From 3M Health Information Systems
AI Talk: 100th blog – A look back
This blog marks my century mark with the 3M Inside Angle team – all in a little over two years! It has been a remarkable period for several reasons, the most notable of course being the global COVID-19 pandemic that still has most of the world in its throes. In my 100th blog, I want to highlight some of the notable AI-related news that has happened in the past two years.
My inaugural edition in March 2019, had a story about GPT-2 from Open-AI. This is a transformer-architecture-based language model (a language model is a model trained to predict the next word given the left context). This is the same technology that constantly corrects your text messages in your phone. In late 2018, we saw the entry of BERT, a transformer model that marked a turning point in how we approach natural language understanding tasks. GPT-2 and BERT marked the evolution of language models and are the core underpinning of what became known as transfer learning in the past few years. The basic idea is that when we train a large language model on the vast amount of content available on the internet, we can then turn around and use this pretrained model and fine tune it on specific tasks like sentiment analysis, language generation, summarization, etc. We now have GPT-3, the largest language model ever built to date, that can generate flawless prose, software code and even produce a summary of itself, GPT-3.
Regulatory approvals for radiological image analysis using deep learning started appearing about two years ago. Now there are scores of such applications that have approval for deployment in practice, from X-ray and MRI images to CT scans. Along with this growth in the application of deep learning has come the awareness that black-box models for diagnosing images are not enough. These models must be explainable to garner trust which is an ongoing research endeavor.
Another explosion in technology that we have seen is in wearables and remote monitoring. This started few years back, but COVID-19 turbocharged this vector. From Fitbit to Apple Watch, to devices that will check your ear for infection, and monitor your diabetes and heart rate using electrocardiogram (EKG) technology. This list continues to grow at an exponential rate. The trick is to devise algorithms and real world solutions that make effective use of this data to provide patients and users with tangible benefits.
Telehealth has been around for decades, but it came into its own last year with the huge increase in both telehealth technology and its adoption as a result of the COVID-19 pandemic. Here is a blog I wrote about telehealth before the pandemic.
My first of many COVID-19-related blogs was in March 2020. In total, I wrote 20 blogs on COVID-19 over the past year, with topics ranging from mRNA vaccines, tech-based contact tracing, disease spread modeling, social determinants of health and how COVID-19 impacted vulnerable populations and, more recently, the different hidden ways we dealt with the pandemic as detailed in a book by Michael Lewis.
IBM’s Project Debater took the stage two years ago. Robots were on the forefront of a number of news events—from comedy and drone delivery to inventory managers and care giving bots.
Many of my blogs also spotlighted the problems with machine learning:
- The use of underpaid online workers to help create training data for insatiable machine learning models. The practice, dubbed ghost work, is still under scrutiny and several researchers are building tools to help online workers better gauge whether they should bid for such work.
- Deep fakes have been a plague and in particular misinformation and conspiracy theories propagated through use of this technology is quite abominable.
- Bias in AI systems has seen growing relevance in the past two years. The term “algorithmvigilance,” was coined, referring to the vigilance we all must have as we guard against the unintended consequence of machine learning models.
Indeed, an eventful two year period that has been an exciting ride. I am reminded of a book review I did a year back: The future is faster than you think by Peter Diamandis. I think he is right.
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.