We look at some of the strangest applications of artificial intelligence. Read next: Hiroshi Ishiguro Wants A Humanoid Future
Judging the looks of CGI people
Hot or Not for fake people has finally arrived. Visitors to Judge Fake People can rate the attractiveness of thousands of faces taken from This Person Does Not Exist, which generates fake portraits by mimicking details found in real photos. “I just wanted to turn it into hot or not,” Judge Fake People creator Mike Solomon explained on his website. “So I wrote a script to download an image from thispersondoesnotexist.com every 5 seconds and built up a collection of around two thousand fake people. Then I made a voting system with php/MySQL and some filters to show the highest and lowest rated faces. And I enabled comments just for fun.” Let us know if you find your face among them. Lonely hearts looking for something more serious than a computer-generated photo may have better luck in one of the robot brothels popping up across Europe. Those that prefer a more human touch could instead swing by Sheri’s Ranch in Nevada for a tête-à-tête with Sybil Stallone, who has spent $500,000 on plastic surgery to become a cyborg prostitute she calls “the sex terminator”.
Coaching non-league football clubs
Data science may be gaining a growing influence on the work of Premier League coaches, but seven tiers below in non-league football it’s already replacing them. After a run of bad results left them fighting against relegation from the Isthmian League Premier Division, Wingate and Finchley FC recruited an AI coach for tactical advice. It made its debut in a crucial February clash against fellow relegation candidates Whitehawk FC. The AI coach studied data on Wingate and Finchley’s players and their opponents, which led it to recommend a 4-3-3 formation. It then took its place in the dugout before the match began. So who would triumph in the battle between human and machine? The result, a 1-1 draw. “Over the last few weeks the AI has been a really useful aid for the coaches at the club,” said interim manager Dave Norman. “With today’s draw, it can now claim to be an unbeaten coach.” Read next: How data analytics is transforming sports on and off the pitch
Crafting pick-up lines
Budding Don Juans and Doña Juanas can turn their tongues from tied to silver with the help of an AI wingman. Research scientist Janelle Shane trained a neural network in the art of seduction by feeding it a dataset of existing pick-up lines that it then analysed for patterns. It returned with lines that ranged from the sweet “I want to get my heart with you” to the surreal “You must be a tringle? Cause you’re the only thing here.”
Creating inspirational quotes
If you’re struggling to find meaning in an imperfect world try listening to InspiroBo, an AI built to generate endless inspirational quotes. At the click of a button, the web application will plaster a new motivational quote on a pretty picture to help a lost soul find its place. Sadly the sage is not immune from its own existential crises, one of which led it to tell Techworld that “the fact that you are ugly, doesn’t necessarily mean that you’re educated.”
Creating TV news anchors
China’s state-run Xinhua News Agency recently welcomed a new employee to the newsroom: the world’s first AI news anchor. The digital figure is modelled on the agency’s human presenter Zhang Zhao, and learns by studying video of the voices and features of human broadcasters. Critics have argued that the anchorman is not a true example of AI as he merely reads a script, but his defenders could claim the same of human anchors reading from a teleprompter.
Scripting car adverts
Lexus claims to have created the world’s first car commercial written by AI. IBM’s Watson AI scripted the 60-second ad after analysing audio, text and visual data from the past 15 years of Cannes Lions award-winning automotive adverts. Oscar-winning director Kevin Macdonald then brought the story to life. The ad is indistinguishable from any car commercial created by a human, little surprise given the notoriously formulaic training data that Watson would have studied.
Creating perfumes
Artificial olfaction has entered the ancient art of perfumery through a partnership between German fragrance house Symrise and IBM Research. The two have created a new method of developing perfumes by analysing existing fragrance formulas alongside historic sales data to understand which type of people were buying each scent. The data is then used to create new concoctions that target specific demographics. Symrise aims to be selling two of the AI-designed fragrances in 2019. “The art and science of designing a winning perfume has been something we at Symrise have been doing for more than 200 years,” said Achim Daub, president of Symrise Scent & Care. “Now our perfumers can work with an AI apprentice by their side, that can analyse thousands of formulas and historical data to identify patterns and predict novel combinations, helping to make them more productive, and accelerate the design process by guiding them toward formulas that have never been seen before.”
Choosing Halloween costumes
Optics research scientist Janelle Shane has developed an algorithm that can pick novel Halloween costumes. Shane trained machine-learning algorithm called textgenrnn that can learn to imitate text on a list of 7,182 costumes. The algorithm studied the examples of costumes to learn how to spell and combine all the words that it needed to design its own outfits. Each round of training taught it new vocabulary, which allowed it to produce increasingly original costumes such as ‘Ballerina Trump’, and ‘Sexy Minecraft Person’.
Telling your fortune
MIT-trained roboticist Alexander Reben built a system that generates one-line predictions by training a neural network on the messages found in thousands of fortune cookies. He calls the results “artificial philosophy” The maxims it produced were surprisingly dark. Reben estimates that around 75 percent of them were very negative messages while others were just weird. His favourites include “no one is listening” and “the first man gets the oyster, the second mouse gets the cheese.” “People think that this sort of computer science is very predictable, mathematical and cut-and-dry, but when you have such a large data set and complicated algorithms, you can get very unexpected outputs,” Reben told the Washington Post. “These fortunes also have a beauty and humor that is all of their own, and there’s this inherent creativity to a lot of these algorithms, as well.”
