The field of artificial intelligence got its start at a conference at Dartmouth in 1956. Optimism ran high and it was believed that machines would be able to do the work of humans within 20 years. Alas, it was not to be. By the 1970’s, funding dried up and the technology entered the period known as the AI winter.
Clearly, much has changed since then. In 2011, IBM’s Watson system beat the best human players in the game show Jeopardy! More recently, Google’s DeepMind had similar success against Go Champion Lee Sedol. Microsoft has also opened a cognitive services division and others are sure to follow.
Yet for all the dazzling technological wizardry, we’ve seen little effect on most businesses. Artificial intelligence, much like PC’s in the early 80’s or the Internet in the early 90’s, remains little more than a curiosity for most managers. So I talked to Josh Sutton, who heads up the AI practice at Publicis.Sapient, to learn about what we can expect in the years to come.
1. Virtual Assistance
The first and most obvious way to use artificial intelligence is for virtual assistance. Many have already had the experience of speaking into a phone to search for answers to basic questions, like who invented the paperclip or directions to the nearest gas station. It’s become so commonplace that we rarely stop to think that it would have seemed like science fiction a decade ago.
Now we’re also seeing companies deploy chatbots for a variety of uses. The most obvious is customer service. Although some believe that this causes more problems than it solves, because the last thing a customer with a problem wants is to get stuck talking to a machine, there is potential in blending machine driven assistance with human customer service.
For example, when there is a spike in demand for customer service, such as when a plane is delayed or when cable service unexpectedly goes out, an artificial intelligence service can be used to answer simple questions, like what other flights are available or when service is expected to be restored. Deployed in this way, human customer service can actually be enhanced, by allowing human agents to focus on more thorny problems.
Marketers are also seeing the value of chatbots. For example the Hello Barbie chatbot allows little girls to have conversations with their favorite toy, while Lionsgate films created a chatbot app to help market its film, Now You See Me 2.
Sutton told me that, “Chatbots and virtual assistants are enabling a fundamental shift in how people interact with technology. I believe that over the next decade we will see virtual assistants become a core part of our normal experience across almost all of the activities that we engage in during a normal day."
2. Generating Insights
Data has been called the new oil because it is essentially the raw material of the digital economy. What used to be trapped in a vast maze of incompatible database and protocols — or worse, filing cabinets — is now quickly becoming a fungible commodity that can be accessed by new technology platforms like Hadoop and Spark.
Yet much like oil, data is largely useless without effective machinery to transform it into something of value. That’s the role that machine learning is starting to play. Today’s systems can not only take in billions billions of data points and analyze them in minutes, they can also learn from the data and get better over time.
For example, Sutton and his team at Publicis.Sapient have used machine learning technology to build a system that can predict a consumer’s demand for a mortgage based on specific life events as well as market-based factors. This allows loan originators to identify high value prospects earlier and increase deal flow.
What makes artificial intelligence systems so powerful is that, unlike purely statistical approaches, they can learn. That allows them to adapt when market behavior changes as well as continually improve performance as more data comes in.
3. Automation of Manual Processes
One of the constants throughout the rise of technology is the automation of work. First, came labor saving machines such as tractors and home appliances and later industrial robots and basic systems to automate things like travel reservations arrived. Today, more advanced robots are even able to work alongside humans in factories.
Yet David Autor, an economist at MIT sees a new era of automation emerging, where the primary division of labor is not blue collar vs. white collar, but routine vs. nonroutine work. In other words, artificial intelligence is quickly automating routine cognitive processes much as industrial era machines automated physical work.
This is a trend that is already very much underway. Smart algorithms are already replacing junior lawyers for legal discovery and companies like Narrative Science can do routine journalistic work, such as summarizing box scores and financial reports. These technologies not only do a very capable job, they can work 24 hours a day without coffee breaks.
Publicis.Sapient helps its clients automate routine processes in a very similar way. For example, an investment bank can use artificial intelligence to automate the very labor intensive process of compiling data from various reports, including industry reports not found in standard databases, and perform an analysis to determine a particular firm’s profitability.
4. Unlocking Unstructured Data
Traditionally, almost all of the data we analyzed was structured data, the kind that gets captured and stored in a database. So we were reasonably good at deriving insights from data generated for that purpose, like sales from a cash register or answers to a consumer survey, but most everything else got lost.
That’s a bigger problem than most realize, because structured data represents only a small part of the information available to us. In fact, it has been estimated that 80% of digital data is unstructured. “Analyzing unstructured data is a small area today, but over the next ten years I think it will be the most impactful,” Sutton says.
To understand the how important unstructured data can be, he offered the example of an energy company that needs to lay a gas pipeline. What was built there before? How has the area been used? What problems had arisen in the past? Much of this information resides in old inspection reports and training manuals that have been sitting on a shelf somewhere for years.
Another area of potential is the ability to understand consumer conversations. One company, Mattersight, uses call center conversations to identify customers’ personality types, so that they can be served by someone with a compatible service style. We are really just beginning to unlock the potential of unstructured data.
An Operating System For Data
In 1936, Alan Turing invented the idea of a universal computer, but for a long time it was just that, an idea. Computers were largely specialized machines that were built to perform a very narrow range of tasks, like scientific computations or back office functions such as handling payrolls. In effect, computers were little more than calculating machines.
That began to change in the 1970’s when operating systems came into wide usage, allowed computers to easily switch from running one application to another. Soon after, the rise of personal computing made it possible for someone to easily switch from writing a document, to balancing a budget to shooting aliens in a video game.
We’re at a similar point now. Theoretically, we have access to mountains of data, but practically we have little ability to derive insights from most of it. We are drowning in data more than anything else. That’s the true potential for artificial intelligence, to be an operating system for data so that we can derive its full benefits.
“If the Internet gave us access to on-demand data, an AI-first world will give the world on-demand insight,” Sutton says. We are entering a new era of cognitive collaboration in which machines become far more than just agents to perform tasks, but help us to understand the world and make better decisions.
