💡 Inside AI Weekly Roundup: The Top Seven Stories To Know:

• Google, Stanford, and MIT claimed the top affiliations of accepted papers at this year's Neural Information Processing Systems (NeurIPS 2020).
• Microsoft unveiled joint efforts to combat the so-called “data desert” and make AI more inclusive for people with disabilities.
• Google's AI chat agent Duplex, mostly known for making restaurant reservations, can now make autonomous calls to salons or barbershops.
• Two software engineers used deep learning models to create an AI that turns people's photos into Pixar-like cartoons.
• Microsoft's latest image-captioning AI can describe photos better than humans in limited tests.
• A rapid COVID-19 test from the University of Oxford uses machine learning to identify the virus in throat swabs in less than five minutes.
• Brazil’s largest AI research center has opened in São Paulo.

Nvidia says it will power the world’s fastest AI supercomputer. The system, named Leonardo, is currently under development by the Italian multi-university consortium Cineca and is expected to deliver up to 10 exaflops of FP16 AI performance.

More:

• The supercomputer will have ~14,000 Nvidia Ampere-based GPUs when it goes online in 2021. Cineca researchers plan to use the computing power to simulate the planetary forces resulting in climate change, as well as molecular movements inside the coronavirus molecule.
• Calculations indicate it could rank among the top three current Top 500 supercomputers. As of June, the top one is Japan's Supercomputer Fugaku followed by IBM's Summit.
• Cineca also unveiled plans for three more supercomputers that will use Nvidia A100 GPUs: MeluXina in Luxembourg,  IT4Innovations National Supercomputing Center in the Czech Republic, and the Vega supercomputer in Slovenia. French company Atos will build Leonardo and two of the other supercomputers. HP Enterprise will construct the fourth supercomputer. All are backed by EuroHPC, which says they'll help power "Europe’s data economy."
• Earlier this month, Nvidia announced that it would build Britain’s most powerful supercomputer. The company will invest around $52m on the “Cambridge-1” supercomputer, which will comprise of 80 Nvidia systems and focus on AI research in health and medical care, including COVID-19 challenges. It's expected to come online this year.

Google is using AI and machine learning to improve its search results, including better spell checking and a tool that matches a person's humming with the correct song. Unveiled at this week's “Search On” event, most of the improvements will roll out in the coming weeks.

More:

• Since 1 in 10 Google search queries are misspelled, the company is updating the feature with a deep neural net-powered algorithm that will offer better suggestions in its "Did you mean" prompt in searches. With 680 million parameters, the tool will operate in under three milliseconds and help correct misspelled words. Google says the change "makes a greater improvement to spelling" than all its improvements in the last five years.
• A new algorithm, due out next month, will also index individual passages from web pages rather than the whole webpage. Google expects it will improve 7 percent of search queries globally. The company will also divide broader searches into subtopics using neural nets.
• Computer vision and speech recognition will soon tag and parse videos into parts automatically. The AI-driven approach allows users to tag key moments in a video and "navigate them like chapters in a book."
• Google says its AI language model BERT is now used in almost every English search query to generate better quality results.
• The company is also making datasets through the Data Commons Project more accessible through search. For example, it will use natural language processing to map a search into a "specific set of the billions of data points in Data Commons."
• Lastly, the ML-powered ‘Hum To Search’ feature is now available through the Google Search app or the Search Widget. Users can sing, hum, or whistle a song for 10-15 seconds and Google will generate matches, ranking them by highest likelihood. To work, Google says its ML models transform the audio into "into a number-based sequence representing the song’s melody," in part by removing details like accompanying instruments.

The U.S. currently lacks a "whatever it takes" attitude to become the world leader in AI, according to former Google CEO Eric Schmidt, who chairs the National Security Commission on AI. During Thursday's Politico AI Summit, Schmidt noted that advancing AI has been a uniquely bipartisan issue for U.S. lawmakers as they seek to counter China and other global AI powers. Along with AI commission co-chair Robert Work, he recently stressed that the U.S. is in an unprecedented "innovation competition" and needs more aggressive actions to develop and incorporate AI.

More:

• Schmidt said the goal is to have America - or at least the Western world - research and invent AI technologies. He stressed the need for peer collaboration among democracies like the U.K., Canada, and Israel to advance AI and fight against what he's called "high tech authoritarianism."
• Schmidt has been warning about the U.S. lagging behind in AI development since at least 2017, when he predicted that China would take the lead in AI starting in 2022.
• Earlier this week, Schmidt's National Security Commission on Artificial Intelligence released its latest report and third-quarter recommendations on how the U.S. can strengthen its AI and work with partners. These include strengthening the "triangular alliance" for AI R&D in academia, government, and industry and establishing new career fields, certifications, grants and more for AI development and the workforce. [Note: We covered many of the recommendations in Wednesday's issue, which is available for Premium subscribers.]
• Schmidt has previously referenced China's development plan to become the global leader in AI by 2030.

