SAS provided their business audience with industry and academic perspectives on deriving value from AI, and then related it specifically to best practices that organizations can use to make it part of their business strategy.
TORONTO – On Wednesday, SAS Canada hosted an all-day Artificial Intelligence [AI] executive event here, focusing on how companies can best combine AI and human intelligence to derive actionable intelligence that impacts business outcomes. While detailed breakout sessions continued the whole day, the themes were laid out in the event’s opening keynote.
The event had four objectives, said Steve Holder, National Strategy Executive, Analytics and AI, SAS Canada.
“We want to look at how AI is playing out in the Canadian market,” he told his audience. “We want to speak to the potential of AI, and what it will bring to your business. We want to give industry insights in use cases where it can be applied. And we want to look at the best practices to take AI from being a buzzword to being a deployable, scalable part of your business strategy.”
Mike Gualtieri. VP, Principal Analyst Serving Application Development & Delivery Professionals at Forrester Research, provided an overview of where AI is at today overall. He said that AI consists of two types – pure AI and pragmatic AI – and that the first is irrelevant to a business conversation.
“Pure AI mimics human intelligence, and that is not what we are doing here,” he said. “A survey of AI researchers produced the view that it has a 50 per cent chance of being achieved in 125 years.”
Pragmatic AI, on the other hand, is not one thing, but several, and includes human knowledge, machine learning, deep learning, robotics and sensory perception, although at this point, machine learning is the most common.
“Nearly 80 per cent of companies surveyed see increasing demand for machine learning models,” Gualtieri said. He noted that there are several inherent limitations to their use, however.
“Models are about probabilities, not absolute, and accurate models may not exist for every question,” he said. “The models are also based on correlation – not causation – even though it’s called data science.”
Good data is also a fundamental prerequisite.
“Algorithms get all the press but it’s the data that matters,” he said. “It’s Garbage In, Garbage Out, the original computer problem. It all requires data.”
Automated interpretation of the data is also sometimes less than fallible. Gualtieri said. He presented a photograph of an airplane on the ground with people milling about, and a smiling man forefront, and showed that Google’s algorithms had categorized the scene as recreational. However, a closer look that a human can do indicated that one of the plane’s wings was touching the ground, and that the scene was, in fact, the aftermath of a minor air crash.
“Google analytics dropped the ball on this one,” he said “Human expertise matters in assessing images. AI is smartest when driven both by humans and by machines.”
Gualtieri indicated Forrester’s advice to companies who want to use AI to examine business processes.
“Stop wasting time and money on unactionable analytics,” he said. “Lots of effort is made on stuff that doesn’t matter, that has no business outcome. Identity high-value business process that are also data-rich and prioritize those. Infuse those applications with those machine learning models – operationalize them. And operationalize AI at scale with ModelOps.”
Ajay Agrawal, an economics professor at the University of Toronto’s Rotman School of Management, indicated that much of the strength of the new economy comes from the fall in some costs, particularly the costs of search and digital distribution. A co-founder of the Rotman’s Creative Destruction Lab, a pre-seed stage start-up program with a heavy emphasis on machine learning, Agrawal said that AI does one thing – predict – and that its lowering of the cost of predictions impacts the costs of complementary and substitute commodities, raising the former, and lowering the latter.
“Data is now newly valuable because the cost of prediction is newly cheap,” he stressed.
Drawing on the new book Prediction Machines: The Simple Economics of Artificial Intelligence, written by himself and two other Rotman professors, Agrawal said that this concept should allow business leaders to understand how their businesses can effectively employ AI.
Prediction problems are ubiquitous, he said, and the key is understanding where AI can be added to the process to solve a problem. For example, he said that some think of HR as a soft science which is instinctual, and a poor place to use AI. But he emphasized that the HR process is exactly about these situations. It involves predicting which of 200 applicants for a job would be the best five to interview. It involves much the same ratios in predicting which of a field of existing employees would be the best candidates for promotion to the next level. It involves predicting which of your superstars would be most likely to leave for a better opportunity elsewhere. The key to leveraging AI for these and many other problems is understanding where and how AI can be best employed.
So how do business leaders do this?
“One way is to focus on tools,” Agrawal said. “Take your existing workflows and slice them up into tasks. Then take a task where there is a core prediction capability, and where there is value in knowing the answer, and drop in an AI. A bank would use this for example to estimate the odds of someone paying back a loan. The key is dropping in the AI where you get the greatest return on investment. You can then determine the cost of building an AI application for each task, and start with those that offer the greatest ROI.”
Understanding your workflows and where you can generate that value is critical, Agrawal emphasized.
“Focus on predictions you want to get rather than how to get gold from the straw in your data lake,” he told the audience. He noted that Amazon had filed patents around anticipatory shipping – shipping things to customers based on predictions of what they will want without them ordering it, on the calculation that the profit from selling two extra things to a customer outweighs the cost of picking up and restocking the four things they didn’t want. That’s an example of this kind of application.
Holder then moved specifically from the industry and economist perspective to the SAS view of how best to unlock the value of AI for business. He emphasized themes of the earlier speakers that AI isn’t a failsafe crystal ball, and that what it is is a science of training systems to emulate human tasks through learning and automation. The goal of SAS in this process is effectively to be an open API, not a programming language.
“Data drives AI and the use cases set the value,” he said.
“The process has three buckets,” Holder explained. “Understand the context, understand patterns, and recognize objects.” He emphasized that AI is a spectrum of functions, from commonly used ones today like machine learning, predictive analytics and rules based systems, through to others like deep learning, as well as those like robotics and computer vision which are harder to attain and which he said is more on the fringe right now.
“The important part is the process for embedding and automating these processes,” he said.
Holder told customers that there are four things they need to consider in their AI strategy.
“How do I explain my AI,” he said. “For that you need transparency in AI systems.
“How do I get control,” he indicated, emphasizing that this is an issue of governance and oversight rather than access. “In addition, AI models decay, and they need to be retrained.
“Obtaining scale is extremely important,” Holder noted. Finally, he emphasized that you have to have choices.
“Not everyone is a coder,” he indicated. “People need to be a huge part of the process and training matters.”