Klil chairman: 'The high-tech-industry gap: 80% AI accuracy? In industry, that’s failure'

High-tech companies and startups have been striving to incorporate artificial intelligence (AI) technologies into their operations, but transferring these solutions to the factory floor poses significant challenges. Tzuri Dvush, a chairman who bridges both the high-tech and industrial worlds, explains that the gap between the idealized environment of AI development and the harsh realities of manufacturing plants is a major stumbling block for adoption. "Connecting the two is full of challenges," he notes.

One of the main issues lies in the discrepancy between AI demos and their real-world performance. Tech professionals often fail to grasp the complexities of manufacturing, such as noisy, dusty, humid, hot or cold conditions, and constant movement from forklifts and cranes. These elements create an environment that is far removed from the controlled, air-conditioned office setting where AI solutions are initially tested.

Additionally, terminology and regulatory standards pose hurdles. Entrepreneurs frequently lack familiarity with the strict regulations governing manufacturing plants and do not speak the professional language of the industry. Terms like "mold" have different meanings across various sectors, requiring a shared understanding to establish the right connection.

Dvush advises that developers use AI tools to familiarize themselves with industry-specific regulations and terminology before engaging with manufacturing plants. This would help bridge the initial communication gap. Furthermore, high-tech industry practices like beta launches and iterative improvements do not translate well to manufacturing environments.

Start-ups often claim an 80% accuracy rate or a learning system, but in manufacturing, even a small error rate can lead to serious consequences, such as defective products, production line shutdowns, damage to expensive equipment, and reputational damage.

The "token economy" and the cost of AI model usage also pose challenges. Without a predefined payment model and clarity on who bears processing costs, factories may face unexpected and high monthly bills. Moreover, legal and regulatory arrangements regarding data ownership fed into AI systems must be established upfront.

Despite these obstacles, the adoption of AI in industry is deemed crucial, and success hinges on realistic expectations. Dvush outlines several key principles for entrepreneurs and factory managers: they should stop promising broad solutions and focus on a specific, deep-rooted problem; define clear success metrics (KPIs) in advance; start with a small pilot in a defined area that doesn't disrupt critical production; and consider partnership models that reduce initial costs or share risks.

Despite these warnings, the market recognizes the potential in overcoming the adoption gap. Incubators like i4Valley and investment firms like IL Ventures are investing in ventures that successfully bridge the gap between AI technology and industrial applications. Dvush concludes that entrepreneurs carry a responsibility beyond their individual projects.

Each AI system failure on the production floor erodes industrialists' trust, making it harder for future entrepreneurs. Transparency about risks and clearly defined boundaries from the outset are essential for building long-term trust.