Artificial Intelligence, known as a miracle helping hand in the field of medical sciences. The screening and testing phases are the departments where AI seems to have a stronghold. This is because of their well-trained Machine learning models which boast professional experience in detecting medical conditions.
But being able to run such an AI-assisted product in the lab is quite different than generating good results in real life. In a rural Thailand clinic, Google learned this fact with practical experience.
— adafruit industries (@adafruit) September 9, 2019
Google Artificial Intelligence medical health: what went wrong?
Google Health is an advanced AI-powered deep learning engine that uses image recognition algorithms by capturing the patient’s eye and detect diabetic retinopathy. But even if the entire engine worked well in a theoretical lab setup. It produced no fruitful results in real life.
This is because of the tool, which has proved its inaccuracy in the practical world despite providing efficient results in the lab. This entire ordeal is frustrating for both, the patients as well as the nurses. This made the team take its failure deeply, and now they are more than determined to make this a better system with high efficiency.
The research paper gives an open insight into the tools used to augment the entire process. The traditional system of diabetes retinopathy requires at least 4-5 weeks of time before receiving final reports. However, Google was intending to provide ophthalmic expertise in a matter of seconds. In a control lab setup, Google was able to identify DR with up to 90% accuracy.
But in real life, many variations in conditions led to the downfall of the product. To be frank, the tool has too many dependencies, all of which cannot be fulfilled by people who are not in the field of research.
Google has cited the entire ordeal in an official blog which shows the exact conditions which caused its failure. But, this does not mean that Google will stop working on Artificial Intelligence products. AI still has a long way to go. Let us know what you think about this in our comment section!
Suryapratim Ray is an engineer, author, robotics hobbyist, and an active Quoran. Being a technical blogger, he covers the good, bad, and ugly of science on a regular basis at Sciencenews18. In addition to his passion for writing, he’s equally keen on learning classical music.