Stanford University School of Engineering
Developing machine learning capabilities will require heavy investment and the cultivation of a generation of developers with a background in data science.
Machine learning and artificial intelligence were the stuff of science fiction when an intelligent computer turned on its creators in 2001: A Space Odyssey. Fifty years later, intelligent algorithms are beginning to reshape many facets of health care, education and commerce – and that process is just beginning, says Jia Li, the head of R&D at Google Cloud AI.
“But machine learning development is a very complex and resource-consuming process. It will require investment and expertise in every single step: Collect the data, design a model, tune model parameters, evaluate, deploy it, and finally update and iterate the entire process,” Li said during her presentation at this year’s Women in Data Science (WiDS) conference at Stanford University.
AI, or artificial intelligence, has the potential to improve the outcome for patients and help clinicians make better decisions, she says. In a sense, AI can help medical teams connect the dots. AI could suggest guidance on everything from patient lifestyles to medications and provide automated monitoring and early assessment of critical conditions by noticing subtle signals that a human would not be able to detect.
Studies have shown that 10 percent of thoracic patient deaths are related to diagnostic errors, and 4 percent of the 400 million or so radiological interpretations conducted each year in the U.S. contain clinically significant errors. Machine learning could improve those outcomes, but developing and training the software is quite challenging, Li says.
Building the models needed to make the software accurate requires board-certified radiologists to label and classify the information in those X-rays, a costly and time-consuming process. Li says that she and other data scientists are working to develop models that are less labor intensive.
In education, artificial intelligence algorithms could help customize courses for individual students based on their past experience, strengths, weaknesses and personal preferences, Li says. AI could free up teachers to work with students by automating chores such as homework and exam assessment.
Although AI and machine learning are hot topics, there are only about one million developers who have a data science background, and far fewer with a background in deep learning, Li says. Google, she says, has a partial solution to the dearth of qualified AI developers: Cloud AutoML is a suite of products that enable developers to train high-quality machine learning and AI models even if they lack expertise in those areas.
Source
