“Genius is one percent inspiration, ninety-nine percent perspiration.” – Thomas Edison

I like to think of myself as an “outdoors” guy, and I have spent a lot of time in the sun over the years, running and riding my bike. As I’ve matured, I find myself more aware of the risk that sun exposure poses to my skin and have become more diligent about scheduling annual checkups with my dermatologist. Perhaps this is why a recent news article caught my attention – Stanford students developed a machine learning program capable of using photos to evaluate and diagnose the risk of skin cancer. Their initial system matches the accuracy of human doctors’ assessments about 91% of the time.

I was really impressed, but I found myself thinking, “There’s probably more to this story than meets the eye.” The first four articles I found on the topic were relatively short, 4 or 5 paragraph stories on major news sites that focused on “what” had been accomplished or “what” the potential impacts to the industry would be. However, there just wasn’t any detail about the “how.” I felt like the students were probably getting short-changed. As I dug a little deeper, I found I was right.

Tracking down the original source, I finally learned just how much effort went into this project. Some of it was taking the Google Inception v3 CNN architecture and then tweaking it and tuning parameters, however, a surprising amount of work was required just to prepare the training and testing datasets. For example, the group had to:

  • Source 130,000 clinical images of 2,000 skin diseases
  • Develop a new, custom taxonomy to catalog all the images consistently
  • Convert image metadata to this taxonomy from multiple languages (German, Arabic, and Latin!)
  • Resize all images to common dimensions
  • Develop tools using metadata and other image analyses to collate related images (same lesion, different angles) and ensure they did not split across training and validation sets

Only after all that work could the real “machine learning” start. As expected, this was no simple or minor undertaking for a weekend; this was months of work for six students, plus a professor.

Practical Impacts

There is no doubt that AI and machine learning are hot topics in business right now. Leaders are striving to come up with ideas to develop automated and autonomous capabilities to out-compete their industry rivals. However, the sound bites and shallow journalism of many advanced-computing stories do a disservice to those making decisions about investing in these, and other, new technologies.

At Thought Ensemble, some of our clients fully understand the challenges they ask their teams to undertake when they kick off an innovation project. A few, though, only have their historical experience for guidance along with perceptions based on what they have read in the news or industry articles. We come to these projects after an executive tires of too many missed deadlines and is looking for a “project assessment and rescue.” Often, there is a hard message delivered back to them: the top-down goals and deliverables were never reasonable, even from the start. Then comes the effort to better understand the required work (with bottom-up input) and reshape expectations. Yes, tackling innovation efforts, whether bleeding-edge AI or just technology new to an organization, can be hard.

Let’s be clear – I am NOT suggesting that companies should forego exploring AI and machine learning solutions, or even that they should only be looking at the easy, already-packaged components. There are absolutely some fantastic opportunities out there, and innovative organizations have the potential to revolutionize their industries. But success will come under leaders who understand the real effort required for innovation and set their teams up with realistic timelines, sufficient resources, and appropriate deliverables.

Oh, and to media folks who might read this… thanks for sharing cool stories with us, like the AI cancer detection; they are inspiring and help motivate us. But if you can, show us a little more “how” to go along with the “what.” All those innovators out there should get the credit they deserve, and the ones that follow need to know what they’re really getting into.