Report: Microsoft and AWS scored $50m in contracts after Google pulled out of Pentagon’s AI drone plan

In Brief Microsoft and AWS reportedly [PDF] scored subcontracts totalling $50m with the US Department of Defense to develop algorithms capable of recognizing objects from military drone footage after Google pulled out from the controversial Project Maven.

This is according to public subcontractor data trawled through by TechInquiry, a nonprofit investigating how big tech quietly works with the US government, first reported by Forbes.

Executives at the Chocolate Factory stopped working with the Pentagon on building AI software for the project back in 2018 after facing internal revolt from its own employees. Many Googlers did not want to be involved in the business of war. The Chocolate Factory, however, said the algorithms were just for object classification and not related to autonomous weapons.

Other big tech companies, however, appear to have no qualms working with the military, according to the report, which detailed Microsoft and AWS scoring $30m and a $20m respectively in subcontract awards in 2019 and 2020 to process footage captured by military drones. IBM also reportedly plays a small part, having won a relatively small $1.9m contract.

New large language model startup founded by ex-Googlers raises $40m

Canada-based Cohere says it has bagged $40m funding in a Series A round to build a large language model API.

These systems, based on the transformer architecture, generate text outputs given a few sentences or paragraphs. The technology was first popularized by OpenAI with its GPT-2 models and was later commercialized by the company via an API when it built the bigger and better GPT-3 neural network.

Competition is hotting up, however, and OpenAI won’t be the gatekeeper to large language model APIs for long.

“No one can get access to these high-quality natural-language models; those who can are the ones with access to a quarter-billion dollar supercomputer, so all the big FAANG (Facebook, Amazon, Apple, Netflix, and Google) companies,” Aidan Gomez, cofounder at Cohere, told Fast Company.

“What we want to do is foot the cost of that supercomputer and give access to all these organizations that otherwise couldn’t build products or features on this technology.”

The upstart is backed by big names like Geoffrey Hinton, an early pioneer of modern neural networks at Google and the University of Toronto, and Fei-Fei Li, a leading academic in computer vision at Stanford University.

Other similar startups hoping to take a slice of OpenAI’s pie include the Israeli AI21 Labs, who recently announced it had built a larger model with more parameters than GPT-3.

US starts new National Artificial Intelligence Advisory Committee

The US Department of Commerce says it has set up a panel of experts to work together and advise the government on all things AI, so that the country remains competitive.

“AI presents an enormous opportunity to tackle the biggest issues of our time, strengthen our technological competitiveness, and be an engine for growth in nearly every sector of the economy,” the Department of Commerce’s Secretary, Gina Raimondo, said in a statement.

“But we must be thoughtful, creative, and wise in how we address the challenges that accompany these new technologies. That includes, but is not limited to, ensuring that President Biden’s comprehensive commitment to advancing equity and racial justice extends to our development and use of AI technology. This committee will help the federal government to do that by providing insights into a full range of issues raised by AI.”

The National Artificial Intelligence Advisory Committee will be led by leaders in the field across industry, academia, non-profits, and federal laboratories. It will be looking at a wide range of issues, including ethics and bias to national security.

Roomba’s latest AI algorithms trained on images of dog excrement to avoid cleaning disaster

The latest model from the circular robot vacuum company, Roomba, has new software designed to recognize and skirt around piles of pet poo.

One woman recounted the horror story from when she woke up one morning to find her adopted pooch’s poop smeared all over the floor, according to CNN Business. Her Roomba device had accidentally trodden on the mess, and, in an attempt to clean the house, it had spread the smelly stuff everywhere it went.

Now the company said it has solved that problem with its latest j7+ model. The latest Roomba runs object recognition software that has been trained on numerous photos of dog poo (and electric cables). When the robot encounters these obstacles, it should, in theory, know to vacuum around them.

The company’s CEO, Colin Angle, said in order to collect the training dataset for j7+, Roomba collected photographs of dog poo and even collected and made fake images too, buying “all the realistic gag poop you can buy on the internet,” and molding Play-Doh models then painting them brown.

The j7+ Roomba costs up to $850 (£614), dog not included. ®

Leave a Reply

Your email address will not be published. Required fields are marked *