Machine learning startups are on a mission to make the world healthier and safer
Entrepreneurial startups are working to harness the power of artificial intelligence (AI) and machine learning to solve vexing social issues, ranging from worker safety to disease diagnosis. The socioeconomic impact of AI is already being felt, from improving health and safety to making online shopping experiences more engaging.
Giving doctors more medical expertise
The business of machine learning isn’t just business, it’s personal for Yosi Taguri, founder of MissingLink. Several months ago, his mother suffered from an aggressive form of non-Hodgkin’s lymphoma.
“I was looking for a doctor that would be able to look on that CT scan on MRI and tell me what do we see here?” said Yosi. “I was willing to pay any amount of money in the world just to have 10 minutes with a doctor.”
The problem Yosi was facing is too common: a shortage of expert medical professionals who can diagnose complex diseases based on large datasets, from images and a variety of test results. “My mission, and our mission at MissingLink, is to empower companies that change the world to actually make it much, much better,” he said.
In a similar fashion, the team at Iterative Scopes is focusing on improving the quality of colorectal cancer diagnosis, especially the ability to identify precancerous polyps. Evan Wlodkowski, who leads regulatory affairs and compliance at Iterative Scopes, points out that about one-quarter of precancerous polyps are missed during screening.
Iterative Scopes is a startup that develops software to detect precancerous polyps in real-time. They are working to address two key challenges: colonoscopy procedures are not usually recorded and the need to make inferences in real time. When the product launches in 2020, it will also include a data analytics platform that integrates electronics and health record information with images that provide more data for a wider array of analysis.
Machine learning algorithms will be able to combine insights derived from images with structured health care information to identify medical insights that are not available from only one type of medical data.
Discovering new drugs faster
On average, it costs $2.6 billion and more than 15 years of development for a new drug to come to market. The result is that not as many new drugs are under development, and those that are being created aren’t being delivered to patients quickly enough.
To solve this, DeepCure is trying to reduce the cost of developing new drugs by using machine learning which molecules potentially could be used to treat various diseases. During his presentation, DeepCure founder Kfir Schreiber demonstrated how his company was able to analyze a billion molecules to find those which could be used to treat the hepatitis C virus.
Using machine learning algorithms DeepCure found molecules that were already being used in treatment, but it was able to do so much more quickly than using typical drug discovery methods.
“We realized that to get to the same point that we reached in just a couple of weeks took roughly two years for the pharmaceutical company,” Kfir said.
An ounce of prevention
In addition to helping doctors diagnose and treat injury and illness, AI promises to help prevent injuries and illnesses. Consider the construction industry. In the United States, 20 percent of workplaces deaths are in the construction industry and the number of deaths has been increasing steadily since 2007, largely because of increased construction.
Sean True, Director of Machine Learning at SmartVID.io, has a plan for turning this statistic around. SmartVID.io develops machine learning-based video analysis software that analyzes construction site videos to identify unsafe practices and dangerous situations. Although construction companies have long collected monitoring videos, they have not had the resources to analyze the large volumes of them efficiently.
For example, Sean showed an instance of a worker not wearing gloves in a construction site. This is a problem because hand injuries are the top injury on construction job sites. These kinds of injuries are not life-threatening, but they can be costly, resulting in lost weeks of work and high medical bills. Without AI to detect these situations in monitoring videos, it is difficult to identify work sites that need to address safety practices.
Improving business efficiency
In another session, David Morczinek, CEO and co-founder of Airworks, discussed how machine learning is being used to increase the speed and drive down the cost of assessing the cost and risk of construction jobs.
Much of the work in construction planning is the responsibility of hundreds of thousands of civil engineers who “are assessing, planning and estimating construction sites every single day,” said Morczinek. “Their main job is to analyze the future construction project, and to create the design.”
Civil engineers need data for their work and it is not the kind of data that can be looked up online, it must be collected in the field. “The current state of the art is to take measurements for each point, point by point with an optical instrument that’s mounted on a tripod on the ground,” he said. “If you look at how this has been done 100 years ago, it doesn’t look much different.”
Airworks is realizing radical efficiency gains. While a human can collect 500 measurements per day, an aerial drone can collect 20 million data points per day. Collecting data from the field is not even the biggest cost. A major expense is the cost of human reviewing images and labeling objects such as roads, buildings, and sidewalks. The labeling work a human can do in 40 hours for can be done in 30 minutes with Airworks software.
AI is also making the world a better place by improving the efficiency of some of the costly things we do.
Cutting costs for consumers
AI is also being used to improve efficiency for consumers, such as streamlining travel planning. Even seasoned travelers may find it hard to know the optimal time to purchase flights. This is where Hopper, a travel app, offers a solution. It recommends when to purchase flights using patterns discerned from analyzing massive amounts of flight pricing data.
“There’s a kind of general anxiety there where we have users who want to make sure that they’re getting a good deal on their flight,” says Matt Dinallo, Data Scientist at Hopper. “And on the other end of the spectrum, flight prices themselves are actually quite volatile, even over a longer period of time.”
Using 15 trillion price quotes over a five year period, Hopper is able to make predictions about prices because of the seasonality patterns within routes. Hopper uses a type of machine-language programming called sequence-to-sequence learning. In this case, for example, it learns to map a sequence of dates to a sequence of prices. Thanks to its ability to process massive amounts of pricing data, the software can predict the prices of a flight each day up to the departure date.
Companies like SmartVID.io, Iterative Scopes, Airworks, and Hopper are developing AI-driven solutions to harness data to improve health and safety. Building AI is an engineering challenge that requires specialized tools. “You cannot run 500 experiments in parallel because it’s simply impossible,” says Yosi. “So we created a company called MissingLink that basically automates that process.”
Watch the presentations on these technologies and learn about other applications of AI at the MissingLink.ai blog. To find out about upcoming Samsung NEXT and MissingLink.aievents, sign up for our weekly newsletter.