Adding More Data Isn’t the Only Way to Improve AI

By Pushkar Apte and Costas Spanos, SEMI

Artificial Intelligence (AI) has the potential to transform society and business in the 21st century. AI has already been deployed to in smart homes, online shopping, natural language processing to make life easier and more efficient for consumers. Still, in business applications, data-driven AI initiatives are struggling to deliver on their promise. Recent studies show that 77% of executives find business adoption of Big Data/AI initiatives a major challenge[1], and barely 10% of companies see significant financial benefits from AI[2].

The challenges are especially formidable in mission-critical applications like health and manufacturing. Not surprisingly, AI’s business impact remains limited despite the hype to the contrary[3]. For business leaders implementing AI projects, these hurdles can lead to frustration with poor results, excessive costs and delays, and even actual harm and lawsuits in extreme cases. Addressing these challenges is critical for realizing the full potential of AI.

HBR logoFundamentally, the intelligence of AI depends on the quality and quantity of data used for training the AI algorithm. Adding more high-quality data can help, but has its limits because there will never be a perfect dataset that captures every nuance of reality. Further, increasing amounts of data require more memory, more processing and more energy – an environmentally unsustainable proposition. The current business mantra of becoming data-driven is necessary but not sufficient. Rather, we believe that a paradigm shift is essential: data-driven AI must be coupled with a scientific and/or intuitive understanding of the physical world.

Our article in Harvard Business Review proposes four ways to drive this shift.

Logo

The focus of the SEMI Smart Data-AI initiative is to realize the full potential of AI in microelectronics, and we are exploring ways to enhance AI with physical insights. Running experiments in multi-billion dollar microelectronics fabs is extremely expensive and time-consuming. To address this challenge, we are exploring the use of physics-based simulators to compensate and complement experimental data in training AI models – an approach that will be critical in improving the effectiveness, sustainability and safety of AI implementations in microelectronics.

  1. “Companies Are Failing in Their Efforts to Become Data-Driven;” by Randy Bean and Thomas H. Davenport, Harvard Business Review; February 2019.
  2. “Expanding AI’s Impact With Organizational Learning;” By Sam Ransbotham, Shervin Khodabandeh, David Kiron, François Candelon, Michael Chu, And Burt Lafountain; MIT Sloan Management Review; October 2020.
  3. “Andrew Ng X-Rays the AI Hype;” by Andrew Ng, CEO and cofounder of Landing AI; IEEE Spectrum; May 2021.

Pushkar P. Apte, Ph.D., is Strategic Technology Advisor and leads the Smart Data-AI Initiative at SEMI. Costas Spanos is Director of CITRIS, CEO of BEARS, and Distinguished Professor of EECS at UC Berkeley. 

About The Author