Hackaday Links: February 17, 2019

February 17, 2019 by Brian Benchoff 43 Comments

There is a population of retrocomputing enthusiasts out there, whose basements, garages, and attics have been taken over by machines of years past. Most of the time, these people concentrate on one make; you’re an Apple guy, or you’re a Commodore guy, or you’re a Ford guy, or you’re a Chevy guy. The weirdos drive around with an MSX in the trunk of an RX7. This is the auction for nobody. NASA’s JPL Lab is getting rid of several tons of computer equipment, all from various manufacturers, and not very ‘vintage’ at all. Check out the list. There are CRT monitors from 2003, which means they’re great monitors that weigh as much as a person. There’s a lot of Sun equipment. If you’ve ever felt like cleaning up a whole bunch of trash for JPL, this is your chance. Grab me one of those sweet CRTs, though.

Last week, we published something on the ‘impossible’ tech behind SpaceX’s new engine. It was reasonably popular — actually significantly popular — and got picked up on Hacker News and one of the Elon-worshiping subreddits. Open that link in one tab. Now, open this link in another. Read along as a computer voice reads Hackaday words, all while soaking up YouTube ad revenue. What is our recourse? Does this constitute copyright infringement? I dunno; we don’t monetize videos on YouTube. Thanks to [MSeifert] for finding this.

Wanna see something funny? Check out the people in the comments below who are angry at a random YouTuber stealing Hackaday content, while they have an ad blocker on.

Teenage Engineering’s OP-1 is back in production. What is it and why does it matter? The OP-1 is a new class of synthesizer and sampler that kinda, sorta looks like an 80s Casio keyboard, but packed to the gills with audio capability. At one point, you could pick one of these up for $800. Now, prices are at about $1300, simply because production stopped for a while (for retooling, we’re guessing) and the rumor mill started spinning. The OP-1 is now back in production with a price tag of $1300. Wait. What? Yes, it’s another case study in marketing: the best way to find where the supply and demand curves cross is to stop production for a while, wait for the used resellers to do their thing, and then start production again with a new price tag that people are willing to pay. This is Galaxy Brain-level business management, people.

What made nerds angry this week? Before we get to that, we’re gonna have to back track a bit. In 2016, Motherboard published a piece that said PC Gaming Is Still Way Too Hard, because you have to build a PC. Those of us in the know realize that building a PC is as simple as buying parts and snapping them together like an expensive Lego set. It’s no big deal. A tech blog, named Motherboard, said building a PC was too hard. It isn’t even a crack at the author of the piece at this point: this is editorial decay.

And here we are today. This week, the Internet reacted to a video from The Verge on how to build a PC. The original video has been taken down, but the reaction videos are still up: here’s a good one, and here’s another. Now, there’s a lot wrong with the Verge video. They suggest using a Swiss army knife for the assembly, hopefully one with a Philips head screwdriver. Philips head screwdrivers still exist, by the way. Dual channel RAM was completely ignored, and way too much thermal compound was applied to the CPU. The cable management was a complete joke. Basically, a dozen people at The Verge don’t know how to build a PC. Are the criticisms of incompetence fair? Is this like saying [Doug DeMuro]’s car reviews are invalid because he can’t build a transmission or engine, from scratch, starting from a block of steel? Ehhh… we’re pretty sure [Doug] can change his own oil, at least. And he knows to use a screwdriver, instead of a Swiss army knife with a Philips head. In any event, here’s how you build a PC.

Hackaday writers to be replaced with AI. Thank you [Tegwyn] for the headline. OpenAI, a Musk and Theil-backed startup, is pitching a machine learning application that is aimed at replacing journalists. There’s a lot to unpack here, but first off: this already exists. There are companies that sell articles to outlets, and these articles are produced by ‘AI’. These articles are mostly in the sports pages. Sports recaps are a great application for ML and natural language processing; the raw data (the sports scores) are already classified, and you’re not looking for Pulitzer material in the sports pages anyway. China has AI news anchors, but Japan has Miku and artificial pop stars. Is this the beginning of the end of journalism as a profession, with all the work being taken over by machine learning algorithms? By vocation, I’m obligated to say no, but I have a different take on it. Humans can write better than AI, and the good ones are nearly as fast. Whether or not the readers care if a story is accurate or well-written is another story entirely. It will be market forces that determine if AI journalists take over, and if you haven’t been paying attention, no one cares if a news story is accurate or well written, only if it caters to their preexisting biases and tickles their confirmation bias.

