AI makes portraits talk, Samsung learns to copy the brain, and other AI news

We aim to inform readers not only about events in the Bitcoin industry, but also about developments in related technology spheres — cybersecurity, and now the world of artificial intelligence (AI).

ForkLog has gathered the most important AI news of the past week.

\n\n

Amazon unveils new AI devices

\n\n

On September 28, 2021, Amazon unveiled a range of AI-enabled products.

\n\n

Among the new products was the home robot Astro, based on the Alexa virtual assistant. It can recognise faces, respond to voice commands, carry objects and monitor its surroundings.

\n\n

\n\n

Also introduced is the Echo Show 15 smart display with sound-detection technology and Visual ID for user identification.

\n\n

\n\n

Amazon updated some Alexa privacy features, allowing the AI assistant to process data locally on devices without sending it to the cloud.

\n\n

The company also showcased the home drone Always Home Cam with motion detection for security and the Virtual Security Guard for monitoring outdoor cameras and activity.

\n\n

\n\n

Release dates for the new products will be announced later.

\n\n

Yandex adds object recognition to mobile app

\n\n

The company Yandex updated its Smart Camera in the official app.

\n\n

When recognizing simple food products, the system indicates the amounts of fats, proteins, carbohydrates, trace elements, calories, as well as sugar and GMOs.

\n\n

\n\n

The work of the ‘Smart Camera’ feature. Data: Yandex.

\n\n

The company also added vehicle recognition with information about their make, model, horsepower, speed and fuel consumption.

\n\n

In addition, the camera was trained to predict the order of actions to solve linear and quadratic equations, and to find tutorial videos analyzing similar problems.

\n\n

Google adds camera-based gesture control in Android 12

\n\n

Google added camera-expression-based control and handwriting recognition features to Android 12. They are intended for people with disabilities.

\n\n

Camera Switches lets users trigger commands with facial expressions, such as opening the mouth or raising the eyebrows.

\n\n

Project Activate reads facial expressions to play music or send a message.

\n\n

\n\n

Lookout recognises handwritten and printed text using the smartphone camera. The feature works for languages using the Latin script. Developers also added currency identification for euros and Indian rupees in currency mode.

\n\n

Samsung unveils brain-copying technology in neuromorphic chips

\n\n

Samsung proposed to ‘copy and paste’ the brain into neuromorphic chips.

\n\n

The concept involves duplicating neural connections with an array of nanoelectrodes and transferring them to a 3D memory-network module.

\n\n

The authors propose using the approach to create a microchip that achieves low power consumption, easy learnability, environmental adaptation, autonomy and cognitive abilities.

\n\n

Russian developers create transcription app for media files

\n\n

TAD Development launched an AI-powered service \”AI-Secretary\” for transcribing audio and video recordings.

\n\n

The system recognises speech on media files and converts it to text, following spelling and punctuation rules. Its accuracy stands at 80%.

\n\n

The app is free and available on the company’s website, as well as in the App Store and Google Play.

\n\n

Cruise and Waymo allowed to operate robotaxi in San Francisco

\n\n

The autonomous-vehicle makers Cruise (General Motors) and Waymo (Alphabet) received permission to operate robotaxi services commercially in San Francisco.

\n\n

The California Department of Motor Vehicles allowed Cruise to operate trips without a safety driver, while Waymo vehicles must always have a safety operator on board.

\n\n

Cruise robotaxi can transport passengers in certain parts of the city from 10:00 pm to 6:00 am, at speeds not exceeding 48 km/h. They were also allowed to drive in light rain and fog. Waymo cars were allowed to operate on public roads with a maximum speed of 104 km/h.

\n\n

Texas police file suit against Tesla over crash with five injuries

\n\n

In Texas, law enforcement filed a lawsuit against Tesla after a Model X with Autopilot engaged crashed into a group of officers.

\n\n

\n\n

On February 27, 2021, a vehicle travelling at 70 km/h crashed into two parked police cars and pushed them into a patrol vehicle. Five people sustained serious injuries.

\n\n

Plaintiffs allege that the accident was caused by ‘design and manufacturing defects’ in the vehicle, and that the autopilot failed to recognise police cars and warn of hazards.

\n\n

Lawsuit against DeepMind over use of patient data in the UK

\n\n

In the UK, a lawsuit was filed against AI lab DeepMind over alleged unlawful use of medical data.

\n\n

In 2015, the company and the Royal Free London NHS Foundation Trust entered into an agreement under which the developers gained access to records of 1.6 million patients to develop applications for doctors.

\n\n

On 30 September 2021, the law firm Mishcon de Reya filed a class-action on behalf of Andrew Prismoll and other UK citizens against DeepMind for obtaining confidential information \”without permission, knowledge or consent.\”

\n\n

AI challenges authenticity of famous Rubens painting

\n\n

AI researchers using AI refuted the authenticity of Rubens’ painting \”Samson and Delilah,\” displayed at the National Gallery in London.

\n\n

\n\n

Swiss experts analysed the painting and found that ‘every fragment was more than 90% fake’. They repeated the experiment and the result did not change.

\n\n

A museum representative said they always take new research into account and will await full publication of the study to assess the evidence.

\n\n

The National Gallery in London bought the canvas in 1980 for $3.4 million, which is equivalent to about $11.5 million today after inflation.

\n\n

AI startup teaches portraits to move and talk

\n\n

The Israeli startup D-ID developed the Speaking Portraits tool that animates a person in a photo and makes them speak.

\n\n

\n\n

To train the algorithm, users need to upload a text file, audio or a short video.

\n\n

The developers created two variants — Single Portrait and Trained Character. The former animates only the head, while the latter generates a character with a predefined set of actions and words.

\n\n

Neural network finishes Beethoven’s symphony

\n\n

A group of musicians and developers finished Beethoven’s tenth symphony with the help of artificial intelligence.

\n\n\n\n

The authors studied the works of the German composer, analysed them, and digitised them. On this data they trained a neural network, into which they later fed drafts of an unfinished work. Over 18 months the team created and orchestrated two full movements of the tenth symphony, each lasting more than 20 minutes.

\n\n

The researchers will present the results on 9 October 2021.

\n\n

Also on ForkLog:

\n\n

• In Russia AI will identify \”dangerous students\” by essays.
• Atari creator will release an NFT collection of classic arcade machines in augmented reality.
• Scientists developed an algorithm to identify animal-to-human transmissible viruses.
• DeepMind developed an algorithm for predicting rain several hours ahead.
• The EU and the US will develop common AI principles.

\n\n

Subscribe to ForkLog’s Telegram news: ForkLog AI — all the AI news!