There’s a philosophical theory from ancient Greece called eudaimonia, which supposes that, if we all live according to our true selves, we will reach our highest personal potential and we will also realize the highest good for society.
Optimists in the field of artificial intelligence think that we’re in a stage of technological development with artificial intelligence that could help us achieve something close to eudaimonia.
OK, here’s another philosophical word for you: qualia, or all the things that add up to your personal, individualized experience of your life.
Why are we throwing all this philosophy at you in an article about AI? Because technologists are currently trying to take advantage of recent developments in AI to create a collective human-meets-machine qualia on the path toward a form of eudaimonism — and you can already feel these effects at work today.
Look, for example, at how the technology company Intuit — which currently has 431 pending and issued patents related to artificial intelligence — uses advancements in knowledge graphs, sentiment analysis and deep learning as it builds capabilities that aid with the more complicated tasks of managing personal finance.
Developing AI solutions in a compliance-driven environment
Personal finance is an ideal space to apply AI, according to Kevin McCluskey, director of innovation, Tech Futures at Intuit. In accounting, taxes and personal finance, “there are explicit rules,” he says. “So the grand challenge is, how do we combine these rules with what we already know about you to make it so that it is very seamless and easy for anyone to comply and run a business or be able to do their taxes?”
McCluskey, who led teams behind Intuit’s TurboTax for more than a decade, says compliance laws give AI a point of control: “There is a fixed set of rules, because these rules are created by humans. There might be a lot of them and they might be complicated, but they’ve been written down and they’re very precise. You know whether you’ve got it right or wrong because it’s very explicit. And we have to encode those rules; we have no choice.”His answer is to capture the relationships among the data. However, the process of capturing these relationships and updating them can be daunting. To overcome this, they also apply AI to read documents published by agencies such as the IRS and automate converting them into knowledge graphs. The knowledge graphs, known as Intuit’s Knowledge Engine (KE), are used to capture the 80,000-plus pages of tax law published each year. Having an ability to help automate this is especially handy when, say, you might have to digest more than 600 rule changes in a matter of months (as was the case this past year when congress passed an act that brought the most changes in tax laws in the past three decades).
And because it uses a combination of AI algorithms, KE can minimize and personalize the user’s experience and explain back how it reached its findings.
“A lot of the breakthroughs we’re working on are really encoding human knowledge in a way that can take advantage of all this data that we have in the world and only surface to a customer what’s needed to get them done with their task,” McCluskey says.
Working towards a ‘frictionless future’
While surfacing all this information is a matter of technological heft, communicating it to a real-life human customer (with sentient empathy included), is a task for Wolf Paulus, a principal software engineer for Intuit’s Tech Futures division.
Paulus’s work focuses on humanizing tech, with a specific interest in the design of conversational user interfaces (CUIs) as they fit into the company’s vision of a “frictionless future.” As far as user interfaces go, “in the last few decades of development, we as an industry have really moved in one direction,” he says. “We’ve moved from [desktop-based OS systems] to the web, then mobile, and now, suddenly we’re going into two directions.”On one hand, he says, are immersive AR and VR experiences — which once you have outfitted yourself with a headset and controllers — are graphical and audible experiences that can put you in a totally different reality. On the other hand, however, is an interface with almost no friction, driven more by your voice than by what you click on or touch.
In a frictionless future, “you just walk into a room and you have this smart speaker, and you just start to speak. We’re not bound to the desktop. But you have nothing that can guide you, you have to either have experienced it before and you remember, or you have to be kind of adventurous and try things out. In this world, it’s what I think is the likability of the experience that will make you choose,” Paulus says.
Paulus often cites an example to illustrate why he thinks a user’s emotional relationship to an app is important: Say you’re driving and your GPS system reroutes you, due to a traffic accident. The voice in the app might cheerfully alert you that you’re now on the fastest route, while also cheerfully mentioning the terrible accident.
That emotional disconnect, while subtle, might be jarring or even offensive to you, and it impacts your overall experience using the program. Likewise, if the GPS voice is stoic or pushy, Paulus believes you will likely switch to a program with a more amenable voice, regardless of how much more capable the pushy AI might be.
To account for this, Paulus calculates the emotional dynamic of the correspondence that is sent to their customers through chatbots. With this tool, he compiles data from various online dictionaries and sentiment analyzers to assign an emotional score to the words, sentences and paragraphs of a written statement.
Let’s say you ask a chatbot, “Can I afford to eat at a restaurant tonight?” And the bot responds: “There are still $70 left in your restaurant budget, but you also significantly overspent in all other categories.”
Through the tool’s analysis, that statement ranks as highly negative. Why? Words such as “still,” “left,” “significantly,” and “overspent” all read as negative, according to data in the dictionaries consulted by the app, and so the chatbot can use different versions of the response until it reaches a more positive score.
“Our products are meant to inform you, to guide you, not to be judgmental. So this tool helps you to verify that text or speech content contains only the intended attitude or sentiment,” Paulus says. Currently the chat responds with, “You still have $70 in your restaurants budget, but please recognize, overall you are in the red.”
Helping humans through machines
Another area where communication becomes a key interaction point between human and machine is understanding when in the user experience a customer needs help — either by another human (as in, customer service), or an assist from an AI program.
“Deep learning is especially helpful here,” says Nhung Ho, the director of data science for Intuit QuickBooks. In one example, she says, “We’re able to apply deep learning when a customer is moving through our product but struggling because they don’t really know what to do. Based on their patterns with our product, we’ll dynamically generate a list of help articles to help them resolve their issue, and that just pops up automatically.”
Ho says that there are many latent signals hidden within a user’s product usage patterns that are difficult to encode, especially those with time and sequence dependence. By using the latest research in recurrent neural nets, Ho says her team of engineers at Intuit were able to create a model that serves up article recommendations in real-time, for each click made by the user.Providing help article recommendations enhances the user experience, however in other cases it can feel cold and disconnected. For an accounting product such as QuickBooks, often users are not well versed in accounting terms and may not be able to articulate the right question, rendering help articles less effective.
“One solution is to marry natural language processing and machine learning,” Ho says. “This allows us to better connect with our customers by allowing them to speak to us in their natural language. Then we translate that to accounting terms to deliver a better, more personalized help experience.”
While applying AI to make the user experience more humanized isn’t breaking news, it certainly is a newer trend in the field of AI applications, and they’re examples of how technologists work to meet these new needs as they appear to a growing field of consumers who rely on this technology in their daily lives.
At Intuit, McCluskey says, “We weren’t looking for AI and finding a spot for it, we were looking for customer problems that were becoming increasingly sophisticated in terms of both the customers’ expectation and the kinds of problems we have to solve for them. And the AI technology has progressed to the point where we can really and truly solve these problems.”
Prosperity means different things to different people. When it comes to designing tomorrow’s technology for the customer of today, Intuit engineers believe that they’re not just solving coding and data problems, they’re leveraging the most recent advancements in technology that change the lives of millions of people. To learn more about how Intuit works to solve the world’s hardest problems with technology like image recognition, CUI and assisted experiences to power prosperity throughout the world, visit intuitcareers.com.