Of course, you, dear reader, are too smart to be duped by such a simplistic view of media engagement. You’re better than that. You’re better than most people, in fact. You’re smart enough to see that most media is just placating your own ego and capitalizing on confirmation bias. That’s why you, dear reader, are the best audience. Please like, share, and subscribe for more of the best journalism on the planet.

NVIDIA’s A.I. Thinks It Knows What Games Are Supposed Look Like

February 4, 2019 by Drew Littrell 45 Comments

Videogames have always existed in a weird place between high art and cutting-edge technology. Their consumer-facing nature has always forced them to be both eye-catching and affordable, while remaining tasteful enough to sit on retail shelves (both physical and digital). Running in real-time is a necessity, so it’s not as if game creators are able to pre-render the incredibly complex visuals found in feature films. These pieces of software constantly ride the line between exploiting the hardware of the future while supporting the past where their true user base resides. Each pixel formed and every polygon assembled comes at the cost of a finite supply of floating point operations today’s pieces of silicon can deliver. Compromises must be made.

Often one of the first areas in games that fall victim to compromise are environmental model textures. Maintaining a viable framerate is paramount to a game’s playability, and elements of the background can end up getting pushed to “the background”. The resulting look of these environments is somewhat more blurry than what they would have otherwise been if artists were given more time, or more computing resources, to optimize their creations. But what if you could update that ten-year-old game to take advantage of today’s processing capabilities and screen resolutions?

NVIDIA is currently using artificial intelligence to revise textures in many classic videogames to bring them up to spec with today’s monitors. Their neural network is able fundamentally alter how a game looks without any human intervention. Is this a good thing?

Continue reading “NVIDIA’s A.I. Thinks It Knows What Games Are Supposed Look Like” →

AI On Raspberry Pi With The Intel Neural Compute Stick

January 31, 2019 by Sean Boyce 58 Comments

I’ve always been fascinated by AI and machine learning. Google TensorFlow offers tutorials and has been on my ‘to-learn’ list since it was first released, although I always seem to neglect it in favor of the shiniest new embedded platform.

Last July, I took note when Intel released the Neural Compute Stick. It looked like an oversized USB stick, and acted as an accelerator for local AI applications, especially machine vision. I thought it was a pretty neat idea: it allowed me to test out AI applications on embedded systems at a power cost of about 1W. It requires pre-trained models, but there are enough of them available now to do some interesting things.

You can add a few of them in a hub for parallel tasks. Image credit Intel Corporation.

I wasn’t convinced I would get great performance out of it, and forgot about it until last November when they released an improved version. Unambiguously named the ‘Neural Compute Stick 2’ (NCS2), it was reasonably priced and promised a 6-8x performance increase over the last model, so I decided to give it a try to see how well it worked.

I took a few days off work around Christmas to set up Intel’s OpenVino Toolkit on my laptop. The installation script provided by Intel wasn’t particularly user-friendly, but it worked well enough and included several example applications I could use to test performance. I found that face detection was possible with my webcam in near real-time (something like 19 FPS), and pose detection at about 3 FPS. So in accordance with the holiday spirit, it knows when I am sleeping, and knows when I’m awake.

That was promising, but the NCS2 was marketed as allowing AI processing on edge computing devices. I set about installing it on the Raspberry Pi 3 Model B+ and compiling the application samples to see if it worked better than previous methods. This turned out to be more difficult than I expected, and the main goal of this article is to share the process I followed and save some of you a little frustration.

Continue reading “AI On Raspberry Pi With The Intel Neural Compute Stick” →

AI Patent Trolls Now On The Job For Drug Companies

January 30, 2019 by Dan Maloney 39 Comments

Love it or loathe it, the pharmaceutical industry is really good at protecting its intellectual property. Drug companies pour billions into discovering new drugs and bringing them to market, and they do whatever it takes to make sure they have exclusive positions to profit from their innovations for as long a possible. Patent applications are meticulously crafted to keep the competition at bay for as long as possible, which is why it often takes ages for cheaper generic versions of blockbuster medications to hit the market, to the chagrin of patients, insurers, and policymakers alike.

