How Can Artificial Intelligence Deliver On Its Promise?

Increasingly, the industry captains and senior executives are realizing that the answer could lie in the use of advanced technology — in particular, machine learning.Business leaders today are faced with an existential crisis: how to increase the quality of their products and services while driving down the costs. Increasingly, industry captains and senior executives are realizing that the answer could lie in the use of advanced technology — in particular, machine learning (ML). ML creates digital labor that can automate a business process, reduce the costs for the process and enhance the quality by the consistency of the outcomes.

I predict those businesses that don’t adopt digital labor will become uncompetitive.

It is for this reason that ML is being seriously considered across many industry segments. While specific small projects are benefitting from ML, there are instances where ML hasn’t delivered on its promise — especially for larger, more ambitious projects for business process transformation. Why is that? Because a couple of prominent issues are holding it back. I'll discuss the issues and how to overcome them to realize the full potential of intelligent machines.

Many experts, including myself, have experienced these issues firsthand in numerous enterprises. Customers have, for instance, complained that lack of transparency has severely hurt the adoption of ML in their enterprise. Analysts, too, estimate that more than half of new investments in AI will be for transparent machines.

Let’s discuss the issues. First and foremost, ML has to overcome a major challenge: How can we make an intelligent machine transparent? ML has gotten a bad rap — it’s incredibly complex and opaque. It does not explain its predictions.

Without transparency, the subject matter experts who own the business process might not trust the machine’s execution of the process. Lack of trust means limited automation. Lack of data means the experts don’t get to benefit from the insights the machine has mined from the data. Lack of data damages executives' confidence in the ability of ML projects to deliver on the ROI promised. Subsequent ML projects, unfortunately, likely won't get funded.

How do we overcome this challenge? I believe the answer is to make ML more transparent. Efforts have been made recently in this regard — explainable AI comes to mind immediately. The explanation of predictions is being implemented by vendors, large and small, with most making progress based on preliminary research in academia and R&D labs.

A few startups have shown promise, making their ML explain the predictions well, especially for classification problems. For example, some ML vendors — from the established Google Analytics to the newer entrants like H2O — base transparency on the research from academia called LIME.

Much of the information we consume is textual. It's no surprise, then, that another major challenge is the inability of ML to reason and derive intelligence from textual data. This handicaps the accuracy of predictions.

For example, suppose a bank is trying to stem attrition of its premium customers. It has a ton of data available about the customers — demographics, types of financial products, volume of transactions, amounts transacted and so on. A particular customer has an issue and calls customer care. The customer doesn’t get the issue resolved and gets frustrated, so they decide to conduct their banking at a competing bank (i.E., soft churn). The bank is stumped; it has no idea why the customer’s account activity all of sudden went down considerably.

The ML model wasn’t able to predict churn proactively; its alert came late. The model failed because it didn’t consider the customer care records for analyzing the customer. ML didn’t have the capability to analyze natural language exchanges between the customer and the contact center agent.

ML has to consider all information, whether numeric or textual, to predict correctly. ML practitioners dump all types of data on the ML platform and let the data objectively tell ML which matters and which doesn’t, and ML accordingly constructs the final model. Unfortunately, most ML platforms can’t process unstructured (textual) data. How can we overcome this limitation?

There’s hope as natural language processing technology matures; only a select group of vendors have an ML platform that can process numeric and textual data. Research — GloVe and LDA, in particular — has given us some facility with NLP semantics.

Last, but not least, is the challenge of the inadequacy of training data. Most of ML is supervisory, which means the intelligent machine uses data to learn about your problem and becomes an expert in your domain, ready to predict. The machine’s training will determine how well it will execute your business process. It depends on the quality of the training data. As ML has painfully discovered, training data is seldom adequate, suffering from a variety of issues, major among them being:

• Incompleteness: The machine is learning to predict an outcome in your process. Not all input variables that influence the outcome are compiled in the training data. The machine’s learning is incomplete, its predictions suspect. To add to the woes, the data is thin, with not nearly enough records required for the machine to learn well.

• Bias: The training data is skewed, capturing one set of events with many records at the expense of other events, making the machine compromise its well-rounded learning. Sometimes, there are unfortunate coincidences (i.E., correlations among independent variables) that don’t reflect reality but inadvertently creep in.

This continues to confound ML experts. Businesses looking to benefit from ML must build the infrastructure for storing tons of data about their operations and make it accessible with sound data governance. Discipline around ML’s best area practices is required. Research is hot in this area.

To conclude, ML’s digital labor is here. I predict that ML will be our colleague in the enterprise, working collaboratively with human counterparts. Kinks that prevent this vision will be ironed out as ML platforms become increasingly more sophisticated. Start to use ML in your enterprise, in small ways, to alleviate risk and see how effective is it for your environment.

The device resembles a stingray, except it doesn’t patrol the water in search of prey.

It’s perched at the edge of the water, and rather than threaten humans, it aims to save them.

It’s much louder, too — emitting a piercing, high-pitched alarm when it sees a motionless human head beneath the pool’s surface for more than 15 seconds.

From a corner of the pool at the Easton/Phillipsburg Branch of the Greater Valley YMCA, the Coral Manta 3000 knows a human head from any old beach ball. The machine, which the branch is testing on behalf of YMCAs across the country, uses artificial intelligence to recognize body parts and learn how humans act in the pool in an effort to prevent drownings.

