Inside | Real news, curated by real humans
Inside AI

Inside AI (Jan 2nd, 2020)

1. An AI system was able to diagnose cases of breast cancer more accurately than human experts, according to a new study that tested the algorithm's accuracy. Google Health and cancer researchers developed the system, which they trained on mammogram images from nearly 29,000 women in the U.S. and the U.K. The AI outperformed six radiologists in reading the images, reducing false positives by 5.7 percent and false negatives by 9.4 percent for U.S. patients. Experts say AI could eliminate the need for two doctors to read each mammogram, as is currently the practice in the U.K. - BBC

2. Privacy advocates in India are concerned about the use of facial recognition to screen crowds at a political rally last month, saying it's a slippery slope into mass surveillance. Police in Delhi used the Automated Facial Recognition System, or AFRS, software during a December 22 rally for Prime Minister Narendra Modi, which reportedly marked the first time that such technology was used during a political event in India. A police spokesman said the system was deployed because of "credible intelligence inputs about possible disruptions." Apar Gupta, executive director of digital advocacy group Internet Freedom Foundation, said using the system to surveil and profile people at public events is "illegal and unconstitutional." India already uses facial recognition at some offices, cafes, and airports as the country prepares to install a nationwide system. - AL JAZEERA

3. An Illinois law that aims to make AI hiring practices more transparent took effect on Wednesday. The state’s Artificial Intelligence Video Interview Act requires companies to notify applicants when they use AI in videos to determine a candidate's "fitness" for the job. It restricts who can view the video interviews (experts only) and forces companies to delete videos within a month if an applicant requests it. As Vox notes, the law only forces companies to gain applicants' consent before using AI, but doesn't require them to provide any alternatives if a candidate chooses to opt out. “It’s hard to feel that that consent is going to be super meaningful if the alternative is that you get no shot at the job at all,” says Aaron Rieke, managing director of the technology rights nonprofit Upturn. - VOX

4. China-based smartphone company Xiaomi plans to invest approximately $7.2 billion into AI and 5G technologies over the next five years, Reuters reports. Thursday's pledge from CEO Lei Jun expounds on his original promise to invest $1.43 billion on AI and internet technologies through 2025. Lei said the investment in AIoT is designed "to ensure we win in this new smart era." Xiaomi, which went public on the Hong Kong Stock Exchange in 2018, competes directly with Huawei, which claims more than 40 percent of China’s smartphone market. - REUTERS

5. During next week's CES 2020 trade show, Ambarella plans to demo its new robotics platform based on its CVflow architecture for AI processing. The platform targets automated guided vehicles, industrial and consumer robots, and emerging industry 4.0 applications. The chip designer plans to demo the highest-end version of the platform – a single CV2 chip – to perform stereo processing, object detection, occupancy grid, key points tracking, and visual odometry. Ambarella also recently announced a partnership with Amazon Web Services, which allows customers to use SageMaker Neo to train machine learning models and run them on the CVflow-powered AI vision system on chip. CES takes place Jan. 7-10. - VENTURE BEAT

6. Similarly, Samsung plans to unveil its AI-enabled human "Neon" at CES. Few details are available online so far, though the company has implied that Neon will surpass the capabilities of its Bixby voice assistant. It still refuses to call Neon a voice assistant or even an AI, describing the system as an "artificial human" and an "artificial intelligence being." Samsung Technology and Advanced Research Labs, or STAR, developed Neon, which it says is "smart enough to become one’s best friend." - BGR

7. Tesla released software version 2019.40.50.1, which offers a preview of the vehicle’s “Full Self-Driving” suite. The update allows select vehicles to recognize stoplights and stop signs, with the vehicle automatically braking if a driver tries to run through a stop sign. The update also added “camp mode,” which allows vehicles to maintain airflow, temperature and interior lighting when drivers want to use their Tesla as a makeshift tent. The update also added new driver profile features, voice-to-SMS functionality and dash-cam clips being saved with a honk, among other features. - MY TESLA ADVENTURE

A version of this story first appeared in Inside Auto.

8. An artist collective developed a GAN that generates and texts people images of fake feet. – an apparent nod to the similarly-generated – was trained on images of feet but doesn't specify what dataset was used. When Motherboard tested the system, it received some normal-looking photos and others that were a "pure nightmare," it said, noting that the GAN "seems more concerned with quantity over quality." Why the AI exists in the first place could be to "possibly to highlight the commodification of feet pics in our extremely online world," notes Vice's Matthew Gault. - MOTHERBOARD

9. A national research institute in Seoul plans to install 3,000 AI camera systems this year as it seeks to crack down on crime in the South Korean capital. The AI software can automatically detect what people are wearing, their gaits, and the time of day when considering how likely a person is to commit a crime. If the captured data exceeds a certain threshold, the system will alert police and the local office of Seoul's Seocho District. The software, which is expected to be finalized in its full version by 2022, will be rolled out to other districts and provinces in the coming years. - ZDNET

10. VentureBeat interviewed some top minds in the AI field, asking them about recent progress and how they expect the industry to change in the coming new year. The interviewees included Google AI chief Jeff Dean, University of California Berkeley professor Celeste Kidd, Nvidia machine learning research head Anima Anandkumar, PyTorch lead Soumith Chintala, and IBM Research director Dario Gil. We highlighted a few of their conversations below:

“I actually think 2020 will be the year when we start thinking [in a more complex way], where it doesn’t matter if your model is 3 percent more accurate if it … doesn’t have a good interoperability mechanism [or meet other criteria].” - Chintala

“All kinds of different iterative algorithms I think are the future, because if you just do one feed-forward network, that’s where robustness is an issue. Whereas if you try to do many iterations and you adapt iterations based on the kinds of data or the kind of accuracy requirements you want, there’s much more chance of achieving that." - Anandkumar

“It’s still so broadly inefficient the way we train deep neural networks with existing hardware with GPU architectures,” he said. “So a really fundamental rethinking on that is very important. We’ve got to improve the computational efficiency of AI so we can do more with it.” - Gil

Written and curated by Beth Duckett, a former reporter for The Arizona Republic who wrote a book about the solar industry and frequently covers hobby and commercial drones. You can follow her tweets about the latest news in artificial intelligence here.

Subscribe to Inside AI