What is artificial intelligence? Ask multiple people and you’ll likely get multiple answers. Is it chatbots? Is it really good analytics on an unimaginably large data set? Is it a computer beating a human at chess?
Everyone has a different perspective, but one thing it’s certainly not is under-hyped.
Panelists at this year’s AI media panel at PARTNERS 2017 debated the ins and outs of AI, reframing the conversation to get past how trendy the topic has become and focusing on the promise it holds today for enterprises.
Overcoming the hype
The panelists agreed that a business must have a strong use case defined before it throws itself into AI adoption.
“Don’t just get shiny new toy syndrome,” warned Andrew Stephen, associate dean of research at the Said Business School at University of Oxford. There is a substantial contrast between buying an Amazon Echo for $100 and placing an eight-figure bet on an AI system that requires a board of directors write-off, he said. The difference is that analytics for enterprise must scale — and executives must learn and adapt quickly to create validity around their AI investments.
Yasmeen Ahmad, director at Think Big Analytics, a Teradata company, agreed that business problems create essential focus around AI for enterprise.
“You need to do some experimentation and pilots, but focus them on a business problem or use case you have in mind,” she said. That gives companies the drive to apply new techniques and do something revolutionary. Then they can build out a sustainable AI roadmap, she said. Key to applying AI at scale, though, is less hands-on interaction from data scientists.
“It’s not just about deploying one model,” Ahmad said. Businesses must deploy “many [models] that can self-learn and adapt.”
Stephen Brobst, chief technology officer at Teradata said automation is the key differentiator between what people are calling AI and machine learning versus its much older cousin, linear mathematics.
“Automation discovers new features. That ‘learning’ [in machine learning] is important, because before data mining was manual.” And in a post-big data world, that model is unsustainable.
Human skillsets in an AI world
So once businesses successfully build an AI use case and invest in machine learning and deep learning applications, what skills will humans need to complement it?
Surprisingly, the panelists didn’t advocate for a future where we all need to learn how to code. Instead, they said to focus on developing critical thinking skills. Then, when AI requires less feature engineering from humans, businesses can rest assured that their models aren’t degrading without anyone noticing.
“For the marketer of the future, that means you have to know how to interact with those machines or model outputs,” says Ahmed. “You will have to understand what machines and algorithms are doing. You have to add a human element to that.”
Ensuring machine learning models aren’t operating in a black box is critically important to Danske Bank right now. The company uses a deep learning model implemented by Teradata to monitor fraud, and if a customer calls wondering why their transaction was turned down, Nadeem Gulzar, head of global analytics for Danske Bank, says he needs to have an answer.
“We have to trust that the AI model will work, but you also have to be critical,” he said. “[If the] model is saying you should go left, but you have the strongest feeling you should go right, question it. Is it really working or is something wrong?”
AI’s top applications right now
Moderator Martin Willcox, Teradata’s director of its Big Data Centre of Excellence, asked each of the four panelists what their favorite application of AI is right now, and there was across-the-board enthusiasm for health care applications.
Brobst described a future where predictive analytics could determine when someone is going to need emergency care, or even intercede to deter future health problems.
Stephen, who has focused some of his research on how social media affects mood, was interested in day-to-day health, helping people spend their time more wisely to maximize their psychological well-being.
Ahmad, who worked in life sciences before becoming a data scientist, anticipates a day where automated experiments could prove as valid as physical experiments, so researchers could virtually simulate the validity of drug trials or gene therapy. Then, health care could pivot from its one-drug-fits-all approach. “To get down to personalized medicine, you need that compute power to analyze at scale,” she said.
Gulzar emphasized the efficacy of health care AI applications versus other industries.
“The amount of data we are able to produce and monitor, you end up saving lives. That’s a big thing,” he said. “Yes you can upsell, but saving lives is actually out of this world.”
For resources on how to implement AI in your enterprise, go to the Teradata AI Insights page.