March 1, 2019
Artificial intelligence (AI) has certainly become one of the big technology trends in manufacturing. After a generation of talking about the possibilities, there is now enough computational power generally available to make AI for manufacturing processes a reality. The general definition outlines AI as computer systems taking over tasks normally involving human intelligence that can be automated. For manufacturing, these tasks include visual perception, speech recognition, navigation and rudimentary decision-making.
“Businesses spent $19.1 billion on AI in 2018, and that number is expected to more than double in 2019.”
Market intelligence research firm IDC estimates businesses spent $19.1 billion on AI in 2018, and expects that number to more than double in 2019. “Interest and awareness of AI is at a fever pitch,” says David Schubmehl, an IDC research director. “By 2021 75% of enterprise applications will use AI,” he adds, in fields that include intelligent process automation.
IDC says discrete manufacturing was the third largest industry for AI spending in 2018, with $2 billion designated toward a variety of use cases including automated preventive maintenance, quality management and recommendation systems.
An Uneven AI Playing Field
Such massive investment has serious implications for any field making the investment. And there is a problem: AI is dominated by a small group of very large software companies including Google, Microsoft, Intel and Facebook. Their seemingly unlimited budgets and industry cachet make them a magnet for the best and brightest in AI R&D.
“There is a universal need for AI services by businesses who lack access to capital to develop their own AI services, and independent AI developers [who] lack visibility and a source of revenue,” says the research team of Ben Goertzel and Gabriel Axel Montes in their introduction to “Distributed, Decentralized, and Democratized Artificial Intelligence” (sciencedirect.com/science/journal/00401625) for the journal, Technological Forecasting and Social Change.
Goertzel and Montes see this “uneven playing field” as one problem holding back widespread AI R&D. The other is a lack of interoperability standards. They propose development of a “distributed, decentralized, and democratized market for AI.” Not only would such an open market allow for more companies to do AI R&D, but it would also “create the infrastructure for coordinated action” between specific AI installations and stimulate the evolution of artificial general intelligence systems (AGIs).
Goertzel and Montes point to blockchain technology as a way to bring distributed, decentralized democratization to AI. Blockchain technology was invented by the anonymous person or group behind Bitcoin; it is a permission-less peer-to-peer system of creating a permanent, immutable, open and trusted record of transactions. In Bitcoin and other cryptocurrencies, the record of transactions is for a form of digital money, but the technology can be used to store and verify any kind of transaction or information of value. (For more on blockchain for engineering, see “Engineering’s Link to Blockchain;” DE, May 2018).
Blockchain developer Rohan Pinto—who spent years in AI research before turning to blockchain—agrees in theory with the idea of using blockchain for AI distribution, but sees four basic issues to overcome. First, AI is generally a centralized application, while blockchains are by design decentralized. Second, most AI work is proprietary while most blockchain work is open source. Third, AI in use is a “black box” while blockchains are generally transparent in operation. Fourth, AI is based on “probabilistic formulas, while blockchain is more deterministic in nature.”
Data Science Talent Needed
In operation, AI programs use domain knowledge as a foundation for doing assigned tasks. Engineering products conglomerate Hexagon Manufacturing Intelligence is taking a close look at how it should bring AI to its customers. “Domain knowledge is as important in AI as the specific technology,” says Claudio Simao, Hexagon chief technology officer. “We have identified 17 specific use cases for AI in our various divisions.”
Hexagon began as a metrology specialist, then branched out into related fields through both acquisition and research. It is now a $4 billion company, having acquired MSC Software (analysis), Intergraph (engineering design and operations), Bricsys (CAD) and Leica Geosystems (imaging and sensors) in recent years.
“There is a lack of management in AI, and a lack of knowledge on how to apply the correct questions.”
Simao says the manufacturing design industry has the same problems as most domains: a lot of data, but no clear method for processing it.
“There is a lack of management in AI, and a lack of knowledge on how to apply the correct questions,” he says.
Hexagon is investing in several areas including autonomous driving, which requires expertise in positioning, modeling, contextualizing and sensor fusion. “We have a team working on AI techniques and AI-driven analytics,” Simao says. Other technologies come into play for autonomous operations, including 5G and edge computing. “Edge-to-edge and edge-to-cloud is a major technology for the distribution of these applications,” Simao says, and AI drives it all.
How does blockchain fit in? “Blockchain can be an enabler for the lack of computing power and big data assessment,” says Simao. “Blockchain enables a more democratic access to data, [and] to compute resources.”
Simao also sees blockchain technology as ideal for guaranteeing the data that is constantly being gathered by sensors, whether it is for self-driving cars or robots in a factory setting. “If you have edge points, [the system] must certify new edges constantly.”
Simao sees plenty of opportunity ahead. “I believe blockchain will be an enabler of many things in this distributed world,” he says, citing such issues as security authentication, sharing data in open environments such as traffic and improving performance for such tasks as cleaning and meshing scan data. “AI for scanning is fantastic,” Simao notes, and blockchain can be the enabling technology for AI’s real-time interoperability with other systems.
Trusting the Black Box
Machine learning (ML) is a subset of AI where a system gains the ability to automatically learn and improve from experience without explicit programming. But there is a reluctance to adopt such systems on a wide scale, say researchers Thang Dinh and My T. Thai, because ML is a “black box” where it is difficult to understand what is inside and why. (See abstract here.)
“Decisions made by those systems are unexplainable to human users and thus cannot be verified or trusted,” write Dinh and Thai. “It is essential that we have an immutable trail to track the development of the data flow and complex behaviors of AI-based systems.”
Dinh (Virginia Commonwealth University) and Thai (University of Florida) say the solution is to have a blockchain “tracking every turn in the data-processing and decision-making chain.” Such a linkage of AI and blockchain will give confidence in the decisions made by the system, providing a clear trail to “trace back the machine decision process, making justification of those decisions much easier.”
With the rise of Industry 4.0 technologies including the Internet of Things, manufacturers will go from not having enough information about product to having more than they can cope with. AI routines need this data to be properly trained and to have enough real-time data to make appropriate decisions. Researchers Pedro Pinheiro and Mario Macedo at the Porto Institute of Technology in Portugal recommend industry use blockchain technology to chart what they call the “intelligent ambient”—the way early explorers charted the unknown seas. (See article here.)
“By looking at industries as an intelligent ambient,” Pinheiro and Macedo write, “where there is a big amount of data being exchanged and created, it is possible to gather data and create knowledge about the interactions.” By noting the types of interactions, and not just the discrete data elements, it becomes possible to “represent an entity in a network of entities” and improve decision-making.
More Hexagon Manufacturing Intelligence Coverage
About the Author
Randall S. Newton is principal analyst at Consilia Vektor, covering engineering technology. He has been part of the computer graphics industry in a variety of roles since 1985.Follow DE