Drug companies now appear poised to benefit from the artificial intelligence revolution to solidify their patent positions even further. New computational methods are being employed to not only plan the synthesis of new drugs, but to also find alternative pathways to the same end product that might present a patent loophole. AI just might change the face of drug development in the near future, and not necessarily for the better.

Continue reading “AI Patent Trolls Now On The Job For Drug Companies” →

Cheating AI Caught Hiding Data Using Steganography

January 3, 2019 by Inderpreet Singh 63 Comments

AI today is like a super fast kid going through school whose teachers need to be smarter than if not as quick. In an astonishing turn of events, a (satelite)image-to-(map)image conversion algorithm was found hiding a cheat-sheet of sorts while generating maps to appear as it if had ‘learned’ do the opposite effectively[PDF].

The CycleGAN is a network that excels at learning how to map image transformations such as converting any old photo into one that looks like a Van Gogh or Picasso. Another example would be to be able to take the image of a horse and add stripes to make it look like a zebra. The CycleGAN once trained can do the reverse as well, such as an example of taking a map and convert it into a satellite image. There are a number of ways this can be very useful but it was in this task that an experiment at Google went wrong.

A mapping system started to perform too well and it was found that the system was not only able to regenerate images from maps but also add details like exhaust vents and skylights that would be impossible to predict from just a map. Upon inspection, it was found that the algorithm had learned to satisfy its learning parameters by hiding the image data into the generated map. This was invisible to the naked eye since the data was in the form of small color changes that would only be detected by a machine. How cool is that?!

This is similar to something called an ‘Adversarial Attack‘ where tiny amounts of hidden data in an image or other data-set will cause an AI to produce erroneous output. Small numbers of pixels could cause an AI to interpret a Panda as a Gibbon or the ocean as an open highway. Fortunately there are strategies to thwart such attacks but nothing is perfect.

You can do a lot with AI, such as reliably detecting objects on a Raspberry Pi, but with Facial Recognition possibly violating privacy some techniques to fool AI might actually come in handy.

Artificial Limbs And Intelligence

December 31, 2018 by Brian McEvoy 1 Comment

Prosthetic arms can range from inarticulate pirate-style hooks to motorized five-digit hands. Control of any of them is difficult and carries a steep learning curve, rarely does their operation measure up to a human arm. Enhancements such as freely rotating wrist might be convenient, but progress in the field has a long way to go. Prosthetics with machine learning hold the promise of a huge step to making them easier to use, and work from Imperial College London and the University of Göttingen has made great progress.

The video below explains itself with a time-trial where a man must move clips from a horizontal bar to a nearby vertical bar. The task requires a pincer grasp and release on the handles, and rotation from the wrist. The old hardware does not perform the two operations simultaneously which seems clunky in comparison to the fluid motion of the learning model. User input to the arm is through electromyography (EMG), so it does not require brain surgery or even skin penetration.

We look forward to seeing this type of control emerging integrated with homemade prosthetics, but we do not expect them to be easy.

Continue reading “Artificial Limbs And Intelligence” →

Artificial Intelligence Composes New Christmas Songs

December 16, 2018 by Lewin Day 20 Comments

One of the most common uses of neural networks is the generation of new content, given certain constraints. A neural network is created, then trained on source content – ideally with as much reference material as possible. Then, the model is asked to generate original content in the same vein. This generally has mixed, but occasionally amusing, results. The team at [Made by AI] had a go at generating Christmas songs using this very technique.

The team decided that the easiest way to train their model would be to use note data from MIDI files. MIDI versions of Christmas songs are readily available and provide a broad base with which to train the model. For a neural network, the team chose to use a Long-short Term Memory (LSTM) architecture. This is a model which is contextually sensitive, which is important when dealing with structured formats like music or language.

The neural network generated five tunes which you can listen to on the Made by AI Soundcloud page. The team notes their time was limited, and we think that with some further work and more adherence to musical concepts such as structure and repetition, it might be possible to generate something a little more catchy.

There are other applications for AI in music, too – like these intelligent musical prostheses.