It’s not that the branch has had any drownings in the last 25 years, branch Executive Director Lori Metz said. Nor will the robot replace lifeguards.

“It’s not reactive,” said David Fagerstrom, CEO of the Greater Valley YMCA. “It’s proactive.”

Sobering statistics

Despite the emergence of fancy pool alarms, camera monitoring systems and drowning prevention interest groups, death rates among children in swimming pools have barely budged over the last 20 years.

Drowning is the second-leading cause of unintentional deaths among children in the United States, second only to transportation-related accidents. The death rate looks small, especially when data from the Centers for Disease Control and Prevention is whittled down to drowning deaths in swimming pools, but it is consistent: about 1 child death per 200,000, or 342 children in 1999 and 385 in 2018, according to CDC data.

But when considering how many kids ride in cars versus swim in pools every day, it’s an “astronomical mortality rate,” points out Kevin Trapani, CEO of insurance company The Redwoods Group.

Drowning as a whole — oceans, pools and bathtubs — claimed the lives of about 3,500 people a year in the U.S. From 2005 to 2014, a 2015 World Health Organization study found.

Fagerstrom said he has been looking for affordable artificial intelligence systems off and on for the last 15 years. In January, he discovered the Coral system, developed by Israeli software developer and budding entrepreneur Eyal Golan, CEO of Coral Detection Systems.

Golan’s product came to market over the summer.

When Fagerstrom reached out, Golan hadn’t heard of the YMCA, but it didn’t matter much. In addition to the device’s $2,500 price tag, Fagerstrom liked Golan’s mission.

Six years ago, while thinking of his own swimming pool and his children, Golan thought, why not use artificial intelligence for pool safety?

“Like most parents, I am a concerned dad,” Golan said.

Two months after he and a partner started developing their idea for a system that could charge itself and require no human intervention, two 10-year-old girls drowned in a backyard pool in a Tel Aviv suburb. Their names were Coral and Or.

Golan decided to name his technology after them.

“Drowning happens so quickly and so quietly,” he said. “We cannot expect people to be robots and machines and never lose sight of children for three, four, five straight hours.”

For the next six months, the Easton branch will test the system and see how it adapts to the six-lane pool. Though Golan developed the technology with private pools in mind — it sees clearly about 10 yards by 10 yards — the YMCA will give Golan feedback on how to better adapt the system to larger facilities, where people of all ages swim from 6 a.M. To 8:30 p.M. The branch paid for four systems — one on every corner of the pool — out of its own operating budget.

After the pilot, the Greater Valley YMCA will consider expanding the technology to its other branches. Metz said YMCA is searching for grant opportunities, and will bring its results to the attention of the Pennsylvania State Alliance of YMCAs.

“There are a ton of YMCAs out there that would love to have something like this that’s affordable,” Fagerstrom said.

‘We want no one to drown’

Golan certainly isn’t alone in developing artificial intelligence systems for pool safety.

Adam Katchmarchi, executive director of the National Drowning Prevention Alliance, said any new technology that aims to make pools safer is a welcome development as statistics remain stagnant.

“Because something does have to change,” he said.

Katchmarchi has seen computer-assisted drowning detection systems dating back 10 or 15 years. In the commercial realm, a product called Poseidon is well known, and he said he’s seen a surge in products that have reached the design stage in the last three to five years. It’s been tough for any one of them to really take off, he said.

“Those systems can be cost prohibitive,” he said. “They’re not cheap.”

A Poseidon system, for example, costs between $150,000 and $300,000.

Trapani, whose Redwoods Group represents about half the YMCAs in the country, said Poseidon approached the group with its technology in the early 2000s. About 15 YMCAs in the country have Poseidon technology installed, he said, but the cost barrier creates an equity issue, where YMCAs in economically challenged neighborhoods are less likely to afford the technology.

“Our mission is not to just save wealthy kids’ lives,” Trapani said. “We want no one to drown.”

His YMCA clients also have beta-tested wearable bands that emit radio frequencies when submerged for too long or when swimmers show signs of distress, but the company that manufactured them closed in January.

The Coral system, on the other hand, is passive, meaning it does not need human intervention to activate or deactivate. It watches swimmers on its own, having filtered through millions of images to learn what human heads look like. The machine emits a chirp when a person enters the pool, then learns who that person is so it won’t chirp again if they hop out and hop back in. The system also connects to mobile devices, which will sound should the alarm get triggered. It powers itself on and off, charging primarily with solar power. In the indoor YMCA’s case, officials charge its back-up battery every few days.

There have been other AI systems Trapani has seen since Poseidon, but none at Coral’s price point and that distinguish between a person drowning and a person diving to the bottom of the pool on purpose.

It’s tough even for lifeguards, who often face high turnover and hot and humid conditions, Trapani said.

“What we ideally want is perfect lifeguards, but even perfect lifeguards have certain barriers,” he said.

Drowning detection technology can only add to layers of protection, which include lifeguards, fences, swim lessons and awareness campaigns, Katchmarchi said.

He can’t say any particular tool will decrease risk by any particular percentage, but he can point to one tried-and-true adage among safety advocates:

“You don’t know which layer of protection will save a life until it does,”