Chess engine with human-like behavior, OpenAI’s new neural network, and other AI news

We aim to inform readers not only about developments in the bitcoin industry but also about what is happening in related technological spheres — cybersecurity, and now the world of artificial intelligence (AI).

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

Amnesty International calls for ban on facial recognition technologies

Amnesty International launched the “Ban the Scan” campaign and urged the global community to prohibit the use of facial recognition technologies for state surveillance.

Amnesty noted that such software often errs. It could also exacerbate discriminatory policing, intensify racist police actions, invade privacy and threaten the rights to peaceful assembly.

As part of the campaign, rights advocates compiled a map of facial-recognition cameras in New York City and announced the release of a tool to submit requests to authorities to learn where the technology is used.

Researchers developed a chess AI engine that plays like a human

A group of researchers from Cornell University, the University of Toronto and Microsoft developed Maia, an AI engine that is not programmed to win a chess game at all costs. Instead, it seeks to reproduce what a human would do.

For Maia, researchers used an open-source Leela system, based on AlphaZero from DeepMind, and trained the model on individual moves from millions of online games with humans. This approach allowed them to tune the algorithm for different skill levels from beginner to advanced.

In December, researchers made Maia available on the chess site lichess.org. In the first week, over 40,000 games were played against it. The researchers found that Maia matched human behavior in more than half the cases.

The developers believe the algorithm could help players study their mistakes and sharpen their own playing skills.

MIT researchers developed a deep learning algorithm for diagnosing breast cancer from a patient\’s mammogram

Researchers from MIT and the Abdul Latif Jameel Clinic developed a deep learning system Mirai capable of predicting breast cancer risk from a patient\’s mammogram. The developers emphasise that the new algorithm performs equally well irrespective of the patient\’s race or ethnicity.

Mirai can model “patient risk at several future time points” taking into account minor deviations. Its predictions can be further optimised if other clinical risk factors, such as age or family history, are available.

The MIT team trained the model on a dataset of 200,000 examinations from Massachusetts General Hospital and then tested its prognostic results on additional datasets. According to the researchers, Mirai is “significantly more accurate” in predicting cancer risk for patients and can correctly identify almost twice as many potential cancer cases among high-risk groups as the currently used diagnostic model.

Mirai also accounts for risk factors that are not necessarily visible on a mammogram, such as patient age, hormone levels and menopausal status. These factors are embedded during training, enabling the model to forecast even if the clinician did not provide the information manually.

Google rolls out AI route planner for electric vehicles on Android via charging stations

Google announced a new feature for electric-vehicle owners that uses AI to sort thousands of publicly available charging stations in the United States and find the optimal route through them.

The algorithm also provides users with recommendations on when to stop charging the EV depending on its location, remaining range, and the type of plug used for charging.

The feature currently operates in the United States on vehicles running Google’s Android Automotive OS. There are currently only two such cars – Polestar 2 and Volvo XC40 Recharge – but the company notes that more models from different manufacturers will appear in the near future.

Researchers developed AI-based technology to aid in smoking cessation

American researchers developed an AI-powered on-board computer “Level” for heated tobacco systems that measures the number and duration of inhalation sessions, the time between uses, and puff strength.

According to the developers, the device could help people quit smoking by predicting when they will crave nicotine, controlling their urge to smoke and reducing the number of sessions.

Researchers hope that in the future information about vapers’ behaviour will also help traditional cigarette smokers quit nicotine dependence. They note that in the US alone about 68% of smokers consume both traditional and electronic cigarettes.

OpenAI unveils a neural network that creates images from descriptions

OpenAI developers presented the generative neural network DALL-E, which can create images of non-existent things, such as an avocado chair, a giraffe-robot or radishes in a ballet tutu.
DALL-E is built on the GPT-3 language model and trained to generate images from text descriptions using a dataset of text-image pairs.

The algorithm understands complex abstract combinations, can alter the style and shape of objects in the image, and create photorealistic pictures.

The network’s name is a blend of artist Salvador Dalí and the robot WALL-E from the popular film.

Subscribe to ForkLog news on Telegram: ForkLog Feed — all the news feed, ForkLog — the most important news and polls.