New Scientist


AI suggests recipe for a dish just by studying a photo of it

New Scientist

Researchers have devised a machine learning algorithm that looks at photos of food and predicts the recipe that created the dish. The AI can also work out from a photo what ingredients went into a food: presented with an image of a plate of biscuits, for example, it knows that they are likely to include flour, eggs and butter. App such as MyFitnessPal already let people track calorie intake, but they have to manually input what they eat. But image recognition algorithms can only go so far, says Christoph Trattner at MODUL University Vienna in Austria.


Robot physical therapist helps people walk again after a stroke

New Scientist

Artificial intelligence is helping people regain their mobility after certain neurological injuries. To avoid persistent difficulties walking after a stroke or spinal injury, walking assistance is crucial. The new system improved the in-harness gait of people following a stroke or a spinal injury. And after a single, 1-hour training session with the smart harness, people with spinal cord injury showed immediate improvement in their gait out of the harness over those given no physio session at all, the authors report today.


Buzz of drones is more annoying than any other kind of vehicle

New Scientist

Amazon, UPS, Domino's Pizza and other companies planning drone delivery services may be heading for discord. A preliminary NASA study has discovered that people find the noise of drones more annoying than that of ground vehicles, even when the sounds are the same volume. "We didn't go into this test thinking there would be this significant difference," says study coauthor Andrew Christian of NASA's Langley Research Center, Virginia. However, Christian points out that simply making drones "only as noisy" as delivery trucks would still mean they are more annoying, meaning companies may need to find ways to make their drones significantly quieter than ground vehicles.


Find your next must-play game by flying through a virtual galaxy

New Scientist

"There are thousands and thousands of games," says James Ryan at the University of California, Santa Cruz. Instead of relying on the opinions of those who might also be trapped in their own filter bubbles, GameSpace drops you into a galaxy where every star represents a game, and similar titles are grouped into constellations. But then he read the cereal game's description and saw it was indeed an adaptation of Doom. Extending it to books and films should be possible for any title with a Wikipedia description available online, says Ryan.


AI doctors should improve healthcare, but not at any cost

New Scientist

It sounds banal until you realise that the trainee might be an artificially intelligent voice-recognition system that requires real-world data to learn its trade. "Data collection and analysis is changing so rapidly that systems of governance can't keep up" Such questions of propriety and custodianship have been asked about data before – but medical information is uniquely valuable and sensitive. As revealed by New Scientist, the deal gave the AI company access to 1.6 million people's medical records to develop a monitoring tool for kidney patients: the ICO ruled that they were not properly informed about the use of their data, among other shortcomings. A report by the Royal Society and the British Academy recently concluded that the collection and analysis of data is changing so rapidly that the UK's systems of governance cannot keep up.


AI coach will train hopeless chatbots to pass the Turing test

New Scientist

US Justice Potter Stewart's famous turn of phrase could also be an apt description of the Turing test, our judgment of whether an AI seems convincingly human. Given a snippet of dialogue between a chatbot and a human, the system predicts how convincingly human you or I would rate the chatbot's response. He chose 1000 short Twitter conversations, and got human volunteers to add a response. After its training, Lowe's algorithm was able to match the judgment of the human evaluators.


Glove turns sign language into text for real-time translation

New Scientist

A new glove developed at the University of California, San Diego, can convert the 26 letters of American Sign Language (ASL) into text on a smartphone or computer screen. "For thousands of people in the UK, sign language is their first language," says Jesal Vishnuram, the technology research manager at the charity Action on Hearing Loss. In the UK, someone who is deaf is entitled to a sign language translator at work or when visiting a hospital, but at a train station, for example, it can be incredibly difficult to communicate with people who don't sign. The flexible sensors mean that you hardly notice that you are wearing the glove, says Timothy O'Connor who is working on the technology at the University of California, San Diego.


Transformer robots can be printed on demand in just 13 minutes

New Scientist

It works by bending wire that already has motors attached into different shapes, using a process its designers call 1D printing. "The idea is that you analyse the current situation, then make a robot on the fly that can deal with it," says Sebastian Risi at the IT University of Copenhagen in Denmark, a member of the team that came up with the system. The system uses evolutionary algorithms that improve their designs bit by bit until they reach one that satisfies all the constraints. This means that anyone could create their own bespoke printed robots with no prior knowledge required.


Brain-training game fails test against regular computer games

New Scientist

The thinking is that this should improve a player's memory, attention, focus and multitasking skills. For this study, Kable and his colleagues recruited 128 young healthy adults for a randomised controlled trial. Those who played Lumosity did show improvements in some cognitive skills, such as attention and focus, but so did those who played the other computer games, and the people who played no games at all. The number of people involved in Kable's study was too small to detect any tiny improvements in performance, so it's possible a small effect was missed.


Neural network poetry is so bad we think it's written by humans

New Scientist

Its best efforts even fool people into thinking they're reading the words of a human poet, rather than the algorithmic output of a cold-hearted AI. Hopkins employed a similar mechanism to persuade the AI poet to write lines that rhymed or followed a particular rhythm. For example, Hopkins could make the AI write poetry in iambic pentameter – the poetic rhythm common in Shakespeare's plays and sonnets. Lack of creativity aside, the neural network still managed to fool some people who thought the poetry was written by a human.