Machine learning is upending everything
No longer confined to the bleeding edge, machine learning has leapfrogged past what we thought computers were capable of.
This article first appeared in Quartz in February 2019.
The era of cognitive technology is upon us. No longer confined to the bleeding edge, AI and machine learning advances have leapfrogged past what we thought computers were capable of, powering not just factory-floor robots and self-driving cars but settling comfortably into our everyday lives.
AI-based apps wake us up in the morning and remind us of the day’s agenda. Traffic apps use predictive analytics to estimate congestion and provide alternative routes. Intelligent assistants offer increasingly human-like responses to help make customers’ lives easier.
In the past decade alone, we’ve seen AI capabilities grow to touch nearly every aspect of a business, including marketing, remote working, data security, and customer service. Businesses are willing to invest in these tools, but are they ready to navigate their risks and roll them out widely? Maybe not.
This was one of the main findings from Capital One’s Are we there yet? The 2019 state of machine learning survey. Capital One surveyed 469 American executives with decision-making influence in IT, business, and software engineering and positioned within companies already leveraging machine learning solutions. As early adopters working in industries from agriculture to hospitality, their answers reveal plenty of eagerness and incentive to adopt ML, but also uncertainty. This report digs into those concerns and surfaces suggestions to address them.
The impact is visible
The machine learning market is projected to be worth $8.81 billion by 2022—up from $1.41 billion in 2017—and survey respondents agree that these innovations deliver strong, even decisive, benefits. Most are already witnessing an impact on IT, operations, and customer service, and the possibilities here are exciting. Take customer service: Instant is the new expectation for support, and AI allows companies to answer questions in a couple of clicks.
For example, Capital One applies ML to enhance fraud-monitoring and prevention tools. With its intelligent assistant, Eno, Capital One can automatically alert customers to potential fraud in real-time, help them report the fraud and lock their card, and then set them up with a temporary solution so they can keep using their account until a new card comes. With vast amounts of historical data, they are able to quickly adapt to new criminal trends while still working to reduce false flags for the customer.
“Machine learning allows us to create a personalized experience like traditional brick-and-mortar retailers, but with all the intelligence of a modern global banking system,” says David Castillo, managing vice president of machine learning and head of the Center for Machine Learning. “So much of that is finding better ways to serve our customers. For us, machine learning not only creates enterprise-wide efficiency and scale in ways not possible before, it also helps our customers have greater protection, security, confidence, and control of their finances.”
The risks are real
While new technologies like ML can help businesses with their customer journey, they aren’t without their own set of unique risks. Two, in particular, are top-of-mind for our surveyed execs: job impact and algorithmic bias.
How machine learning will shape headcount and hiring depends on who you ask. Unsurprisingly, software development execs are the most optimistic. ML is a software-based venture, after all. Nearly 64% said their staffs would either remain the same or grow due to ML adoption. Business and IT decision-makers, however, are more trepidatious. Nearly 30% of business execs said they believe ML will bring unpredictable changes to staffing, while 20% of IT leaders noted they will likely shed employees as a result.
The second fear surrounding this emerging technology is bias. Bias is one of AI’s biggest flaws because it’s one of humans’ biggest flaws. This may explain why there’s so much disagreement on the best way to tackle it. Of the survey options listed, from developing tools that detect partiality to using outside vendors for auditing, no single answer was chosen by more than a third of respondents.
Regardless of the specific approach, transparency should be a guiding principle when developing ML applications. It’s hard to trust what you don’t understand, which is why many AI and ML initiatives have had a “black box” problem: Stakeholders, from employees to customers, cannot fully explain how an AI model comes up with its decisions. That’s why “explainable AI”—the ability to articulate how models reach any given output—is so important. If you’re seeking to build fair and equitable results, you have to understand how those systems are constructed.
The plans aren’t in place—yet
Execs understand the risks and rewards of ML, but how do they integrate it? To successfully adopt any new process, you need two things: the right plan and the right people.
But a no-hitch ML strategy is not widely in place, according to the survey respondents. Less than half said they had a roadmap to address the impact of machine learning on their company. In fact, just 36% of the business execs said they had a change management plan to address the impact of machine learning on their company. And 19% of IT execs were unsure if they had a plan or not. This is a concerning data point considering these businesses have already integrated ML tactics.
While the majority of execs said that they are prepared to train their staff, which could mean anything from learning how to build models to creating ethical guardrails and new organizational protocols, the endorsement of preparedness is underwhelming. Respondents prefer to bring on in-house talent to meet new ML needs, but they’re open to outside help. Most interestingly, the software engineering execs noted that they were likely to leverage academic talent to help supplement their team’s training (58%). Working with academia to support team development is not a common trend but may become one soon.
All of this serves as a reminder that training should not be a reactive measure. From the onset, prepare and support your teams so that they are equipped to navigate shifts in their roles and responsibilities. The more proactive you are communicating with and empowering your team, the more successfully you can implement ML.
A way forward does exist
Whether internal or external, respondents agree that senior leadership is critical to the success of ML in their enterprise. Getting widespread buy-in for any new software can be a challenge, but, when evangelized consistently and clearly from the top down, innovations like ML can unite an organization around solving a whole new set of challenges. Company leaders should detail how ML will impact current processes, structure, and goals, pinpoint the stakeholders, and create ownership.
It’s still early days for this technology. Our strategies around it will continue to evolve. Those serious about harnessing its powers will have to establish new priorities, rethink systems, and invest in a workforce willing—and excited to—adapt to change.
Learn how to handle a new machine learning project, or whether to approach it at all.