Unfortunately, Commercial AI is Failing. Here’s Why.

What happens when a product fails to justify its hype? It fails. In the 90s, interest was booming and many companies wanted to take advantage, but could not. Only a few survived, and when the dot-com bust happened, those companies had to shut down due to lack of effort. In the mid-2000s, cloud computing was the hot topic. Several companies tried to gain momentum but failed because they could not move their data to the cloud.

Commercial AI products are not booming as expected. This is leading to disappointments not only to the artificial intelligence developers but also to the industry and businesses who wanted to employ these products. This is also known as AI fatigue, when a product is unable to deliver the results as promised by its hype, informational, and sometimes, misinformation. For example, when companies were developing a chatbot for Facebook's Messenger, they observed a 70% failure rate in handling user requests. According to a research report by McKinsey Global Institute, 45% of work activities can be automated, of which 80% is enabled by machine learning. Companies in sectors like manufacturing and health care have captured less than 30% of the potential from their data.

The Reason Behind The Failure

One of the reasons why AI products fail to make an impact on a commercial scale is the lack of deep learning. Deep Learning is a subset of AI. Most often, it is used to classify data problems that involve finding data patterns. But many in the AI industry have found it challenging to build artificial intelligence products with deep learning. This issue can be tackled by producing scalable AI products.

What Is Scalable AI?

If an AI is both accurate and powerful, it is known as scalable AI. In this context, powerful refers to AIs ability to adapt itself to any business model. For example, a medical imaging AI should work in different clinical settings and for patients worldwide. Silicon Valley investor Andreessen Horowitz, who worked with a range of AI companies wrote in his recent article about the lack of scalable AI. It is becoming a challenge in the AI industry to make a program scalable for commercial use, once it's out of the lab. If we look deeper, the problem is not with AI, though. The problem lies in the way these AI applications are made for commercial use. On paper, it has a different perspective than when it is put to use.

To build scalable AI for commercial use, the industry needs to shift its focus from data quantity and AI accuracy to data quality, diversity, powerfulness, and knowledge about the industry to fix the problem.

Data Quality

Scalable AI cannot work with poor quality data. It affects both the accuracy and powerfulness of AI. Even a 1% error in data can impact AI accuracy. As a practice, AI practitioners must "clean" the data. Having effective data cleaning methods to improve data quality is a significant factor to build a robust scalable AI.

Data Diversity

A globally diverse dataset is essential for testing and validating the AI. As scalable AI should be robust and powerful for it to function universally, the need for additional investment and efforts to take a global approach has arisen. In healthcare, AI can be biased to a certain section of people or clinics. Healthcare problems are global, hence it is crucial to take this approach. But collecting global data is complicated. The easiest way for AI companies is to collect data from one or more clinics to have a large dataset, preferably from a prestigious clinic that has larger patient data.

Why is scalable AI important? Because people have started doubting the credibility of AI in the commercial space. An industry transformation is needed for companies to believe in AI and employ them. This means a big change in the company's organizational DNA. AI companies that will succeed in making commercial AI will be the ones that focus on creating scalable AI for business learning and business functions. This will harness the transformative power of AI.