The biggest problem with AI is also its biggest opportunity
Much of this still sounds like something out of a sci-fi movie, and that’s evidence of how far this technology has come in recent years. Businesses small and large are already contemplating how AI can support their goals, and some are already implementing it. There’s just one problem: Often, we have no idea how it works.
The intelligence is a so-called “black box”—meaning that we can’t track how it thinks or gets from point A to point B—and that has put a cap on many practical applications of AI. It’s great as a suggestion machine, but does little to actually work with humans or enhance communication.
But by solving this black box problem, we might be able to chart a new path for AI moving forward.
The bothersome black box
The AIs we all know—Alexa, Cortana, Siri—are great at sifting through information and performing specific commands such as playing music, executing Google searches, and relaying the weather forecast. With these lower-level AI tasks, it’s OK to not know how it arrived at the answer. For industries just looking for suggestions, comparisons, or analytics, this is fine.
But for something like Watson, which is coming up with personalized health care plans, helping target demographics for marketing, and finding errors in legal documents, that poses some problems. It’s excellent at giving suggestions based on personal or company information, and organizing huge amounts of data, such as helping with tax returns, but when it comes to high-risk, cognitive decision-making, it’s hard to feel comfortable relying on Watson completely.
Even if an AI is accurate 99 percent of the time, it’s difficult to accept that an answer, especially one to a complex problem, is correct when you can’t explain the reasoning behind it. Just like in math class, AI needs to show its work before we will truly be able to trust it can go beyond data crunching to actual decision-making.
By better understanding the associations AI is making, and tracking and editing its logic, we can open up new uses and applications, from diagnosing medical ailments to demoing a new product, offering certifications, and handling customer service. By baring the neural pathways of AI, we can help it learn and grow its role in the workplace.
The employee enigma
Automation always sparks some fear—if machines can perform a job as well as humans, then why spend money on employees? Headlines about potential job disappearances in the coming years and decades stoke those doubts.
In reality, the idea of such a drastic takeover isn’t such a threat. Automation has removed the need for hundreds of workers from predicting crop yields, for example, but that just means that the USDA can use those employees for higher-level tasks, while the machines focus on the time-intensive, repetitive labor.
AI could do the same thing, for example, for training and supporting employees at a hardware store, who all have different expertise: plumbing, electricity, power tools, and so forth. If a customer asks a plumbing expert about electrical equipment, they could ask a coworker—or, they could ask a store-specific AI. This would mean less time searching for answers they are unfamiliar with, and more time demonstrating tools or materials, or offering more helpful, trade-specific advice. The same goes for training—instead of having to dedicate time and resources to getting new members up to speed, the AI could cover most of the basic information.
The most successful AI wouldn’t remove the human element but enhance it, scaling employees’ knowledge, sharing it with the rest of the workplace, and removing the smaller, time-consuming tasks. But in order for it to do so successfully, we’ll need to know how it thinks so we’re sure the help it’s providing is sound.
In theory, AI could become a valuable coworker. If it has all the answers, we can spend less time figuring them out or searching for them, and more time doing practical tasks. However, AI isn’t ready for that kind of undertaking.
The way AI works right now is great for data crunching, pattern spotting, and generating suggestions. But the technology struggles with communications and actual knowledge sharing. Part of this comes from the way it learns. The other part comes from the fact that we don’t know how it learns, making it difficult to expand it to more sensitive or nuanced applications.
Automation doesn’t have to be scary, and it certainly makes for more efficient workplaces. The trouble is, until we can break open the black box—and make sure AI works with us, rather than for us—its functions will remain limited. The automated future will undoubtedly reshape your organization. But AI has some big steps to take before that happens.