This audio was created using Microsoft Azure Speech Services
It is fairly easy to conceptualize that computers can compare, analyze and compartmentalize data according to rules dictated by a human created program. But, this is just a small part of what computers can now do. Digital machines have been built that actually learn to make decisions about new situations that exceed the experience of the initial programming. The science of getting computers to learn beyond data specifically programmed is called Machine Learning (ML). For example, a machine can watch a person tidy up a house and then appropriately clean up a different house with a different mess. Building machines with this type of decision-making capability underlies the concept of artificial intelligence.
The secret of ML is based on mathematical equations called learning algorithms that can consider an infinite number of attributes, or cues, to make predictions. This process mimics the neural networks in the human brain. Interestingly, early ML tools still widely used, such as the decision tree, were developed in the 1960’s by a cognitive psychologist who was interested in understanding how humans make decisions.
Two basic categories of ML with somewhat alarming titles, at least to those of us with children or pets, are Supervised Learning and Unsupervised Learning.
The most common problem solving programming is supervised learning, where continuous outputs are created based on the information given by pre-determined classifications. The classifications are used to determine the error of the network which is adjusted by an ongoing feedback loop. Today, computational biologists, aviation controllers, and data centers designers use this type of programming.
A second type of ML is unsupervised learning. This is used when anticipated answers are not based on historical outcomes. One outcome from unsupervised learning is called clustering where the goal is simply to find similarities in the data. Because nothing is known in advance and there is no “right” answer, the machine is not told what to look for in the data. Instead, new patterns or clusters are revealed. The clusters discovered might match an intuitive classification, or they could reveal new patterns that we have not suspected.
A rapidly learning and unsupervised computer could help hospitals identify:
- Nodes of voice confusion in the operating room or the pharmacy
- Novel sound control strategies in the patient room
- Unexpected building designs and management associated with lack of staff safety, employee under-productivity or decreased student performance
- Distant weather patterns or land management that brings microscopic pathogens into the building
- Unknown information necessitating new building codes
- Medical knowledge from electronic health records that could dramatically increase our understanding of an entire disease process
- Predictive energy modeling and continuous energy efficiency opportunities
- Preventative building maintenance needs
What new safety solutions can you envision ML bringing to your facility? Let us know in the comments below!
