digests /

Last Week in AI #72

Deepfakes for good, AI's carbon footprint, and more!

Last Week in AI #72

Image credit: Rebecca Heilweil / Vox

Mini Briefs

How deepfakes could actually do some good

Deepfakes, synthetic media that are often indistinguishable from the real thing, are arguably one of the scariest technologies available today. Indeed, by being able to create videos of politicians that could seem real, mimicking voices and appearances, deepfakes have the potential to spread massive amounts of disinformation. While the cat has long been out of the box on deepfakes, there may be some small redemption in finding positive uses for the technology. A new HBO documentary, Welcome to Chechnya, provides such an example. Chechnya’s LGBTQ population has faced “significant persecution, including unlawful detentions, torture, and other forms of abuse.” While survivors cannot safely reveal their identities, deepfake-like technology allowed the makers of Welcome to Chechnya to overlay volunteers’ faces on top of survivors’ faces to allow those survivors to speak out about their experiences. Just as deepfakes can allow the film to “shine light on human rights abuses while minimizing the risk for victims involved in the production”, they may have other uses that could help bring attention to the unfortunate, and sometimes horrid, issues that many of us remain unaware of.

AI’s Carbon Footprint Problem

Among the adverse side-effects brought on by the era of deep learning has been an immense carbon footprint. The massive amount of compute it takes to train deep learning models, compounded by the need to iterate on models and reproduce experiments, does not come without cost: a 2019 MIT Technology Review article notes that training a single AI model can emit as much carbon as five cars in their lifetimes. With so much attention being paid to the climate crisis, AI’s environmental impact has come under scrutiny as well. In efforts to stimulate change, a movement called “Green AI” seeks to prioritize leaner, less power-hungry AI models that can achieve similar performance on other benchmarks. But progress requires measurement: by how much can we reduce AI’s carbon footprint? To that end, a team of researchers from Stanford, Facebook AI Research, and McGill University has developed a tool that measures the carbon emissions of a machine learning project. As machine learning systems become more ubiquitous, they will inevitably represent a greater share of carbon emissions. Tools such as this one, paired with a commitment to including low carbon emissions as an objective for machine learning systems, will become more and more important in the future.


Check out our weekly podcast covering these stories! Website | RSS | iTunes | Spotify | YouTube


Advances & Business

Concerns & Hype

Analysis & Policy

Expert Opinions & Discussion within the field

  • Reflecting on a year of making machine learning actually useful - For those of you who don’t know my story, I’ll give you the short version: I did machine learning research for two years, decided not to get a PhD at the time, and became the first machine learning engineer at Viaduct, a startup that provides an end-to-end machine learning platform for automakers.

  • Yann LeCun Quits Twitter Amid Acrimonious Exchanges on AI Bias - This is an updated version. Turing Award Winner and Facebook Chief AI Scientist Yann LeCun has announced his exit from popular social networking platform Twitter after getting involved in a long and often acrimonious dispute regarding racial biases in AI.

Awesome Videos

That’s all for this week! If you are not subscribed and liked this, feel free to subscribe below!

More like this
Follow us
Get more AI coverage in your email inbox: Subscribe