AI Talk: Avatars, carbon footprint and manufacturing AI

June 14th, 2019 / By V. “Juggy” Jagannathan, PhD

This week’s AI Talk…

Avatars as a proxy for human contact?

A few months ago, the New York Times featured an article about the role of technology in mimicking human contact. A company called has an interesting approach to communicating and helping elderly folks. They have an avatar of a cat or a dog on an iPad interface that engages the elderly patients in a conversational way. Behind that interface is a whole army of healthcare workers around the world who engage in the conversation and provide 24/7 psychosocial support! From the patient standpoint, they are simply engaging a talking cat or dog with expressive faces. Speech from a patient sent remotely results in a chat text response from the remote healthcare worker which is translated into speech again on the patient side. The back and forth communication provides real support to this population. The article also notes that, while screen time is increasing in the poorer segment of society, the opposite is true for the rich. Interesting. I guess I must categorize myself in the first camp—my iPhone keeps saying my screen time is increasing!

Carbon footprint of Deep Learning

I ran across this blog post in MIT Technology Review last week. It had the alarming title of “Training a single AI model can emit as much carbon as five cars in their lifetimes.”  Well, it’s true. Deep learning has a large carbon footprint. And the model they compare it to is one that is quite popular these days: the Transformer model, with its hundreds of millions of training parameters. Transformer models are popular in the natural language processing community for a reason—they have been shown to be effective for a wide range of problems (review the full paper from University of Massachusetts). The authors’ message is two-fold: First, these models are not only costly in terms of carbon-footprint, but they are costly—period. Second, the sheer expense is leading to increasing privatization of research. The authors advocate for the creation of cloud services by the government (specifically for researcher use) and to explore more efficient (less costly) algorithms, along the lines of supercomputing resources available to the research community in the past. Google, however, has a different answer on the carbon footprint issue. The company proposed the transformer model. You can look at one of Google’s latest blogs on their use of electricity. They have been getting all their electricity from renewable sources—100 percent—for the past two years! The implied message from Google: We are using only renewable sources for energy and the carbon footprint for renewable energy is extremely low—so go ahead and train a transformer model on our Google cloud platforms. But of course you need to fork over a lot of greenbacks! Now the question is what are Amazon and Microsoft doing on this front? And will the government setup a cloud platform for academic researchers?

Another story on carbon footprints!

While we are in the topic of carbon footprints, I saw this article this week in Futurity. It proclaims that one swap of beef to chicken or turkey halves the carbon footprint of one’s diet!  It also says cutting down on food waste also helps with decreasing your carbon footprint. Well, I am a vegetarian, so I must focus my energy into not wasting food. Hmm, I guess I can also become a vegan!

AI and Manufacturing: another book review

I saw a reference to a new book in Microsoft’s blog: The future computed: AI and Manufacturing by Greg Shaw. It is an interesting promo for Microsoft technologies and a good collection of engaging customer stories. It includes use cases from wide ranging industrial domains—from predicting when an elevator is going to fail, to using AI to farm efficiently, to sustainable manufacturing. I found the chapter on retooling the labor force enlightening.  By analyzing the 600 million LinkedIn users’ skill profiles, the authors judged AI skills penetration in Manufacturing to be 5th behind IT, education, networking and finance. Health care was a distant seventh place. I wonder if they had considered us as part of the manufacturing bucket? Microsoft seems to have an impressive array of educational training opportunities—probably in line with this statement: “Often what is needed is more than a high school degree but less than a four-year bachelor’s degree.” European-style apprentice programs could be useful here in the U.S. Another idea is to have simulated environments: The lab is a public-private partnership in which workers learn new skills inside a simulated smart factory.” Other parts of the book are a bit of a rehash from other material I have seen from Microsoft related to ethics and fairness. A good collection of references, though I must admit I didn’t review those!


My colleague Anna Abovyan pointed me to the article in the New York Times.

I am always looking for feedback and if you would like me to cover a story, please let me know. “See something, say something”!  Leave me a comment below or ask a question on my blogger profile page.

V. “Juggy” Jagannathan, PhD, is Vice President of Research for M*Modal, with four decades of experience in AI and Computer Science research.