Why Google’s AI Failed In Medical Diagnosis
Google is one of the most famous brands to have ever existed. Every day, millions of users input countless search inquiries into Google, ensuring its resiliency as a household name is maintained. These days, Google is also getting involved in a myriad of high-tech pursuits like the development and perfection of artificial intelligence for a diverse set of purposes, including handling medical diagnoses that have historically been managed by people. Early lab results were immensely promising, but recent results from real-life show that Google’s AI has many hurdles to overcome before it can truly be relied upon.
Why did Google’s AI fail in medical diagnosis focused on the real world, and how is the company vowing to do better? Here’s why the future of medicine may not be turned over to the robots as quickly as some think.
The AI fell short
There’s no denying that Google’s recent effort to employ AI in the diagnosing of real-life people fell far short of what many artificial intelligence proponents expected. According to TechCrunch’s excellent breakdown of the failures, Google Health’s deep learning system was supposed to scan images of the human eye to identify the presence of diabetic retinopathy, which can cause vision loss. While the system performed well in test runs, its deployment in the real world proved lackluster, as the deep learning system proved incapable of assessing many of the images sent to it.
Many previous studies have found that AI is equal with humans when it comes to diagnosing issues with patients. The vast majority of these studies have been under lab conditions which are very dissimilar to the real world, however, and are thus often incredibly misleading. Real life situations, such as inclement weather, poor operating conditions, peculiar lighting, and workers who aren’t super familiar with the technology in question can all impact AI’s ability to actually achieve its intended goals.
In other words, when you next pursue an Express MRI, consider relying on human expertise instead of a robot. These AI systems may perform splendidly under ideal laboratory conditions, but the real world of medical pain and human suffering is ugly, complicated, and unlikely to become easier for machines to navigate anytime soon. This isn’t to say that AI is worthless, or that it won’t get better, but simply to assert that many miraculous predictions of AI’s current might and capabilities aren’t taking enough real-life scenarios into consideration.
The problem isn’t only one of AI, but also one of engineering.
Engineering failures will persist in the near-future
Google may be able to design an impressive AI that can match humans when it comes to certain diagnostics, but engineering artificial intelligence equipment that can consistently be relied upon in real world scenarios is another task altogether. Engineers need to be prepared to produce products that can withstand a wide range of physical conditions and be wielded by individuals who may lack the savvy tech expertise that Google’s lab workers possess. That’s going to be a time-consuming process that’s also liable to be incredibly expensive.
Google’s recent failures stemmed, in part, from a failure to get images to upload to the deep learning system in the first place. Many of these failures came from the fact that the image quality was lackluster, but other issues had to do with human input that wasn’t as ideal as that which we see in lab conditions. That problem isn’t going away anytime soon, so engineers will have to navigate around it for the foreseeable future.
None of this is to say that AI doesn’t still have a very promising future. Google has relied upon AI with great success elsewhere, for instance, especially when it comes to using AI to fight spam. Healthcare professionals in hospitals are also eager to use AI to diagnose patients when they’re feeling overburdened by work, which is a huge problem right now with the public health pandemic. In the forthcoming years, however, we should be cautious when greeted with claims that artificial intelligence is coming to replace human healthcare workers – the real world is far more complicated than laboratory testing, and AI and human errors alike will stymie our progress every step of the way.
The post Why Google’s AI Failed In Medical Diagnosis appeared first on The Startup Magazine.
Discover Past Posts