From healthcare to agriculture to energy, AI has the power to radically transform the biggest industries on earth. Speaking on Tuesday at the EmTechDIGITAL 2016 conference at the St. Regis Hotel in San Francisco, a panel of business leaders explored how, and why, AI can make an impact.
The panel included Anthony Goldbloom, co-founder and CEO of Kaggle; Naveen Rao, CEO of Nervana; Tye Brady, chief technologist of Amazon Robotics; Erik Andrejko, vice president of science for The Climate Corporation; and Colin Parris, vice president of GE Software Research–a mix of startups as well as business giants. Here’s how each company sees AI changing each industry.
Goldbloom is head of “The world’s largest community of data scientists”–to be exact, there are more than 550,00 data scientists involved, which is a quarter of the world’s total population. The startup hosts AI competitions, from cancer detection to school performance, to evaluate machine learning platforms. First, training data is used to train algorithms, then the algorithms are applied to test data. Kaggle then evaluates how successful the solutions are.
“The way Kaggle runs machine-learning competition is you can take the problems, put [them] up on the web, and data scientists can download problems and their solutions,” said Goldbloom. There’s a scoreboard of all the competitions, where you can see your rank.
SEE How to prepare your business to benefit from AI (TechRepublic)
“This gives us a clear sense of what’s possible at the edge of machine learning,” said Goldbloom.
Goldbloom gave the example of using AI to identify promising chemical compounds for drug discovery. “The pharmaceutical industry needs solutions,” he said. By using “rules” to let machine learning algorithms “loose,” different solutions can be discovered, more than what humans can come up with.
Another impressive way AI can solve problems is by predicting seizures through EEG readings. One machine learning system figured out whether a seizure would happen an hour ahead of time in 80% of cases.
Kaggle has conducted over 300 machine learning competitions. “Before they were brought to us, I didn’t think the machines would do a good job,” said Goldbloom. “But these are areas I’ve been surprised at what machine learning is capable of.”
“Humans have no chance at the EEG challenge,” said Goldbloom.
The key to the success of AI? “Any profession with a lot of repeated work will be changed fundamentally by AI,” said Goldbloom, “if it exists at all.”
Naveen Rao, founder of the startup Nervana, wants to provide a platform for companies to incorporate deep learning. Rao’s models are meant for companies to train on their own data, like image, video, text, and speech, through a deep learning framework.
“We need to build solutions for specific problems, not general ones,” said Rao.
Nervana provides a simple, open-sourced platform, in which users can upload data.
One type of task the system can perform is speech-detect, or audio to text, which is sometimes better than humans. “It’s incredible to me,” said Rao, “that the technology has come so far in just a few years.”
Harris spoke about how a “digital twin” can reduce redundancies and increase efficiency when it comes to maintaining machinery. The model aims to “increase insights to deliver specific business outcomes, data from manufacturing, maintenance, and understand how it runs.”
One real-world problem Harris talked about is increasing efficiency in servicing parts. For example, in the past, a part would be serviced based on how long it had been used.
A plane, for instance, that flies in a hot, harsh environment will have contamination buildup on its blades, which can cause problems for flying. So, the company needs to inspect planes every 200 flights. But what if nothing was wrong with it? The time would be wasted in a maintenance shop, or wasted because it wasn’t in use.
But, if a “digital twin,” a digital version of the physical thing, is constructed to represent the original object, they can “use data from environment and produce a cumulative damage model,” said Harris. That way, they’d only bring in the part when needed–which could mean saving tens of millions of dollars.
The Climate Corporation
Andrejko showed how AI can bring “actionable insights to farmers.” Because of the ability to collect data from different sources in an environment, like machines, sensors, public data sets, there is now the chance to “integrate data sets together through models that draw inferences and deliver insights back to farming equipment,” he said.
“We can deliver data directly to the edge, to the fields and paddies where farmers are making decisions,” said Andrejko.
This is not just about the tech. It’s about solving a fundamental problem: “Staving off a pending food crisis.”
SEE: Facebook: Our AI will give everyone ‘superpowers’ (TechRepublic)
By 2050, Andrejko said, we will have a 60% shortage of food production, if we are going at our current rate.
While we have had several revolutions in agriculture–the green revolution, the biotech revolution–Andrejko sees us at the edge of a third revolution: The digital ag revolution.
Right now, AI can be used to optimize yield production, said Andrejko, to increase yield and decrease use of natural resources.
“This is not just hypothetical,” he said. In 2014, a Georgia farmer yielded 502 bushels of corn per acre–nearly three times as much as the average in the US, which was 172 bushels per acre.
“If we have all this data, why haven’t we done it yet?” asked Andrejko. Because “it’s not just about the technology,” he said–a critical piece is partnerships with government.
Tye Brady, head of Amazon Robotics, talked about how robots are transforming the warehouse and shipping industry. The Amazon fulfillment centers, he noted, are massive–over a million square feet. So, the time people used to spend getting to products is drastically reduced by having the robot bring items to the human worker.
“Bots have changed the game from hours to minutes to fulfill orders,” he said. Here are five observations Brady offered that will help us “fully realize the robotics potential”:
- Robots should be closer to humans, not further away. The fact that they’re accessible, a personal computer on you, has changed the game, said Brady. There’s a critical need for AI systems that allow robots to be in a natural human environment.
- Robots should be easy to use and should work for everyone. We need to develop natural and convenient interfaces.
- Robots should create a learning environment for humans. Brady believes we must “allow humans to be curious, to understand the capability of the robot.”
- Robots should perform tasks in a predictable and obvious manner. Robots must be, like roboticist Manuela Veloso said, predictable and transparent.
- Robots should have pointed purpose to reduce complexity and extend human capability. They should “do one thing, and do it extremely well,” he said.
Ultimately, AI is poised to make drastic changes across every industry. The final concern? “Will machine learning be deployed so quickly we can’t retrain our workforce?” Anthony Goldbloom asked.
“I’m not sure.”