AI Orchestration Enables Decision Intelligence
According to Bar-Lev, a broad definition of AI orchestration would be “managing the set of tools, processes, data and talent related to the application of AI within an organization such that it becomes a part of the operational day-to-day as opposed to staying within research or an experimental process. This definition takes into consideration the larger set of decisions and resources required for this to happen — such as management, application of budgets and resources, etc.” And this is the same definition used in this report.
In short, AI orchestration provides the infrastructure to enable automated decision intelligence in the daily operations of the business.
AI greats such as Lorien Pratt and Google’s chief decision scientist Cassie Kozyrkov are quick to say that the ultimate business advantage in using AI is decision intelligence — the automation of the full action-to-outcome process (Figure 1–1). As Kozyrkov writes, “Decision intelligence is the discipline of turning information into better actions at any scale.” And therein lies a top reason to use an AI orchestration approach.
“If you’re a big traditional company with a stable business, you have to do AI differently than the glitzy tech companies like Google or Tesla that can turn on a dime,” says James Taylor, author of Digital Decisioning (Meghan-Kiffer Press) and CEO of Decision Management Solutions. “Other big companies can’t just throw out a large and stable part of their value proposition just because the data said so. At least not without serious consequences.”
That is why the failure rates of AI deployments are so high, says Taylor. AI failure rates have been estimated at 50% by IDC and higher than 85% by Gartner. Traditional companies are experiencing high failure rates, says Taylor, because they are trying to mimic tech companies instead of staying true to their own strengths.
“Instead, start with a business problem and decide what a ‘better’ outcome would be, then work backwards to determine what technologies and processes you need to get there,” Taylor says.
AI orchestration helps put the pieces together to accomplish the goal — the well-defined and predetermined “better” outcome you identified at the beginning.
And there are many pieces, not the least of which are the necessary changes in processes. The best resource for how to handle those is the open standard for analytics processes called CRISP-DM, which stands for “cross-industry process for data mining.” The CRISP-DM methodology was developed in the European Union and is free for anyone to use. Hundreds if not thousands of academic papers have also been written about it. And books, like Brown’s Data Mining for Dummies, summarize it in clear, easily understood language.
“It is the standard for management practices in integrating the technical aspects into the business, so there is no need for anyone to reinvent the wheel here,” says Brown. “It’s more than a diagram, it’s 75 pages of fine print on exactly what to do, and it was created by practitioners from around the world.”
CRISP-DM is a “structured technical playbook, but it doesn’t tell you how to extract your data or create models in a framework,” Brown says.
In this holistic approach to AI, success itself is redefined. In terms of processes, data from AI feedback further perfects them, ensuring a progression of successes. In terms of business success, the measure is in what this decision changed in the business.
“I challenge the notion of success in many modern AI projects,” says Taylor.
Consider AI finding more cancerous tumors, and sometimes earlier, in medical scans than doctors have. Show me one patient whose treatment was changed by that. That is a successful research project because we already suspected cancer in those patients and the AI indeed found it. It is not a successful business project because it is of only marginal use. We still don’t know what decision to take to improve patient survival rates or reduce treatment costs, for example.
Taylor’s meaning is further illustrated in the current public health crisis.
Take, for example, Dr. Anthony Fauci’s recent dilemma wherein he had the ability to tell the public what the data says about the spread of coronavirus but was unable to determine whether it’s safe for kids to go back to school yet. All eyes were on the director of the National Institute of Allergy and Infectious Diseases (NIAID), but he could not deliver the decision they most wanted to hear. It’s not a position any business leader would want to find themselves in.
“It’s a completely different discipline to go from discoveries and knowledge to decisions. And it is completely missing in the use of AI today,” explains Pratt.
The motivation behind the effort to mature and orchestrate AI is simple and easily understood: when all is said and done, most of us don’t really want to know all the details on how the novel coronavirus spreads; we really want to know whether to send the kids to school in the morning.
Business decisions flow in a similar vein: executives need to know a specific decision to make to effectively solve problems and bring about positive change.
Learn faster. Dig deeper. See farther.
Join the O’Reilly online learning platform. Get a free trial today and find answers on the fly, or master something new and useful.
Pam Baker is the author of eight books and hundreds of technology articles published in leading online and print publications. Her latest book, Data Divination: Big Data Strategies, was featured at the prestigious National Press Club book fair and listed as recommended reading for executives on the National Chamber of Commerce’s reading list. Onalytica ranked Baker as a Top 50 Big Data Influencer in 2015. She is a popular speaker and industry analyst as well. Baker is a member of the National Press Club, the Society of Professional Journalists, and the Internet Press Guild.
