Want to invest wisely? Check your prior beliefs at the door

The AI Playbook: 6 steps for launching predictive AI projects

5 ‘extraordinary turnarounds’ for global well-being

Credit: Mimi Phan / Irina Shi / Shutterstock

Ideas Made to Matter


The business value of curated model collections


It’s common practice for museums and galleries to have curators who are responsible for acquiring, caring for, and developing collections so they remain key assets. Engineering organizations can take a page from the art world and embrace curation concepts to better manage their own models.

Companies building complex systems such as naval battleships, jetliners, and cars are switching from traditional, paper-based processes to digital and model-centric engineering workflows as part of digital transformation. Product-related models, simulations, and data are used to digitally represent design concepts, mimic system behavior, and test and optimize product performance under varying conditions.

While model-based practices are now a staple in product design and engineering workflows, models are seldom managed as an enterprise collection, limiting their value, according to Donna Rhodes, a principal research scientist at the MIT Sociotechnical Systems Research Center. By establishing new curator roles and curation workflows, companies can maximize the utility of models as a true enterprise asset, enabling greater reuse, improving design efficiencies, and seeding more innovation.

“We’re talking about management control and preservation as you would in an art museum, but also about the active enhancement of models in the collection,” Rhodes said during a recent MIT SDM Systems Thinking Webinar Series event. “Models are becoming so valuable to enterprises that ultimately they may have greater value than the physical assets themselves — that is, if they can be repurposed.”

What is model curation?

Not every model belongs in a curated collection. For example, the myriad models used in early-stage product development aren’t necessarily applicable elsewhere. However, mature models that function as a digital twin — a digital representation of a physical asset or system — are valuable additions to any curated repository, Rhodes said. In addition, models that can be easily repurposed for functions beyond their initial intended use have a place in a collection.

A curated model collection requires some foundational building blocks. Governance is key and should include policies that specify access controls and permissions, and a board or committee that makes key decisions. A governance board specifies what models are allowed to come into a collection (a process known as accession), what models need to be removed (known as deaccession), the valuation of key assets, and the strategic road map for evolving the curated model collection. Establishing the proper cybersecurity controls is another requisite, as is adopting technologies that enhance usability in areas like search and model discovery.

While curated-model use cases are still evolving, one example with broad applicability is what Rhodes referred to as a “digital demonstrator” — a capability that could allow a company to demonstrate an offering to potential customers as part of a competitive bid process. “Imagine reaching into a model collection, pulling out some models, and putting them together in such a way that you could actually demonstrate what a solution looked like,” Rhodes said.

What to consider when building a curated model collection

As organizations advance on this journey, several factors are central to establishing a collection of curated models and ensuring that it delivers desired benefits, Rhodes said.

Issues to consider include the following:

Establishing trust in the models. For models to be useful, potential consumers must trust that they are credible, especially when they’ve been developed by someone else. Model verification and validation practices can help build that trust. It’s also important to create mechanisms that provide visibility into model origins, establish transparency into how a model was created and how it might have been repurposed, and context for related decisions.

Consumers that have more experience and expertise working with models are more likely to view them as credible, as are those who have a higher propensity for trust in general, Rhodes said. Model credibility is also influenced by an individual’s trust in the model developer and how easily models can be discovered and retrieved from a repository. “We’ll increasingly find ways that we can curate models for specific consumer needs, and the ability to do so will be very important in whether this idea of a shared model repository would be useful to an enterprise,” she said.

Related Articles

How data fuels the move to smart manufacturing
MIT expert recaps 30-plus years of enterprise architecture
10 capabilities to accelerate digital transformation

Addressing governance challenges. Creating strong governance standards is one way to help establish trust in curated model collections. Organizations should establish a board of standing members, domain experts among them, that is tasked with making decisions about what models come into a collection as well as what models are removed. Rhodes cited work supported by the U.S. Department of Defense Systems Engineering Research Center to help establish criteria to determine whether a model is a candidate for curation. Pertinent attributes include relevance to the enterprise, completeness of the metadata and documentation accompanying a model, completeness of the model pedigree, the potential for reusability, the uniqueness of the model, and the economic business case.

Governance efforts also need to recognize that people won’t want to give up their own models and often prefer to use them on an individual or localized program level. As a result, companies may want to consider a federated approach, which keeps some models local as opposed to centralizing them in one enterprise repository. “Think of it as a system of systems of models in repositories,” Rhodes said. “There are many things we are thinking about in terms of the composition of governance.”

Considering technology as an enabler. The entire concept of a curated collection hinges on users’ ability to discover and access models. Emerging technologies in areas like data science, visual analytics, and machine learning can help consumers discover pertinent models and aid in the curation and reuse processes, Rhodes said. For example, augmented intelligence could provide context for how a model could be repurposed for an entirely different context, facilitating human decision-making, she explained.

“We know if we’re going to achieve this idea of enterprise model collections, the value of these collections needs to outweigh the investment it takes to create and maintain them,” Rhodes said. “But I think they can be very powerful, especially when we get into this world where every system has a digital twin.”

Watch the webinar “Investigating the Future of Curated Model Collections”

For more info Sara Brown Senior News Editor and Writer