In recent discussions on the progress of AI, a great deal of focus has been placed on model efficiency, which refers to “reducing the compute needed to train a model to perform a specific capability.” While methods such as deep neural networks and neural architecture search have produced great improvements according to metrics like accuracy, they use vast amounts of compute. This raises a number of issues, including the environmental impacts of using so much compute power and the inequitable access to the compute needed to reproduce research that employs these methods.
In the spirit of aiding efforts to introduce model efficiency as a salient objective, OpenAI announced on May 5 that it would begin tracking machine learning models that achieve state-of-the-art efficiency. By publicly measuring model efficiency, OpenAI hopes to paint a quantitative picture of algorithmic progress, which it believes will in turn “inform policy making by renewing the focus on AI’s technical attributes and societal impact.”
Deep reinforcement learning has been, more or less, the “superhero” of recent advancement in AI. We’ve seen it accomplish incredible feats, from beating the world Go champion to teaching robots to walk. But can the secret ingredient to beating video games help run the economy? Scientists at US business technology company Salesforce have developed a system called the AI Economist that uses reinforcement learning to identify tax policies that maximize productivity and income equality for a simulated economy. While still rudimentary, the scientists hope the tool can provide a sanity check for existing economic models, and, in the future, replicate existing theoretical results.
Advances & Business
How A.I. Steered Doctors Toward a Possible Coronavirus Treatment - Specialists at the London start-up BenevolentAI helped identify the arthritis drug baricitinib, which is now part of a clinical trial. In late January, researchers at BenevolentAI, an artificial intelligence start-up in central London, turned their attention to the coronavirus.
An AI algorithm inspired by how kids learn is harder to confuse - Drawing inspiration from the way parents teach children to identify things by introducing increasingly finer categories, researchers at Carnegie Mellon University created a new technique that teaches a neural network to classify things in stages.
Written And Directed By: Artificial Intelligence - As impossible as it seems, it won’t be long before artificial intelligence is writing and creating films. As a lead up to this eventuality, here is a list of just some of the creative endeavors that AI has already accomplished.
Concerns & Hype
AI and the Far Right: A History We Can’t Ignore - The heads of two prominent artificial intelligence firms came under public scrutiny this month for ties to far right organizations.
Clearview AI to stop selling controversial facial recognition app to private companies - Controversial facial recognition provider Clearview AI says it will no longer sell its app to private companies and non-law enforcement entities, according to a legal filing first reported on Thursday by BuzzFeed News.
AI Software Gets Mixed Reviews for Tackling Coronavirus - Hospitals hope technology can ease pressure on doctors facing the pandemic, but some have noticed errors.
France is using AI to check whether people are wearing masks on public transport - France is integrating new AI tools into security cameras in the Paris metro system to check whether passengers are wearing face masks.
Pandemic Robots Deployed in Singapore Parks to Remind Humans of Their Own Mortality - The Boston Dynamics robot known as Spot has been deployed to a park in Singapore to remind people they should follow social distancing guidelines during the pandemic, according to a new report from Singapore’s top government technology agency.
Expert Opinions & Discussion within the field
Yann LeCun and Yoshua Bengio: Self-supervised learning is the key to human-level intelligence - Self-supervised learning could lead to the creation of AI that’s more human-like in its reasoning, according to Turing Award winners Yoshua Bengio and Yann LeCun.