimer at the end of the article acknowledges that "I currently have a vested interest in eight Vision AI companies."
Re: (Score:2)
Yeah, I think their examples were pretty bad. But I do tend to agree with the sentiment that this is a big growth field. One example I can't wait to see reach maturity is optical pest and weed control, wherein pests or weeds are looked for on camera, then killed (such as with a laser or targeted blasts of boiling water, pesticide / herbicide, or whatnot). This doesn't just apply to agriculture, either - there's work on laser mosquito killers, which can track mosquitoes from as far as dozens of meters away
Re: (Score:2)
I work in the field by the way of (yuck!) vision AI or as we know it "computer vision" or "machine vision" if I want to buff my grouchy old man credentials.
One example I can't wait to see reach maturity is optical pest and weed control, wherein pests or weeds are looked for on camera, then killed (such as with a laser or targeted blasts of boiling water, pesticide / herbicide, or whatnot).
Well within the bounds of possibility now. I was chatting to a prof I know abut 3 years ago, who had a master's student
Re: (Score:2)
Your rant is both misplaced (replied to the wrong comment, probably in the wrong discussion even) and factually wrong. You refer to "unaccountable groups" but what we call that here in meatspace is "public opinion". Cancel culture is capitalism's response to free speech and the free market. Capitalism cancels anything that threatens the bottom line. The decision to cancel stuff is never being made by the people who are ranting against it, it is always made by some corporate executive whose job it is to dest
Re: (Score:2)
There are a lot of different systems in testing already, one just repeatedly punches the weed under the ground (it dies after two or three repeats), which I found interesting. A lot of the research is happening in European organic farms and in Japan because the cost of field labor is so high there. La Migra leaves farm workers in the US pretty much alone since they don't want to piss off the big agribusiness conglomerates, but field labor is still getting more expensive and there are some tentative deploy
Re: (Score:2)
I hadn't heard about a system that just punches weeds into the ground, but there is logic in that, if the system is run often enough. It'll basically use up its carbohydrate stores repeatedly trying to regrow - so long as it's not given enough time to recover via photosynthesis.
Re: (Score:2)
Ah, here it is. By Bosch, of all companies.
https://spectrum.ieee.org/auto... [ieee.org]
The stamping tool is 1 centimeter wide, and it drives weeds about 3 cm into the soil. It’s designed to detect (through leaf shape) and destroy small weeds that have just sprouted, although for larger weeds, it can hammer them multiple times in a row with a cycle time of under 100 ms. Testing on a real carrot crop, which has carrots spaced about 2 cm apart and an average of 20 weeds per meter growing very close to the carrots
Temperature sensing (Score:2)
Temperature sensing using a regular camera? That would be cool if true, how accurate is that? Good enough to detect fever reliably?
Re: (Score:2, Funny)
Re: (Score:2)
That's where the summary lost any credibility it might have had.
Re: (Score:2)
Re: (Score:2)
Well, the biggie right at the beginning: if you want to measure the temperature of something using a camera, you need to measure *emitted* light, not reflected. There are some very peculiar circumstances in which certain substances change their reflectivity in a meaningful way depending on their temperature, but it's not a common thing.
Secondly, infra-red doesn't mean heat. IR is just a part of the electromagnetic spectrum, like "red" "green" or "radio". IR is conflated with heat because things that are at
Re: (Score:2)
If the phone has a thermal sensor, it can certainly do that. I believe someone has been banging on about a CAT-branded phone which has one.
Of course, that will be largely worthless for detecting Covid, but I presume you have other applications in mind as well.
Nope (Score:3)
No, developers won't be learning Vision AI as a matter of course.
But 90% of the world only has 10% of the money (Score:1)
Re: (Score:2)
Computer vision can potentially replace all the jobs that need human visions. There is a story above this one about how it can spot various serious diseases on an x-ray. Driving down the cost of diagnostics like that will improve healthcare.
A lot of jobs are vulnerable to this. A lot of QC inspection is already done with machine vision as it's relatively easy, just look for major differences between each unit and a known good example. In some manufacturing it is used to sort and orientate objects, the main
Re: (Score:2)
You should check out the advances in recycling technology. It's a horrible and dangerous job for a human that doesn't pay shit, so perfect for robots.
No. (Score:1)
Slashdoomed.
I use vision APIs almost every day. (Score:3)
No (Score:1)
This is nothing more than data gathering and once the code to do it is written and presented as an API its no different to any other form of input and will have limited scope anyway such as trawling images for particular correllations perhaps for medical or security reasons. But most devs probably won't go anywhere near that sort of thing.
"just the way we taught millions to code the web"
Code the web? You mean toss out a dogs dinner of HTML, javascript, CSS and whatever else is web flavour of the month this
Re: (Score:2)
Code the web? You mean toss out a dogs dinner of HTML, javascript, CSS and whatever else is web flavour of the month this week just to present a GUI in a window?
Well, to an awful lot of people, a dog's dinner of HTML, JS and CSS is an awful lot better than not having anything at all due to say a 10x increase in cost not being affordable.
Re: (Score:2)
False economy. Getting a bunch of monkeys to create crap just to get something out the door is never a good long term solution.
Re: (Score:2)
False economy. Getting a bunch of monkeys to create crap just to get something out the door is never a good long term solution.
Is it though? A lot of people don't have very exacting needs. And that money could well get a better ROI spent elsewhere.
Re: (Score:2)
This is nothing more than data gathering and once the code to do it is written and presented as an API its no different to any other form of input and will have limited scope anyway such as trawling images for particular correllations perhaps for medical or security reasons. ....
Well, yes, sort of.
