A.I. May Not Be That Smart, But Its Business Use Keeps Growing

Few of them care to say it with much force, but a number of specialists in artificial intelligence (A.I.) and machine learning (M.L.) admit that A.I.’s true potential may be based on something besides human-like displays of intelligence. To them, the technology is a powerful (though still-evolving) toolset that allows for faster and more sophisticated data analysis and automation.

And while it may be mislabeled in this context—it’s more about speed and automation than “intelligence”—businesses have found a real use for A.I., even in these early days.

“A.I. really is the science of training systems to automate tasks to a certain, higher level,” said Wayne Thompson, chief data scientist of the SAS Institute in Cary, N.C. “It could be as simple as recognizing a $1 bank withdrawal as a fraud, but we think of it as being more analytical.” For example, a chatbot has to understand what a user is saying in order to find the true answer. “The key word here,” he noted, “is ‘true.’”

John Harney, CTO of New York-based DataScava, a developer of an unstructured data miner, sees A.I. as being “a combination of interpreting natural language and using truth and logic trees to find output that hasn’t been declared a mistake yet.” He doesn’t see any components that are “anything akin to [human] thinking,” he added, and notes that capabilities such as remembering mistakes, purpose-specific apps and vision “are really support functions.”

In fact, none of the experts we interviewed are thinking about building self-aware machines. “At IBM, we’re focused on building A.I. technology that enhances and scales human expertise for the enterprise, as opposed to attempting to replicate human intelligence,” said Rama Akkiraju, distinguished engineer at IBM Watson in San Jose, Calif. The purpose of A.I., she continued, is to train systems to process large amounts of structured and unstructured data and “turn it into actionable insights which can be used to enhance human decision-making.”

Harney strips A.I. to its most basic level when he suggests it “creates output by trial and error.” And humans must be involved to extract real value from the systems. “A human is required to indicate which outcomes are desired, which are mistakes and which are neutral,” he said. “This testing of output is an ongoing process. Each correction is another logic route.”

How Much is Hype?

If some people say A.I. is in the “buzzword stage,” it’s because the technology is “already helping us to solve big, complex problems, and it’s just getting started,” Akkiraju said. “Studies have shown that 80 percent of the world’s data is ‘dark,’ meaning it’s currently unsearchable. A.I. holds the key to revealing that data, and delivering real, actionable insights to businesses.” That’s a real value, he said, which means “the technology is not just hype.”

“I’m very bullish about this,” Thompson said. “Look at autonomous cars. Machines can see better than you can.” As systems advance and context becomes a factor in natural language processing, software will be able to interpret phrasing so it can answer questions properly, and even anticipate issues. To use a simple example, a question about “weight” is likely different than a question about “wait,” and machines will recognize the difference.  

To Harney, that means something like an automated receptionist is the ideal application for A.I. “Using natural language processing and logic trees, the system can use a defined set of inputs to calculate which of the defined set of outputs are most likely to be inferred,” he said.

At the same time, Harney isn’t as enamored with A.I. as many of his colleagues. “It’s hard to think of a use case that really benefits from using A.I. but can’t effectively be addressed in other ways,” he said. “Automated homes, vehicles, communications can all operate without A.I.” In fairness, he added, it’s not that A.I is bad. It’s that its reputation has been over-hyped and “it’s being offered as a solution for many applications it simply doesn’t have the components to address.”

Bullish as he is, Thompson seems to think along similar lines. “The problem is machines can’t reason,” he said. “They’re not conscious and I’m not sure we’ll ever get there. A kid can see a desk as something to stand on or hide under,” depending on what’s going around her, he explained. A.I. isn’t near that level of reasoning.

Tech’s Misperception

That example points to the biggest misperception about A.I.: that it can think. “A human can take a set of facts, tools and a desired objective, then use seemingly unrelated references to derive an approach,” Harney said. “A program moves in logical steps and so lacks the ‘artistic’ ability to imagine entirely new approaches.”

“A.I. is more a way to automate fundamental tasks like driving a truck,” Thompson said. “It’s very focused.” That’s a lot different from the idea of “hard A.I.,” where machines align with human capabilities.  

“Today there’s a notion that you feed information into an A.I. engine, which then ‘analyzes’ the information to come up with an innovative solution,” Harney said. “In fact, there’s no component that performs any form of analysis.”

But such limitations have to be kept in perspective. Regardless of whether “artificial intelligence” is an accurate term for the technology, the value of its capabilities is apparent. According to IDC, spending on A.I. systems will increase from $12 billion in 2017 to $57.6 billion by 2021. “A.I. has the power drive real economic growth and change the way we work,” Akkiraju said. “As the technology develops and businesses continue to adopt A.I., we’ll see even more benefits.”  

In addition, Akkiraju argues, implementing A.I. solutions is becoming more straightforward. “While the technology does require new tools and ways of working, many A.I. services are readily available to developers, and it’s simple to get started quickly,” she said. “In many cases, those looking to get started with A.I. can use out-of-the-box services to easily build powerful, A.I.-enabled solutions, such as conversational agents that can assist in customer service.”  

As has happened with many technologies, as A.I. becomes more complex under the hood, it may very well become simpler to implement and use. Data scientists and software engineers aren’t the only ones who should be paying attention: User interface and UX designers, hardware engineers and the like are sure to see growing opportunities coming their way, as well.