Making pizza
MIT students and researchers have built an AI system that generates new pizza recipes. The team trained a recurrent neural network on hundreds of artisan recipes and cooked its creations in a 900-degree wood-fired brick oven. Some of the recipes lacked vital ingredients or included ones that that didn’t exist, so the team chose to recruit some human assistance. Tony Naser, the owner of Crush Pizza in Boston, stepped up the challenge, adding a sauce to one recipe and removing “wale walnut ranch dressing” from another. The results ranged from a delicious blend of sweet potato, beans and brie, to a more divisive concoction of shrimp, jam and Italian sausage. It was up to Naser to deliver the final verdict. “I think it’s pretty good because maybe it’ll put together a combination that you won’t think of,” he said. “But then you tweak it from there.”
Finding Waldo
Marketing agency Redpepper has built a robot to find Waldo, known as Wally in the UK. Researchers trained Google’s AutoML Vision to identify the elusive illustrated character by feeding it images of the peppermint-striped traveller. A robotic arm equipped with a camera then takes a photo of the pages of the puzzle book and compares it to the model. When it finds a match, it slams a plastic hand on Waldo. There’s Waldo is still only a prototype but already it can find the bobble-hatted figure faster than most five-year-olds.
Learning to play basketball
AI is now teaching animated basketball players how to dribble. Researchers from Carnegie Mellon University and DeepMotion Inc used deep reinforcement learning to copy the skills of human basketball players caught on motion capture data. The programme had to firstly master locomotion and then learned how to use its arm to control the movements of the ball. After millions of trials, the motions of their arms movements had become so closely coordinated with the bounce of the ball that they could dribble between their legs and behind their backs like a pro.
Flipping burgers
Flippy the burger-flipping robot can grill150-300 patties per hour. It monitors the patties as they cook, switches spatulas when required and even cleans the grill. Manufacturer Miso Robotics calls Flippy “the world’s first autonomous robotic kitchen assistant that can learn from its surroundings and acquire new skills over time.” It may learn its job too well. Flippy had barely begun its career before it was forced to retire, as its human colleagues couldn’t assemble the burgers as fast as Flippy could grill them. The result represents an early victory for human workers over the machines that want their jobs.
Judging beauty contests
Beauty.AI is trying to prove that beauty is not as much in the eye of the beholder as it is in the data by launching the first beauty contest judged by artificial intelligence. Humans take a selfie and submit it for robots to evaluate their wrinkles, face symmetry, feature proportions, skin health and a range of other parameters. The judges compare the submissions and then pick their beauty queen and ing. Unfortunately, the robot jury had a flaw that’s common in both AI systems and human beauty contest judges. Only one of the 44 winners had dark skin, leading to accusations of racism.
Rapping
AI is already producing paintings, music and poetry, but the world of rap has thus far spurned its disruptive charms. That may change when AI rapper DeepBeat steps on the stage. The programme uses machine learning to combine lines that rhyme from existing rap songs into new lyrics, as the creators explained in a research paper titled “DopeLearning: A Computational Approach to Rap Lyrics Generation”. DeepBeat’s early recordings suggest it’s not quite time for the likes of Jay-Z to hang up their mics. Read next: If computers can be creative what does that mean for humanity?
Gaydar
Researchers from Stanford University claim to have developed a facial recognition system that can identify whether someone is straight or gay just by looking at a photograph. The algorithm was trained on a sample of more than 35,000 images posted on a dating website to identify features, expressions and hairstyles that humans think are typical of gay men and women. The model correctly predicted the sexuality of 81 percent of males and 71 percent of women, outperforming the human accuracy of 61 percent for men and 54 percent for women. The results raised concerns that similar software could be used to discriminate against homosexuals, but other scientists have debunked the claims. Analysis by Google researchers suggests that the results were not due to appearance but the angle at which gay men and women typically take selfies.
Writing Harry Potter
J.K. Rowling crushed many dreams when she announced there would be no more Harry Potter books, but AI means that Potterheads no longer need the author. A digital entertainment group called Botnik studios has written a new chapter by training an algorithm on the previous seven Potter novels. The Verge reported that the predictive text programme had some help from multiple human writers to construct the four-page chapter, titled “Harry Potter and the Portrait of What Looked Like a Large Pile of Ash.” The combination adds a surreal touch to Rowling’s prose, which yields gems such as Dumbledore’s reaction to a surprising guest at Hogwarts. “The pig of Hufflepuff pulsed like a large bullfrog,” the chapter reads. “Dumbledore smiled at it, and placed his hand on its hand: ‘You are Hagrid now’.”
Brushing your teeth
Kolibree claims to have invested the first AI-powered toothbrush. The dental technology company unveiled the device known as Ara at CES 2017. “Patented deep learning algorithms are embedded directly inside the toothbrush on a low-power processor,” explained Kolibree founder and CEO Thomas Serval. “Raw data from the sensors runs through the processor, enabling the system to learn your habits and refine accuracy the more it’s used.” The toothbrush is designed for adults, but Kolibree doesn’t want kids to miss out on the fun of dental hygiene. The company also produces an interactive toothbrush called Magik that uses augmented reality to help children brush their teeth.
Brewing beer
IntelligentX sells the world’s first beer that’s brewed by artificial intelligence. After necking one of the company’s four bottled beers, drinkers submit their tipsy thoughts via an online feedback system. Algorithms turn the data into insights that are sent them back to the brewery. Every beer you crack open is another chance to improve the recipe, so bottoms up.
