We aim to inform readers not only about developments in the Bitcoin industry but also about what is happening in adjacent technological spheres — cybersecurity, and now the world of artificial intelligence (AI).
ForkLog has gathered the week’s most important AI news.
- The Russian government will allocate 3 billion rubles to support AI developers.
- Apple has scanned user emails for CSAM content since 2019.
- A supercomputer capable of training an AI model with 120 billion parameters has been unveiled.
- An American spent 11 months in jail due to an error in the ShotSpotter gunfire-detection system.
- Researchers trained robots to interact with one another during a joint task.
- In Japan, an AI-powered doll for lonely seniors was developed.
The Russian government will provide state support to AI developers
The Russian government will provide state support to developers of hardware-software AI systems for the advancement of AI technologies. In 2021 authorities allocated 1.5 billion rubles from the federal budget.
Under the decree, engineers will receive up to 3 billion rubles for projects lasting up to six years, at 500 million rubles per year. Companies may use the funds for research and development of products, payroll, equipment and the creation of high-tech jobs, production of prototypes of hardware-software complexes (HSCs) and their pilot batches.
Organizations must undergo a competitive selection process to participate in the program. Developers must also commit that revenue from the project will be at least twice the amount of the subsidy received.
The Russian Interior Ministry plans to use AI to identify serial offenders
The Russian Interior Ministry plans to build an AI-based system to identify signs of serial criminal and administrative offenses. The ministry issued a tender and allocated 23.1 million rubles for the project.
Law enforcement is considering the applicability of speech recognition and optical character recognition for handwritten and printed texts. They are also exploring the possibility of identifying named entities — full names, dates and places of birth, gender, kinship, distinguishing marks, mentions of drugs, weapons, method and place.
According to officials, the police currently manually establish links between serial offenses. AI software will automate the analysis, they added.
Turkey unveils AI development strategy
The Turkish government unveiled a road map for AI development. Officials highlighted six key priorities:
- training AI specialists and increasing employment in the field;
- support for research, entrepreneurship and innovation;
- ensuring access to high-quality data and technical infrastructure;
- taking measures to accelerate socioeconomic development and adaptation;
- strengthening international cooperation;
- accelerating governance transformation.
The program aims to raise AI’s contribution to GDP by 5 percentage points from 2021 to 2025 and to provide employment in the sector for 50,000 people.
Apple began scanning iCloud Mail for CSAM in 2019
Apple began scanning iCloud Mail for CSAM content in 2019.
Company representatives confirmed the information. According to them, the algorithms analyze cloud-linked email accounts and examine incoming and outgoing messages.
They also clarified that the algorithms do not scan photos in iCloud or backups of devices in the cloud.
Cerebras CS-2 can train AI models with up to 120 trillion parameters
The Cerebras CS-2 can train AI models with 120 trillion parameters.
CEO Andrew Feldman said the technology will increase the size of the largest existing neural networks by a factor of 100.
CS-2 runs on the Wafer Scale Engine 2-generation chip, designed for machine-learning workloads. It contains 2.6 trillion transistors and 850 thousand cores optimized for AI tasks.
Snapchat gains object search in augmented reality via the camera
The company Snap updated the augmented-reality Scan feature in Snapchat.
The scan button is now centered in the app and can recognize real-world objects. Users can identify clothing, dog breeds, plants, cars, and product compositions.
The company also added a shopping-by-screenshot feature and camera shortcuts that, depending on the scene, recommend the optimal shooting mode, soundtrack, and filter.
American spent 11 months in jail due to AI error
A Chicago court acquitted a man who had spent 11 months in jail because of evidence produced by the ShotSpotter gunfire-detection AI. The neural-network data that detects gunfire and relays information to police proved faulty.
Public defenders said the algorithm initially identified a fireworks display as the sound of gunfire from 1.5 km away from the incident, after which ShotSpotter staff classified it as a gunshot at the scene. The company maintains that it did not alter the model data to bolster the case.
Due to insufficient evidence to prove guilt, prosecutors withdrew the ShotSpotter report, and the judge dismissed the case against the man.
Scientists develop a neural interface restoring fingertip sensitivity in paralyzed individuals
Researchers at the Feinstein Institute for Medical Research have developed a brain implant that restores fingertip sensation in people with paralysis and neuropathy.
Researchers applied stereo-electroencephalography and implanted electrodes in the brain sulci of two volunteers with drug-resistant epilepsy. After a minimally invasive procedure, the patients reported tingling and a sensation of electricity in their hands and fingertips.
According to Dr. Kevin J. Tracy, the approach could help people recover functions lost due to spinal-cord injuries or strokes.
Researchers taught robots to collaborate as a team
Researchers from the University of Cincinnati taught two robots to move a sofa to learn to work together as independent participants.
Researchers trained them to use \”fuzzy logic\” rather than directing one another and to share a pre-built strategy. In a virtual environment, the devices carried a sofa around two obstacles and passed through a narrow doorway. They completed the task successfully in 95% of cases.
According to the project authors, their goal is to train more than ten robots to cooperate with each other to perform complex tasks, such as moving furniture.
In Japan, an AI-powered doll for lonely people was developed
Japanese company Takara Tomy has developed an AI-powered doll named Ami-Chan to alleviate loneliness among elderly people during periods of isolation.
The doll can blink, imitate sleep, remember conversation partners’ names and their birth dates, talk and understand speech. It has a vocabulary of 1,600 Japanese words, and it continually learns new ones. The doll knows 40 songs, poems and tongue twisters, and a facial-recognition algorithm helps it understand whom it is interacting with.
According to government data, Japan has almost 36 million elderly people, and 6 million of them live alone.
Also on ForkLog:
- Britain will continue the investigation into the Nvidia-ARM deal.
- Study: AI is not capable of recognizing human emotions.
- In the United States the use of facial recognition technologies will be expanded.
- Fetch.ai launched an NFT marketplace for AI-generated artworks.
- Alethea AI, backed by Mark Cuban, attracted $16 million in a token sale.
- Media: Clearview AI provided a trial version of facial recognition to law enforcement worldwide.
- Ministry of Digital Transformation and Ukroboronprom will work on AI development in Ukraine.
- In South Korea, Facebook was fined $5.5 million for illegal biometric data collection.
Subscribe to ForkLog’s AI news on Telegram: ForkLog AI — all the AI news!