Executive Summary
Many TLT readers are still skeptical about using artificial intelligence (AI) in their research and development (R&D) work. Many recognize AI as useful for collecting and organizing data, as well as analyzing reports, and some companies are working on ways to use this rapidly evolving technology internally. While a few readers report they have no concerns about using AI, issues such as reliability, trustworthiness, security and safety impacts were listed as potential concerns with using AI for R&D.

Q.1. What are some strategies you or your company are using to incorporate AI/generative (Gen) AI technologies in R&D?

Try it and see.

AI is a hype in our field. I see no meaningful benefit beyond “toy” problems. 

None at the moment.

We don’t use for lubricant formulations but have used in other areas. Many times they are very useful. You can have some initial idea without going through expensive experimentations.

Data sifting and repetitive paperwork tasks.

We have used some AI for multi-factor design of our latest greases, particularly in the formulation of a biobased grease meeting the NLGI HPM certification. The use of current AI for simulation of test performance is still in the formative stages; you still need actual testing.

Nothing specific yet.

We use the technology at the exploratory/feasibility stage and as a starting point for a deeper literature review. AI is never used to make critical business or engineering decisions.

Not much. Impact of AI is not yet seen other than rapid answers to some basic commercial and trade information.

Our company has started projects on such innovations.

Looking for limits and representative values in oil analysis reports. Resource to know root cause of failures and to make more accurate diagnostics.



More support in nature to do mundane tasks or accelerate report generation or topic searches. No resources to really develop outside of pure business functions, which I would not classify as R&D, like market trends, etc.

The main action was to have an internal AI licensed package available to the team and provide some clues to get the best results.

We use AI for oil analysis.

Utilization of proprietary AI software and existing software such as ChatGPT.

We use AI and GenAI in R&D mainly to accelerate research review, support ideation, analyze technical data, optimize designs and draft documentation faster. The goal is to reduce time spent on repetitive tasks and improve decision speed, while keeping expert review in place for accuracy, confidentiality and quality.

In the future we plan to use GenAI or AI Alchemi, or both for our research.

Use AI as a sounding board for R&D projects to get a general idea of the pitfalls.

Concluding some results about performance with the real data to claim the next strategic accomplishments, related to biodegradable lubricant.

Our strategy is to effectively use AI in our R&D as much and as quickly as possible. Although it is still in early stage, our direction is firm. AI will not only be used in AddPack development but component development.

Identifying the potential gaps in AI-based R&D.

Would you trust the output of AI or GenAI simulations of tests to be representative of real-life performance?
Yes 44%
No 56%
Based on an informal poll sent to 15,000 TLT readers.
We use machine learning as a tool in many projects to see whether it yields better insights than traditional statistical methods.

Preliminary bibliographical studies, methodologies and coding.



We really have not used AI for R&D at this point.

Still learning.

Current AIs are generally limited to language models, which are not designed for technical problem solving beyond pure mathematics.

I do not use AI.

At this stage not a lot. We tend to use available AI programs through Meta, Google, etc.

I do not know about official strategy, but I try to avoid it.

Q.2. What are your concerns in the use of AI/GenAI for lubricant formulation and testing?

Hallucinations.

No concerns; it’s an opportunity.

The bubble issue with AI.
 
The amount of raw materials of different molecular makeups and how making an “AI” grouping of any particular molecules could potentially misrepresent said molecules in different lubricants base oils/charges/interactions, etc.

Having a good set of data to train the AI model.

Not always reliable.
 
Believing “correlation” means “causation” and getting pointed in wrong directions.

Arbitrary output.

I don’t see it as capable enough at this time because the available database is limited. Because of trade secrets, no one is going to share breakthrough data. Sometimes you just need 45 years of experience to prevent going down an R&D rat hole when developing a new lubricant.

Garbage in, garbage out. I have used AI and it is only as good as what it can find to analyze. Biggest concern is senior leadership not understanding AI is only as good as the people who use it and can benefit as a tool. Many think AI will replace people, which is a bad assumption, especially around key resources.

The tendency to trust the results without doing due diligence to confirm them is my greatest concern. There are also concerns around securing intellectual property (IP). Many employees do not have a clear understanding of our policies around submitting information in an unsecured environment and what tools have been made available to safeguard this information.

Formulation is an art by itself. My guess is for a real life formulation situation, it is hard to use simulated information for lubrication formulation. Wrong predictions for transportation, especially in automobile and aviation, may create real safety issues.

The primary concern is that the technology is not yet sufficiently refined to demonstrate its viability.

Sometimes it is too general and not specific for each component and operational context.

So long as bulletproof algorithms are crafted, AI could have a positive influence.

Incomplete understanding of finished fluid spaces that take hands on experience to develop. Fundamentally it can set basic shapes but not much better than guidelines provided by additive companies. Testing could be more interesting as the massive amount of data export from tests could.

As lubrication is not part of the core of our industry, besides it is a very critical subject, the main concern is to filter the “commercial” info from what really matters to a decision maker.

My concern is the same about using AI for everything that it is currently being used for—that there must be follow-up with the presented answers/solutions to prevent errors. I’ve seen multiple times, in different applications, that AI has been incorrect. Also, people become dependent on AI to solve their problems and will eventually be 100% reliant on AI information.

Test automation.

What benefits of using AI and GenAI in lubricant formulation and testing R&D have the maximum impact?
Time saving in reducing unnecessary or laborious tasks          68%
R&D acceleration due to rapid testing and virtual experimentation 59%
Cost saving                                 46%
Sustainability or reduced CO2 footprint                             16%
New and surprising insights                           43%
None of the above—GenAI doesn’t have a place in lubricant product development 19%
Other not mentioned above                                 3%
Based on an informal poll sent to 15,000 TLT readers. Total exceeds 100% because respondents were allowed to choose more than one answer.
Doubts that problems may arise because they are unable to make sound judgments.

The key concerns are poor data quality, weak interpretability, false correlations and unreliable generated outputs. In lubricant formulation, small compositional changes can strongly affect tribological and rheological behavior, so AI recommendations must be validated experimentally. We also worry about confidentiality, compliance and the risk of GenAI producing technically plausible but incorrect guidance.

Answers that seem convincing have to be supported by real test data.

In tribology and finding substitutes for next generation technology.

As AI application in component and AddPack development is still in its early stage, we have to carefully choose AI tools which are capable of doing the job. Hopefully there will be more suitable AI tools available in the market in the years to come so we can pick the right one for us.

It should always, in the last instance, be approved by human intervention.

Applying the algorithms to genuinely new data often results in relatively poor predictive accuracy.

Not concerned.

Accuracy of information.

Cannot accurately predict synergistic or antagonistic behavior of new molecules/new mixtures.

Real-world solutions are always multi-disciplinary. I have yet to see an integrated chemical/physical/structural solution. AI funding is currently put toward the low hanging fruit.

Nothing beats real experience, and that is what is relied on most in my company.

The quality of the information and variety of modern-day lubrication requirements.

Nothing. I have concerns with AI generally.

Editor’s Note: Sounding Board is based on an informal poll sent to 15,000 TLT readers. Views expressed are those of the respondents and do not reflect the opinions of the Society of Tribologists and Lubrication Engineers. STLE does not vouch for the technical accuracy of opinions expressed in Sounding Board, nor does inclusion of a comment represent an endorsement of the technology by STLE.