AI Talk: Trends, lawsuit, word associations and black or white?

July 12th, 2019 / By V. “Juggy” Jagannathan, PhD

This week’s AI Talk…

AI Healthcare Trends

Last month, Forbes featured an article summarizing the trends in health care impacted by AI. The author, Terence Mills, is the CEO of AI.io—a company that claims it is building white box AI. That term, however, is bit misleading. White box is a term that is used to refer to explainable AI—however, here it is simply referring to intelligent apps. The general AI market is set to balloon from $25 billion in 2018 to $200 billion by 2025! According to Mills, four major trends are germane to health care: First, large numbers of people suffering from chronic conditions are being helped by AI-based-smart apps. This trend is helping put patient’s in charge of their health—a patient-centered approach, which is a good thing. Second, the amount of data being generated annually is set to double every two years and currently stands at 4 trillion gigabytes! All this data is a rich source for data mining and AI. This is leading the way for personalized medicine. Third, the area of medical imaging is being transformed by AI. Every imaging modality is now being screened by AI algorithms. Lastly, the area of communication between patient and physician and physician with other caregivers is getting attention from AI solutions. Reasonable characterization of trends!

Lawsuit for using EHR data

Last year, Google published a landmark paper analyzing 46 billion data points from EHR data to predict length of stay, readmissions, mortality, etc. A year later, there is a lawsuit against the University of Chicago and Google, charging that patient consent was not obtained and claiming Google can easily identify the patients, even though the records were de-identified. This class action lawsuit currently has all of one patient behind it, but it is not an unusual tactic, apparently. Of course, Google and the University of Chicago defend their right to use the de-identified data to benefit patients and research. A reminder that we live in a litigious society and a reminder that all the abuses by tech companies are coming back to haunt them.

Word associations to new knowledge

In 2013, Tomas Mikolov of Google wrote a landmark paper on technique dubbed “Word2Vec.” Using a completely unsupervised learning technique, this algorithm crunched millions of text documents and learned to represent each word as a vector of numbers—also called its embedding. The representations learned were so powerful that one could solve analogy problems like man -> women; king -> ? (the answer came back as queen). And similar topics can be gleaned by looking at the angle between their vector representation—referred to as the cosine similarity. The embeddings research is its own active area of research now, with advancements happening almost daily. I was pleasantly surprised to see reference to this technique in the MIT Technology Review, applied to scouring scientific literature to provide new insights. Last week, Nature published new insights after applying the Word2Vec algorithm. The work was done by a researcher associated with the Lawrence Berkeley National Lab and Google (where else?) running the algorithm on 3.3 million scientific abstracts published between 1922 and 2018 related to material science. What is the big deal here? Here is a quote from their paper: “A ranking of thermoelectric materials can be produced using cosine similarities of material embeddings with the embedding of the word “thermoelectric.” Highly ranked materials that have not yet been studied for thermoelectric applications…” Yes, they managed to identify new materials that have not yet been studied for a specific property! All by running this algorithm on journal abstracts!

Black or white?

No, I am not referring to the famous Michael Jackson song! The question is: Should you wear black or white on hot summer days? That’s indeed a pressing question in the middle of summer! This great Wired article has the complete scoop on this wardrobe issue. Before you jump to the conclusion that it must be white, read the article. Bedouins in the middle of the desert apparently wear black robes! So, what gives? Well, the answer (spoiler alert if you are planning on reading the article), white is still the best to wear on hot summer days—if it is tight fitting like a t-shirt. If you want to wear black, make sure it is loose fitting like a robe. Who knew?

Acknowledgements

My colleague Zach Taft pointed me to the lawsuit article.

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 an AI Evangelist with four decades of experience in AI and Computer Science research.