Next Steps in Climbing the AI Ladder
Kelly Combs, Director of KPMG's Emerging Technology Risk practice, speaks at the Journey to AI session
The vast majority of organizations perceive AI as a strategic opportunity. While this is true, the reality is that most have only committed to random acts of AI as many find it difficult to scale the technology across their organization. To better understand some of the challenges business leaders must overcome, we went to “Accelerating the Journey to AI” at Think 2019. The session was rich in insights as Chief Data Officers and Chief Innovation Officers gave a behind-the-scenes look at their approach to deploying AI.
A $15.7 trillion opportunity
“Artificial Intelligence is a $15.7 trillion opportunity,” said Rob Thomas, General Manager of Data and AI at IBM. The problem? “We’re only at a 4 percent adoption rate right now.” Every company understands that AI can drive costs savings and revenue increases, but “for AI to truly scale, there are many challenges we need to overcome.”
Data is what fuels digital transformation. According to Thomas, there are four things companies need to do with their data to climb the AI ladder: Infuse it, analyze it, organize it, and collect it. And, one lesson that can be drawn from successful organizations is that you need a unified information architecture that delivers everything an organization needs for AI, on any cloud. Thomas hit the mark when he said, “There’s no AI without IA” (information architecture).”
But that’s not all. In order to deploy AI at scale, businesses “need the right tools and services, open source to combat vendor-lock in, and trust and transparency.” Addressing the widening AI skills shortage is another imperative. To address these challenges, IBM just announced that it has made Watson available anywhere, on any cloud. With “Watson Anywhere,” businesses can deploy Watson wherever their data resides—public or private clouds, from any cloud vendor, as well as on-premises IT.
Lessons from the field
There are a lot of myths around artificial intelligence. “AI is not magic; it’s computer science,” Thomas noted. One way of dispelling the myths is to look at real-world cases of companies applying AI—and that’s what the rest of the session was about.
Reena Ganga, a UX Designer at IBM, explained how AI models are being created to identify the patients that are at the most risk so that doctors can prioritize their time where they’re needed the most. This is no small feat. With 6,000 people for every cardiologist in the world, AI can help fill the gap.
Laurent Prudhon, Cognitive Factory Leader at Crédit Mutuel, shared stories of how his company is putting AI to work. Crédit Mutuel turned to IBM Watson to provide “faster and personalized services” to its customers across 5,000 bank franchises worldwide, he said. According to Prudhon, the biggest success factor was “working with IBM to break up data silos early on and make those accessible for AI.”
Other guests included Guy Taylor, head of data-driven intelligence at Nedbank in South Africa, who described how Watson is used by Nedbank to predict ATM machine outages and optimize custodians’ route to refill ATMs.
Think 2019 attendees also heard from Jack McCarthy, CIO at the State of New Jersey Judiciary, on using AI to reform the criminal justice system. In the past, a bail amount was set in determining if an arrestee was able to go home before appearing in court or sent to jail. With this approach, 14,000 people out of approximately 45,000 were being sent to jail each year for sums under $2,500 that they could not afford. The New Jersey Judiciary decided to use AI to automate risk profiling in helping judges evaluate these situations. The AI system provides a score based on nine factors that the judge uses to make a final decision.
“We were able to go from a process that would manually take 4 hours and bring it down to 3 seconds,” McCarthy said. He emphasized that “ensuring that those decisions are fair is absolutely crucial.”
KPMG’s use of Watson OpenScale to develop trust in AI decisions
The question of trust and transparency is probably one of the most important hurdles organizations must overcome to scale AI. AI’s progress has been slowed because people don’t always trust what they see as a black box where the factors that determine an outcome are not apparent. And the risk of a world filled with AI black boxes is that important decisions made by AI to assist people will lack accountability.
Kelly Combs, Director of KPMG's Emerging Technology Risk practice, spoke about her company’s use of Watson OpenScale to increase the trust and transparency of AI decisions. “There’s a lot of excitement about AI, but also a lot that can go wrong,” Combs said.
To deal with this pressing challenge, KPMG developed a methodology called ‘AI in control,’ which rests on 4 key pillars:
- Integrity – understanding the lineage of data
- Interoperability – the technical robustness of the AI tool
- Explainability – ensuring the outcome is what we expect
- Fairness – checking that the AI system is free of bias
Combs noted how establishing the right level of trust in AI is top of mind at the board level of many companies as AI continues to be a strategic imperative. “It’s great to have tools like Watson OpenScale that gives us actionable insight to detect unfair outcomes.” Ultimately, Watson OpenScale “lets us understand what our AI systems do and whether we can trust it,” she added.