In practice curating good representative training datasets and properly labeling them is an enormous challenge and an enormous amount of work. Also, if you can figure out good outside-the-box ways to do exactly that you can write your own ticket.
How do you mean frontier? (Score:3, Insightful)
This has been around for many years in many different industries, automobile may be a good example.
We call it process automation, years ago this was as simple as having a camera match a product on a conveyor belt.
The challenge imho is in making sense of the data, not the type of sensor used
A very very hard problem, understanding the world, when you see it.
Ian Betteridge (Score:1)
Nobody has quoted Betteridge's not-a-law of headlines yet.
Is this how AI startups successfully stand out from the rest? Stick the word Vision in front of the acronym AI? As if it's somehow magically better? Maybe even visionary? I'm pretty sure the answer is no.
The 2020's AI startups appear to be seeking a new marketing term.
AI / ML is specilist areas (Score:1)
Web pages allowed people who could write to get their writings on the Internet. I think most of those that wanted a website back then just uses Facebook today, or other means of getting their info out, or even CMS.
I remember back in the days, when I developed a CMS for advertising agencies / customers to keep their own webpages updated, that was fun times. When I moved on a former employee took the idea and made it into a CMS that is still out there, but I was likely the first to implement it around here.
Ap
Re: (Score:2)
This. Building a useful AI model -- whether it's an ANN, k-NN, SVM, or whatever else -- is hard. It requires domain expertise and the ability to do research to find good models. Beyond the hardware resources you mention, good AI needs a good algorithm and a good training set. In the case of deep learning, the "algorithm" includes choosing all the metaparameters about depth, layer sizes, and activation functions and the training set includes both labeled truth and a way to generate relevant false or erron
Re: (Score:1)
Building first AI that beats best human in GO is hard. I think we should just leave this to Deepmind.
Using framework (released by the first group) to create AI that can sort cucumbers is not that hard [1]. I think this will be the future of average developers. E.g. "create AI that counts how many cars drive", "create AI that sorts lego bricks".
You can also train AI using human input, like Google does with those traffic light pictures. This is a pretty good short term job for non-educated cheap labour, but i
Vision AI could be since (Score:2)
The only thing that makes me wonder is.
Has Big Tech in conjunction with their Chinese partners already done most of the job. And they are getting ready to do the install now.
AI / ML only as good as its application (Score:3)
Many companies are using AI/ML to improve their offerings, but this really requires both the data and the operational insights into the specific problem it's trying to solve. Computer vision has many applications, we use this in automotive for everything from driver drowsiness detection & alerting to collision avoidance to risk scoring, for example, but it's still only one small part. The examples provided in the post are things that could be done, but the missing part is a real business case for doing this in the first place. CS has never faced a lack of things that could be done with technology, computer vision is not unique in this regard. It's certainly worth having CS students learn the basics of data science and AI/ML tools and methodologies, as it's unlikely to go away in the future, but it would be a mistake to focus only on this. AI startups further face the challenge that they're quite removed from the datasets held by the companies that will ultimately be applying the technology, so they're still going to face a hurdle in going from a proof of concept to actual application. While I can appreciate that someone interested in early-stage investment wants to get in early, I suspect they're chasing the wrong side of the hype cycle for now, just as they did with blockchain.
I read the title as "Weird Al" (Score:2)
Not quite (Score:2)
"in the 2000s everyone was learning HTML and making a website. "
Unfortunately most of them used Flash instead.
What hardware platform? (Score:2)
I took a vision AI/Machine Learning in 2007. My project for class was creating program that allowed you to use hand gestures to control a television. Couple years later took similar class online. It was one of the first tranche of classes offered through coursera. It's been around.
Two challenges need to be addressed before it can take off.
1. API -- no such thing as a generic AI/ML computer vision program. Also, not sure if even many low level libs exist. Lots of work here before the average programmer can g
this nice slashvertisement (Score:3)
Not at all like HTML (Score:1)
Re: (Score:2)
We see this sort of thing a lot in the tech world, 'Wally Wonderful Startup' has a great demo that brings in investor bucks, but fails utterly when applied to the real world. If Wally's lucky one of the big tech companies pays too much for his startup and he can retire to Borabora and sip umbrella drinks all day. In the end the technology gets blamed for being insufficient when more than anything it's an issue with the programming staff not understanding the industry they're supposed to be aiding.
IMNSHO p
Right. ! ? (Score:2)
Vision AI has made great strides. Vision AI is way overhyped.
Re: (Score:2)
Correct on both counts . . .
Lots of positives, but... (Score:2)
Being the pessimistic type, I'm concerned about the negatives. Completely automated surveillance being the main thing. Just wait until LE gets their hands on this, especially given the biases we are training into AI. Pre-crime will be a realistic possibility.
Nope. I predict an imminent backlash... (Score:3)
In a world already saturated with too much information for anyone to be able to cope with, I honestly believe there's going to be a backlash movement against tech at some point.
The volume of data gathering and analysis has entrenched itself into every single part of our lives.
We are monitored, compared, pushed into code and shunted onto graphs - and the outcome is not pretty.
Our privacy has been invaded to such a massive extent, it feels we have lost it almost entirely.
What we have gained seems amazing - the fact that I can type this message and that 100 people can read it, who I don't even know, is astounding.
But it is also alarming - because software is also reading it.
My online profile is being built up with every interaction I make online - Google sometimes knows more about me that I do myself.
I'm not so sure the tradeoff is worth it.
I do wonder at what point people just scream "enough" - this has gone too far.
Well, I think nature will decide for us anyway, in the next few decades, so it's a moot point...