Ever Wonder Exactly What AI Learns and What it Does Not?

Tilak Kasturi , CEO & Founder, Predii, Inc.

This is the second in a series of Predii Perspective articles addressing the concerns surrounding AI. We are looking to help the industry understand how artificial intelligence can be leveraged to drive benefits for your organization in a safe and secure way.??

One of the more common discussions we have with customers is around protecting intellectual property (IP).? IP can be: proprietary data Q&A interactions with an AI model, proprietary configurations for your business made to the AI Model, and insights produced by an AI Model. Precautions need to be taken while using “licensed content” from 3rd parties – such as OEM content - within your firewall. Your IP is a valuable asset, and the idea of AI certainly raises questions around the exposure of data as well as the potential for your data to be leveraged to make your competitors smarter.?

The information provided here attempts to address these concerns: What does AI Learn? How can you protect your intellectual property? AI can be implemented in ways that allow you to unlock the value of this new technology while protecting your business and competitive advantage.?

LET'S MEET AT AAPEX, Nov. 5-7: If you want to discuss more about how your intellectual property is protected when adopting AI, please contact Mark Seng at [email protected] and let’s meet in Vegas.??

What does AI learn??

AI Assist systems use large language models and customer-specific data to answer questions. Large language models are trained on datasets.? AI Assist systems can understand questions being asked using a fine-tuned large language model. “Fine-tuned” domain-specific large language models are like Automotive specialists who understand the automotive domain knowledge (industry language, industry terms, industry context) and can dialog with a technician, service advisor, parts counterman or even the consumer. AI providers like Predii use their own internal data assets (not customer data) and valid public industry sources to fine-tune their large language models.?

AI providers should ensure LLM models are not trained on customers’ proprietary data and provide necessary guardrails to isolate the proprietary data assets. AI providers should ensure AI Assist systems leverage proprietary datasets only for the purpose of answering questions.

How can you protect your intellectual property?

Intellectual property is a competitive advantage for a business. Intellectual property assets grow over time and customer-specific AI Assist systems will become smarter as they leverage your IP - for your business purpose only - in a protected manner.??

AI providers should offer transparency into how they are protecting IP assets for any single business. There are some easy ways to provide this assurance:?

1. Regular Audits: Auditing and monitoring systems that track what data has been used for fine-tuning and ensure that customer IP is not being misused.??
2. Data Usage Agreements: Clearly outline in your contract how data will be used, for what purpose, and under what security conditions.??

AI Assist systems should have isolated pipelines enabling IP protection while getting the benefit of leveraging automotive domain-specific LLMs for your business as shown below.?

A real-world example?

Predii’s mission focuses on developing automotive-specific AI assist solutions capable of interacting in an “industry aware” manner when responding to questions from service advisors, parts counter professionals and technicians.? At the same time, your expectation is that Generative AI deployed inside the business produces insights and supports decisions that are specific to that business based on its own proprietary data without those insights and decisions being shared outside the organization.??

Yes, artificial intelligence – by definition - “learns,” but AI providers can put safeguards and restrictions on what is needed to make the AI stronger and more accurate while ensuring that specific data and private information is not retained and shared with the next user.? “Learning” can be limited to the customer-specific private instance.

For example, in the world of repair orders, the data will contain customer complaints or symptoms and the component names which were replaced to fix the problem. The AI Large Language Model (LLM) can be allowed to retain the component name and the symptom wording to recognize them going forward, but the correlation between the two – the actual “fix” for the problem – is not. The known fix is an insight allowed to be leveraged by the owner of that data only. Additional data points like pricing, labor times, shop name, location, and diagnostic notes created by the technician are also considered proprietary insights retained only by the data owner as well.??

The good news is that your proprietary data – in its “raw” state or even post normalization & aggregation by the AI model – is not used to assist or influence the insights or decisions of another company. You as the data owner and customer of the AI solution, are assured that your intellectual property is safe, secure and protected.? While the AI model’s taxonomy – or the dictionary of recognized terms – does learn and grow, your customer-specific information is never shared or aggregated with another customer’s insights.??

For example, the automotive-specific Predii Large Language Model is fine-tuned to recognize our industry’s language, the types of questions service advisors, parts counter people and technicians ask, and what data is required to answer those questions. The actual answers to the questions or the predictions the model is making are all reliant on the breadth and strength of the individual customer’s datasets and their use is restricted to the specific customer as well.??

These are just some of the ways to address concerns around making sure your data is safe, secure and not being leveraged to help the competition and to know that your intellectual property is not being aggregated in any way to assist or influence the training of, or insights from, other AI solutions or “bots” being used by other businesses.??

The key is in the AI-provider partner you select. Be sure they are at least following the guidelines included here and are focused on maintaining the security and privacy of one of the most important assets to your organization – the data!??

LET'S MEET AT AAPEX, Nov. 5-7: If you want to discuss more about how your intellectual property is protected when adopting AI, please contact Mark Seng at [email protected] and let’s meet in Vegas.

#automotiveinnovation #automotiveai #autocare #mema #ETI #SEMA #aftermarket #aftersales #fixedoperations #AAPEX

Big thanks to our board advisors: Dr. Michael Becker , Arwed Niestroj , John Redgrave , Tim Eisenmann , Babak Hodjat , Adam G.

AAPEX Specialty Equipment Market Association (SEMA)