Could Unsupervised Learning be THE Key to Enterprise AI Adoption?
84% of Enterprise AI projects have not scaled – is supervised learning the problem?
By Ryan Welsh, Founder & CEO, Kyndi
Enterprise AI requires a new AI approach. That is our belief here at Kyndi, and that was a central theme at NeurIPS 2020, which was held virtually from December 6th – 12th. Among the AI community, it is now an accepted fact that supervised machine learning is prohibitive for most business situations. That’s because supervised machine learning requires too much technical knowledge, requires too much data (both quantity and quality), and adds too much business risk in most cases because it is a black box.
First, enterprise AI requires new AI technology because supervised machine learning requires too much technical knowledge. In a KPMG survey, the five companies with the most mature AI capabilities had, on average, 375 full-time employees working on AI and paid them $75m annually. Those same companies expected to increase their AI staffing to between 500 – 600 full-time employees by 2022, which would cost between $100m – $120m annually. So for the Global 2000 enterprises to develop mature capabilities, they need to hire 1.1m full-time employees to work on AI and pay them $220B annually. Unfortunately for the Global 2000 and the millions of smaller businesses, there are less than 500,000 AI professionals globally.
Additionally, supervised machine learning requires too much data from both a quantity and a quality perspective. Data is the lifeblood of AI, and while the world is awash with data, a lot of it isn’t suitable for feeding to machine learning algorithms. Current estimates are that “data-wrangling of various sorts takes up about 80% of the time consumed in a typical AI project.” That’s because training machine learning algorithms require large volumes of carefully labeled examples, which humans have to apply. So of the $220B the Global 2000 will be paying full-time employees to work on AI, $176B of that will go toward mucking around with data and not adding any business value.
Finally, supervised machine learning adds too much business risk in most cases because it is a “black box.” Supervised machine learning, specifically deep learning, is largely a “black box.” Once a neural network has been trained, not even the developer knows exactly how it is doing what it does. Embedding these “black box” systems into business operations introduces risks that far outweigh the return.
Thus, while supervised machine learning has delivered value in specific, narrow use cases, most business use cases require a new AI approach. At Kyndi, we’ve pioneered such a system. It is an unsupervised method with explainability built-in from end-to-end. Our comprehensive APIs allow any software developer to integrate enterprise-ready AI into the applications that power their business, no Ph.D. needed (see IDC Planscape case study). The Kyndi Platform simplifies and accelerates Enterprise AI adoption, reduces cost and risk, and delivers intelligent systems that are flexible enough to meet evolving needs.