AI Talk: Year-in-review top 10

December 17, 2021 / By V. “Juggy” Jagannathan, PhD

This is my last blog for 2021, so I thought it would be appropriate to pause a beat and look back. What an eventful, tumultuous year. Here, I hope to summarize various trendlines and advances that happened in 2021. Just for fun, I have organized this blog as a top 10 list, David Letterman style!

  1. Regulating algorithms. Regulation, as it pertains to artificial intelligence (AI) algorithms, was a big topic in 2021. Given that AI has permeated every aspect of society, it is no wonder there are a range of multifaceted regulation efforts. Below, I discuss regulations in three separate buckets—general approaches to algorithm regulations, specifics surrounding big tech and health care. I have done a number of blogs on this topic starting from the beginning of this year. This recent Harvard Business Review article provides a thoughtful overview of what companies need to do to prepare for regulations. Basically, build systems that avoid bias and discriminatory decisions, study the impact of outcomes of decisions and ensure explainability and transparency of the decisions reached. Explainability is an extremely active area of current research in machine learning (ML). Current regulatory guidance is based on rules related to anti-discrimination, but more is certainly underway in the U.S., Europe, China and worldwide.
  1. Regulating big tech. A hot topic in the news when Facebook whistleblower Frances Haugen’s revelations shocked the nation’s conscience. Then there were the revelations surrounding mental harm caused by Instagram posts. Underlying all these social media platforms are AI algorithms that decide which posts to promote, what advertisements to show, etc. Lawmakers are contemplating changes to Section 230 of the Communications Decency Act that shields platforms from lawsuits stemming from content posted by subscribers. Undoubtedly there will be lot of policy change efforts by Congress in 2022—what form they will take is anybody’s guess. This recent New York Times article compares the work done by Congress to regulate social media to the effort to curb big tobacco and reveal the harmful effects of smoking!
  1. Regulating AI in health care. Two regulations come to mind on this front:

    During the COVID-19 public health emergency (PHE), telehealth services were covered and paid as normal inpatient visits. States loosened licensure restrictions and allowed physicians to practice across state lines. These changes turbo-charged a sleeping industry. The Centers for Medicare and Medicaid Services (CMS) has signaled that telehealth services will continue to be supported next year. Telehealth has spawned an AI and tech-driven remote monitoring industry as well. There are lots of startups in this area—take a peek at CBinsight’s report on this front.

    Another regulatory step was one taken by the U.S. Food and Drug Administration (FDA). The FDA issued guidance for AI and ML in their Software as a Medical Device (SaMD). This recent PEW report provides a fairly detailed analysis how the FDA regulates AI and ML embedded in devices. One area where AI is used often is in clinical decision support. The FDA has signaled recently that they will start regulating this area starting in 2022. About time!

  1. Wearables: When you hear the term wearables, you probably think of Fitbit and Apple Watch. But no, there is an entirely new category of devices which appeared in the landscape this year. Smart (AI-enabled) clothes that can convert athletic apparel into monitoring devices are entering into this space! This site lists 15 smart clothing apparel brands that you can buy now. Another category of wearable is hearable, which I covered in one of my blogs this year. This category includes smart ear buds which can monitor various functions in your body.
  1. Digital twin: Fundamentally, a digital twin is a high fidelity simulation of reality. It can be applied to study supply chain disruptions, or it can be applied to health care to personalize care. This recent VentureBeat article talks about 21 ways digital twins will transform health care.
  1. Data woes: “Data is the new oil”—a phrase coined in 2006 by Clive Humby and is eerily true after a decade and a half. ML algorithms have an insatiable appetite for data. In this blog post from a few months ago, I highlight various approaches to address the data paucity issue that practitioners currently face.
  1. Deepfakes: The alarming rise of deepfakes and their malicious impact has been well publicized. However, there is one story that surfaced recently showing this technology can be a power for good. Like many hundreds of millions, I am a huge fan of the Beatles. The latest documentary on Disney+ chronicles the famous group during their last hurrah in 1969. The movie, “The Beatles: Get Back” was edited from 150 hours of recordings filmed in 16mm format. Peter Jackson, the director of the Lord of Rings series, was hired to create the documentary. The technology used to restore the old film clips? Deepfake.
  1. Future of health care: I wrote a blog on this topic in the summer based on the book by Nicholas Webb: “The Healthcare Mandate”. His basic message is advocating for a “big shift”—which refers to a refocus on disease prevention and promoting wellness, as opposed to treating preventable diseases. How can we achieve this vision? AI, a population management dashboard and continuous remote monitoring of what he calls “constituents” in his “Constituent Healthcare Operating System.” You are not a patient, if you are not sick—hence the term constituent. I cannot agree with his vision more!
  1. AI and climate change: Not a day goes by when we are not reminded of the effects of climate change—the recent devastating tornadoes is just one example. This past summer the United Nations issued a detailed report on climate change. In my blog on this topic, I highlight various ways AI can be put to use to counter the effects of climate change—based on a report done by Forbes. I highlight two areas where AI is already being used: predicting and preventing forest fires and precision agriculture and vertical farming.
  1. Brain and language models: To me, and I am biased, the discovery that ML-based language models are quite similar to the human brain in how they process language tasks is groundbreaking. It informs us of the limits of such ML models and guides us to look elsewhere to create thinking and reasoning capabilities. Just last week, DeepMind released fresh results on advances in language modeling. Undoubtedly, we will see a lot of advances on both language models and brain models in 2022.

Happy holidays and see you all in 2022.

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

V. “Juggy” Jagannathan, PhD, is Director of Research for 3M M*Modal and is an AI Evangelist with four decades of experience in AI and Computer Science research.