Published by Peter Lin on
Oct 30, 2018 12:00:00 AM
It's 2018...where's my smart robot?
Machine learning is happening, but not as fast as anyone anticipated. How can we accelerate progress?
The concept of “machine learning” means slightly different things to different people. Without getting supremely technical, we consider machine learning to be the science of progressively improving computer system performance without explicit programming. For having such futuristic connotations (did you ever watch the movie A.I. Artificial Intelligence? Talk about scarring), machine learning is surprisingly pervasive – you encounter it daily without even realizing.
Machine Learning is Challenging
While machine learning is being applied in our day-to-day lives, many wonder why we haven’t progressed further. It’s 2018 and no one has a near human-like robot doing their manual labor.
The short answer? It's challenging. Progress was hindered in part by PR hype, which led to an “AI winter” in the ‘70s and ‘80s. A key factor preventing major strides in machine learning is hardware. Until Nvidia created fast video cards and tools to perform billions of calculations per second, training a computer took months, if not years.
Going back several decades, the USPS began using hand writing recognition to sort mail. An impressive feat, if you’ve ever seen some people’s sloppy penmanship. Today, we use such technology regularly, but probably wouldn’t consider it to be true "machine learning."
Consider an average day, where you might wake up and ask Alexa, “What’s the weather outside?” and she will spout off a forecast while you select the appropriate outer layer. Then you grab your iPhone X, use your face to unlock your phone, and dash off an email before punching in the location of your first meeting to Google Maps. Between the voice recognition, facial recognition, and GPS navigation, you’ve already immersed yourself in a world of AI…without even realizing it.
Where is Machine Learning Headed?
In the near future, machine learning will be used in a wide swath of applications across industries:
Car manufacturers will move towards autonomous vehicles
Diabetic research will advance with retinopathy
Farmers will take advantage of crop disease recognition
Medical researchers will progress drug discovery
Consumer goods providers will come out with higher tech games and better smartphones.
The caveat? It will take a good ten years of failure to get there. After that, things will settle down and move at an even pace.
Get Your Developers Involved
While there are a lot of lofty applications, practical uses for machine learning are everywhere. As no single individual can keep pace with all the research out there, it’s important for developers to share knowledge and advance together. Companies should create a sandbox environment where their developers can play with Google Tensorflow and PyTorch and encourage their team to join those communities. Understanding the tools and setting up an environment takes several months, but creating Virtual Machine images is a great way to reduce the startup cost. The most important takeaway is to start – even if it’s small.