Breakthrough in Quantum Computing, Sarcasm Detector and Other AI News

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

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

IBM accelerates data processing 120× with quantum computers

At the THINK conference IBM announced a breakthrough in quantum computing and unveiled the Qiskit Runtime service.

Rather than storing data on a physical quantum computer, the company proposed using a hybrid cloud. This enables a 120-fold speedup.

Qiskit Runtime enables quantum systems to perform complex computations, such as chemical modelling and financial-risk analysis, in hours instead of weeks. For example, it simulated the lithium hydride molecule in nine hours, whereas previously it took 45 days.

IBM also unveiled a CodeNet dataset for training neural networks in programming.

It comprises 14 million fragments of code and 500 million lines of code across 55 languages—from COBOL and FORTRAN to C++ and Python.

Developers can use CodeNet for code search and clone detection, or use it as a benchmark dataset.

Each sample of code is timestamped with CPU time and memory usage, enabling researchers to develop automatic code-correction systems.

IBM made CodeNet publicly available and placed the repository on GitHub.

The company demonstrated a number of AI-based tools, including Watson Orchestrate, AutoSQL and Cloud Pak.

Tesla says it cannot deliver a fully autonomous car by year-end

The board informed the California Department of Motor Vehicles that their autopilot would not reach full autonomy by the end of 2021. In January, Elon Musk claimed that engineers would push the technology to Level 5.

“Tesla is struggling to say whether the pace of improvements will deliver a fifth-level by the end of the calendar year,” the memo said.

In the company’s view, their autopilot remains at Level 2 autonomy.

Nevertheless, Elon Musk tweeted that Tesla will begin beta-testing Full Self-Driving (FSD) within the next two months.

He said that in the coming month the company would also open access to FSD via a monthly subscription.

Meanwhile, the US National Highway Traffic Safety Administration (NHTSA) opened an investigation into a fatal Tesla crash in California on May 5.

The electric car collided with a truck that had rolled over five minutes earlier. The Tesla driver died, the second-car driver was injured, according to the California Highway Patrol.

The report does not indicate whether the autopilot was engaged at the time of the crash.

NHTSA said it had opened 28 special investigations into Tesla crashes, 24 of which are under review.

Against regulator’s statements, the company’s shares fell 4.4%.

In California, a man was arrested for using the Tesla autopilot in the back seat

In California, 25-year-old Paras Sharma was arrested for using the Tesla autopilot from the back seat of the car. No one was behind the wheel during the drive.

The young man had previously come to the attention of law enforcement.

Sharma also frequently posts videos on Instagram, where he drives on public roads, fully trusting the autopilot.

The young man will face trial on July 6. He said he has no desire to change his lifestyle and plans to return home from prison “in the back seat of his Tesla.”

Media: Pentagon plans to terminate contract with Microsoft over Amazon’s lawsuit

The Pentagon may terminate the JEDI project amid Amazon’s court battle, according to reports.

The agency argues that in 2019 Trump interfered in the AI technology competition for the department worth $10 billion and pressed for Microsoft.

According to Amazon, the ex-president sought to settle scores with the CEO, who owns The Washington Post. The publication has often criticised Trump’s actions.

Amazon contends that its technologies outperform Microsoft’s and are better suited to taxpayers’ needs.

Microsoft acknowledged that the dispute has taken a toll, but the company remains ready to implement the project.

US Army develops a sarcasm detector

Researchers from the University of Central Florida, at the DoD’s behest, developed a tool that “almost perfectly” recognises sarcasm online.

They trained neural networks to trace dependencies between keywords and produce a classification score for sarcasm presence in messages.

According to the researchers, the algorithm will help identify user sentiment and flag signals for information operations online. It will help identify topics of concern or possible attacker actions.

Researchers developed a tool to protect selfies from facial recognition

At the International Conference on Artificial Intelligence (ICLR), researchers introduced two free tools to shield selfies from facial-recognition systems.

Fawkes was developed at the University of Chicago. It adds barely perceptible pixel-level distortions to images that interfere with algorithms identifying people.

However the authors note that their tool cannot fool systems already trained on unprotected photos.

Fawkes can be downloaded for Windows and Mac.

Another tool — LowKey — was developed by researchers from the University of Maryland. It distorts images to such an extent that it can fool commercial AI models such as Clearview AI, AWS Rekognition, Microsoft Azure, and Face++.

LowKey is available as a web service.

Ukraine approves national AI development plan

The Cabinet of Ministers of Ukraine has approved the AI development concept for 2021-2024.

According to the annex to the resolution, among the main priorities the cabinet listed:

• legal regulation of AI use;
• promotion of AI among youth;
• government support for AI use in priority sectors of the economy;
• integration of AI into cybersecurity, defence, healthcare, and public administration;
• defining measures for the social reintegration of convicted individuals via data analysis.

Prime Minister Denys Shmyhal instructed the relevant ministries to ensure plan implementation and to report annually to the Ministry of Digital Transformation on its progress.

American company develops algorithm to detect armed threats

The American company has developed computer-vision technology to detect armed threats even before the shooter enters the building.

According to the developers, they trained the algorithm on hundreds of different weapon types and thousands of hours of video. The system can detect a threat at a distance of three to five metres.

If it detects a weapon, the image from the cameras is immediately sent to an external monitoring service where a human assesses the threat and decides whether to call the police.

Russian scientists train drones to recognise Sosnovsky’s hogweed

Researchers from the Skolkovo Institute developed AI to analyse drone images and recognise Sosnowsky’s hogweed — a poisonous weed dangerous to plants and people.

A fully convolutional neural network handles processing. It has been optimised to run on a single-board computer, enabling near real-time field analysis.

According to the developers, their model surveyed 28 hectares in 40 minutes at an altitude of 10 metres, not missing a single plant.

Researchers say their technology will enable a cheaper fight against hogweed. According to some data, in the last six years the Moscow region alone spent 1 billion rubles to destroy hogweed.

Also on ForkLog:

• The US Senate proposed amendments to the AI funding bill.
• Military experts: the AI frenzy will harm the Pentagon.
• At MIT developed a system for automatic cleansing of messy data.
• Study: patients do not trust AI doctors who know their names.
• Neural networks helped a paralyzed man write text messages.

Subscribe to ForkLog news on Telegram: ForkLog AI — all AI news!