ML Journal

A hybrid approach combining generative and symbolic AI brings control and explainability when interpreting complex, real-world data.

 

TAKEAWAYS:
● Trust in AI depends on explainability, especially when working with complex, high-risk manufacturing data.
● Generative and symbolic AI combine flexibility with deterministic logic to make complex data usable and reliably interpreted.
● Manufacturers can accelerate AI adoption by starting with targeted, explainable use cases rather than waiting for perfect data.  

 

Artificial intelligence (AI) is advancing rapidly across manufacturing, but adoption remains uneven. While many organizations are experimenting with AI, fewer have embedded it into day-to-day workflows such as quoting, engineering validation, or order processing.

The challenge is not access to AI technology, but trust in how it processes and applies large volumes of complex, fragmented data. It’s still a “black box” for many, and trust in outputs is impacted by mastery over that data.

Manufacturing leaders consistently point to the same concern: AI can accelerate tasks, but can it be relied on when accuracy is critical? In environments where errors lead to rework, production delays, or margin loss, even small inconsistencies have significant consequences.

For AI to move beyond experimentation, it must operate within the constraints and realities of manufacturing, mastering existing data and applying it in a controlled, explainable way. By combining generative AI with symbolic AI, your internal, rule-based logic can provide guardrails for generative AI outputs.

The Data Challenge: Structure Over Volume

A common belief is that AI requires perfectly clean and complete data. This often leads organizations to delay adoption, waiting until their data environment is fully standardized.

In practice, most manufacturers already have the data they need. The issue is that it’s unstructured and fragmented across enterprise resource planning (ERP) systems, CAD files, spreadsheets, emails, and the experience of engineers and sales teams.

This lack of structure creates risk in operational workflows. For example:

• Sales teams may interpret customer requirements differently, leading to inconsistent quotes.
• Engineers may need to manually validate configurations, delaying response times.
• Pricing decisions may rely on incomplete or outdated information.

Without structure, AI systems—particularly generative models—can produce outputs that appear reasonable but do not align with engineering or commercial constraints.

Improving data structure, even incrementally, allows manufacturers to apply AI in a more controlled and reliable way.

Generative Versus Symbolic AI

Different types of AI play distinct roles in manufacturing operations.

Generative AI (GenAI) is effective at working with unstructured inputs. It can interpret customer specifications, summarize technical documentation, or identify patterns across historical data. This makes it useful in early stages of workflows, such as translating customer requirements into potential solutions.

However, GenAI alone is not sufficient for execution. Its outputs are probabilistic, meaning they are not guaranteed to follow engineering rules, pricing logic, or manufacturing constraints.

Symbolic AI addresses this gap.

Symbolic AI: The Oldest Form of AI

Symbolic AI—a form of AI that has existed for decades—is grounded in the way manufacturing already operates: through rules, constraints, and engineering logic. It encodes expert knowledge into systems using structured relationships such as dependencies (“If option A is selected, option B is required.”) and exclusions (“These two components cannot be combined.”). This allows it to function much like a digital extension of engineering expertise, for example, ensuring that every configuration, design, or pricing decision follows the same logic used on the shop floor.

Unlike pattern-based AI, which infers possibilities from data, symbolic AI helps to guarantee that outputs are valid and manufacturable. In practice, this means that sales teams, for example, can generate accurate configurations without repeated engineering checks, and operations teams can trust that what is approved upstream can be built downstream without rework or delay.

Creating a Deterministic Solution

A hybrid approach combines these strengths into a deterministic solution.

GenAI can propose options based on customer inputs or historical patterns. Symbolic AI can immediately validate those options against predefined rules. Instead of relying on manual checks, validation happens in real time.

This combination is particularly relevant in operational workflows.

A Practical Example

Consider a quoting workflow for a highly configurable product.

• Before: A sales engineer interprets customer requirements, proposes a configuration, and sends it to engineering for validation. Multiple back-and-forth cycles delay the quote, and errors may still reach production.
• With hybrid AI: GenAI helps translate a customer’s RFP into a proposed solution. Symbolic AI validates it instantly against technical and regulatory constraints. The sales engineer reviews the result, focusing only on exceptions.

The outcome is faster response times, fewer errors, and reduced engineering bottlenecks without removing human oversight. The timed freed up can be focused on other, high-value activities.

Moving Forward Without Waiting for Perfection

Many manufacturers assume they need advanced data maturity or proprietary AI models before applying AI in operations. In reality, early value often comes from focused, practical use cases. Leading organizations are starting with areas where:

• The process is repetitive or time-consuming
• Errors are costly
• Rules and constraints are well-defined.

Examples include gathering business requirements from multiple sources and document formats, validating product needs earlier in the buying process, or assisting customer service teams in retrieving verified technical information.

These applications share a common characteristic: they combine AI-driven insights with clear rules and human oversight.

Importantly, progress does not require perfect data. Combined with rule-based systems, existing data can be structured and made more usable. This will also provide insights on how to make incremental improvements over time. This allows manufacturers to evolve the maturity level of the data as it is being used.

Addressing Organizational Skepticism

Adopting AI in manufacturing is as much a people challenge as it is a technical one.

Within most organizations, there’s a mix of perspectives. Some employees are eager to adopt AI, while others question its reliability, data security, or impact on established expertise. In operational environments, skepticism is often grounded in experience; teams have seen what happens when systems fail or sensitive data is mishandled.

Building trust requires transparency, secure data handling, and practical exposure. Organizations that succeed typically:

• Provide foundational education on how AI works with existing data and where it applies;
• Start with low-risk use cases that demonstrate clear value; and
• Ensure data is handled securely within enterprise-grade environments and governed within existing systems.

When teams see AI reducing manual effort, such as eliminating repetitive validation steps or improving order accuracy, while maintaining control over data and decisions, they’re more likely to adopt it in broader workflows.

The Path to Trustworthy AI

For manufacturing organizations, trust will determine how quickly AI moves from experimentation to everyday use.

Explainability is central to that trust. Teams need to understand how decisions are made, especially when those decisions affect production, cost, or customer commitments.

A hybrid approach—combining generative and symbolic AI—provides a practical path forward. It allows manufacturers to benefit from AI’s ability to handle vast amounts of data while maintaining the control and transparency required for operational reliability.

Rather than waiting for perfect data or large-scale transformation, manufacturers can begin with targeted applications that improve accuracy, reduce manual effort, and deliver immediate value.

In doing so, AI becomes less of a black box and more of a reliable extension of the systems and expertise that already drive manufacturing success.  M

 

About the author:

 

Jesper Alfredsson is Chief Product Officer at Tacton.