Enterprise AI: How to Meet the Challenges
By Dr. Anand S. Rao, Global AI Leader, PwC
As artificial intelligence (AI) moves from consumer to business uses, companies will face significant challenges. Here are some guidelines for adopting AI technology in your company.
Over the past several years, AI has moved from a research curiosity hotly debated in academic circles to a topic that is widely discussed by national governments, international bodies, and the boards of numerous companies. Consumer adoption of AI in its various forms, ranging from book and movie recommendations to voice assistants and chatbots, is moving rapidly. As a result, companies focused on direct-to-consumer products—whether they are physical assets or informational—are leading the charge in AI adoption.
However, as AI transitions from B2C products to B2B or B2B2C products, enterprises must deal with significant challenges. This article outlines some of the key challenges faced by companies that are adopting AI for their business customers.
We recently conducted a survey of CEOs and their top challenges in adopting AI. More than three-quarters were concerned about increased vulnerability and disruption to their businesses from AI solutions. A similar percentage felt that there is a significant risk due to bias and lack of transparency from AI, and they stressed the need for AI governance. As enterprises decide to adopt—or at least experiment with— AI, they are faced with additional challenges.
1. Where should you start?
Different departments in an enterprise are approaching AI from different directions. CIOs and COOs often come to AI, especially natural language processing (NLP) and machine learning (ML), after they have implemented robotic process automation (RPA). Business executives in different functional areas—marketing, sales, product development, operations, customer service—arrive at AI from an advanced analytics perspective, moving from predictive to prescriptive and cognitive analytics. Chief data officers approach AI from the world of data warehouses and big data architectures.
Regardless of which direction you come from, in order to leverage the full benefits of AI, you need to integrate these different viewpoints to automate your processes, derive insights from massive amounts of data, and make better business decisions.
2. What data should you use?
In the era of big data and the internet of things (IoT), organizations are gathering, distributing and storing huge volumes of data—both structured and unstructured. However, compared to consumer data, the availability and use of business data is still very fragmented and has lower levels of veracity. In addition, most enterprise data is not labeled, and there is very little, if any, open-source business data.
As a result, one of the biggest challenges in enterprise AI is having enough labeled data to train machine learning models.
3. Who will build your AI systems?
With AI taking center stage in board room and senior management discussions, it’s essential to cut through the hype and present a realistic view of what can and cannot be accomplished by AI—and within a specific timeframe. Acquiring, training, and retaining critical talent who can be multi-lingual (business, technology, data, analytics and AI literate) is a big challenge for enterprises.
Unlike consumer AI, the domain knowledge required to build enterprise AI is highly specialized. Buy-build-partner decisions are also critical for rapid adoption.
4. How can you build safe and robust AI?
To address all the above questions, you have to progress toward building AI systems in production. However we are still in the very early days of full-scale, production-quality AI systems deployed on a 24x7 basis. Production-quality AI systems need to contend with issues related to the fairness, accountability, and transparency of their recommendations or actions. For example, an AI system that grants or rejects a mortgage needs to explain its rationale to both a consumer (most likely with little or no mathematical training) and a data scientist (arguably with a good knowledge of statistics or advanced mathematics).
Given the concerns expressed by CEOs, this is an area that requires serious consideration by enterprises management.
5. How can you make AI more human?
AI is often viewed (and sometimes feared) as a replacement for people and the tasks they do. While AI will take over some of the repetitive, routine manual and cognitive tasks humans perform, this technology should be viewed as augmenting, rather than replacing, human decision making. Often, the combination of humans and machines does a much better job than either one could do on their own. Certain actions are better done by humans (e.g., showing empathy to customers), while others (e.g., analyzing historical patterns of stock market movement) are handled more effectively by AI machines.
As a result, it will become critical to deploy human-centered AI: designing AI systems that learn from human interactions, and systems that allow people to teach AI systems (e.g., annotating documents or images for labeled data set creation).
In summary, adoption of AI in enterprise applications or B2B scenarios is still in its infancy. To increase AI adoption, organizations need to adapt techniques and practices that have worked well in consumer-oriented AI, but also focus on issues related to fairness, accountability, transparency and ethics. Taking full advantage of the opportunities offered by artificial intelligence, while mitigating some of the risks, will ensure the continued success of this technology.