Amazon Web Services (AWS) agreed to a data management partnership with Cohesity. As part of the agreement, Cohesity will analyze, manage, and protect the data of its customers using AWS. Cohesity customers will now get access to different AWS services such as Amazon SageMaker, which allows software developers and data scientists to build machine learning models.

More:

• Prior to the agreement, Cohesity would design and sell its data management software to its business customers and let them run it themselves.
• Cohesity customers can also use Amazon Macie which uses machine learning to protect data. Cohesity currently has 2,000 business and government customers, including NASA.
• 35 trillion gigabytes of new data will be created in 2020. If AWS and Cohesity are able to offer data management as a service tool that's an extension of AWS, IT departments would be able to focus on building services for employees working remotely during the pandemic.
• Cohesity was founded in 2013 and has raised $650m in venture capital funding from investors such as SoftBank's Vision Fund and Sequoia Capital. Amazon participated in Cohesity's $250m Series E funding round earlier this year.
• The global database management system market generated $55b in revenue in 2019, up from $46b in 2018. Cloud service companies like AWS made up 30% of this revenue, up from 7% in 2016. Cloud services are expected to account for more than 50% of this market within the next three years.

A version of this story first appeared in Inside Business. You can read the full issue here. 

Cruise received approval from California’s Department of Motor Vehicles to test its completely driverless vehicles on roads in the state. GM’s autonomous driving division is the fifth company to receive a permit to operate self-driving cars without a backup safety driver, following Waymo, Zoox, AutoX and Nuro.

More:

• The permit allows Cruise to operate a driverless fleet of five vehicles on specified streets in San Francisco. The vehicles cannot operate on roads with a speed limit over 30 mph and cannot be tested during times of heavy rain or fog.
• Cruise CEO Dan Ammann said Cruise intends to be the first autonomous developer to test driverless vehicles, without backup safety drivers, in the dense, populous city. The other four developers have opted for more suburban areas throughout the state.
• 60 tech companies and auto developers currently have an active permit to test self-driving cars in California, but are required to have back-up safety drivers.

A version of this story first appeared in Inside Transportation.

Feature: The pandemic's impact on AI

(Note: This paid-only premium content first appeared in the Sept. 30 issue of Inside AI. You can sign up for premium here.)

A new survey of IT directors shows most large companies spend $1m or more on AI projects and actually plan to raise their AI/ML spending due to the pandemic. Algorithmia, which sells ML operations and management software, asked over 100 IT directors working at large companies about their AI/ML trends before, during, and after the pandemic. Organizations are now more worried about a lack of funding and executive buy-in, they found.

The findings:

• Two thirds (66%) said they realized that AI/ML projects matter more than they thought before the pandemic. 23% said they think it should have been their company's highest priority.
• Half said they plan to spend more on AI/ML as a result of the pandemic. 21% plan to spend a lot more.

• More than half of large enterprises have more than 20 AI/ML projects going on right now. The largest segment (31%) had 50 or more projects.

• The top priorities are projects that generate short-term cost savings and/or insights about customer behavior. Long-term process automation and back-office projects have fallen priority-wise.
• Many of the businesses have refocused their AI/ML efforts toward projects that support remote workers and impact customer experiences. 17% reduced their project staff size.

• Large enterprises (64%) have AI/ML teams of over 50 people and a quarter have over 100 people.
• 91% of survey respondents said they were already spending at least $1m annually before the pandemic. 30% had budgets above $10m annually.

• A lack of in-house staff with AI/ML skills was the primary challenge for IT leaders. The most important job skills will be security (69%), data management (64%), and systems integration (62%).
• Organizations are more worried about a lack of funding and executive buy-in because of the pandemic. A third of companies still face challenges from insufficient tools for managing their AI/ML projects and operations.

QUICK HITS

• A University of Waterloo research paper says AI should be able to identify more objects than the number of examples it was trained on - a process known as “less than one”-shot or LO-shot learning.
• Scientists mapped 1.8 billion tree canopies in West Africa using AI, which analyzed a database of satellite images.
• New data obtained by The Information shows UPS has only invested $2.5m into autonomous trucking startup TuSimple.
• As Apple device adoption grows in the enterprise, organizations are looking for ways to seamlessly manage and secure devices while extending the world-class Apple experience users expect.*

*This is sponsored content.