AI Formulators are coming are you ready?

It is undeniable that Artificial Intelligence (AI) is poised to disrupt nearly every industry and the cosmetic industry is no exception. As a cosmetic chemist who has been in this industry for more than 30 years, I’ve seen a number of promising trends come and go with little real impact on my job. Sure epigenetics or the microbiome are interesting areas of science, but they haven’t actually changed the formulator’s job much. We still use the same occlusives, humectants, and emollients to make skin care that people want to use.

So, when AI came on the scene I was skeptical. It seemed more like a marketing gimmick than something that would really impact formulators. And the early entries on the market only seemed to reinforce my belief. This pertains to companies that collected quiz data and maybe skin or hair measurements, then put the information in an “AI” to predict precisely the products that are perfect for the consumer.

As a formulator, I know whether a product is good for a consumer or not is much more complicated than an AI is ever going to be able to predict. I always wondered how would knowing a consumer thought they had sensitive skin would impact the amount of petrolatum put in a formula? Honestly, tools like these are little more than marketing gimmicks as are the “customized” formulas that result from using them.

There are also a number of formulator-focused tools. These are meant to help cosmetic chemists choose chemicals, track ingredients, solve stability problems, and create formulas. While these tools offer some potential benefits, they also face limitations due to the industry’s secretive nature and data scarcity. It’s difficult to create a good stability AI prediction tool with the limited amount of stability / formula data freely available. And it’s unlikely that companies want to share their formula information with everyone else in the industry.

But chatbots, these tools are different.

Chatbot Chemists?

I first heard about Chatbots, and specifically ChatGPT, back in February of last year. This is when version 3.5 was introduced generally to consumers. The premise of these tools is simple. You start a discussion with them and they continue it. The responses they give are powered by a Large Language Model that is really good at predicting the next word in a sentence. When properly trained, these tools can give responses that seem amazingly accurate.

Of course, I was curious. The first thing I did was to ask ChatGPT to give me the formula for a shampoo. In fact, I typed “Give me a formula for a shampoo.” And it did exactly that. Here is the exact formula it told me.

• Pure Castile Soap – 1/2 cup
• Distilled Water – 1/2 cup
• Jojoba Oil – 2 tablespoons
• Essential oils – 30 drops

Now, as a professional cosmetic chemist, this made me laugh. First, it gives measurements in cups and tablespoons. And basing a shampoo on soap and jojoba oil? That’s ridiculous. I was not impressed. It wasn’t surprising to me because the chatbot was trained on data from the Internet and most of the formulation information there is created by DIY formulators who make products in their kitchens.

Getting Improved Responses

But I learned that a chatbot like this is a tool and there are better ways to use it. Rather than simply asking it for a shampoo formula, you get better results by giving more information about what you want. If you tell it that you are a professional chemist and that you want a formula that would be suitable for producing in a factory, then it gives a more reasonable response.

When I specifically started the conversation as “I am a professional cosmetic chemist. Please give me a shampoo formula with all the ingredients listed in terms of percentages. Additionally, list the manufacturing procedure and the appropriate specifications.”

With this input, I was amazed at the results. The formula that it gave me resembled a real formula on the market.

• Water – 60 – 70%
• Primary Surfactant (SLS or SLES) – 10 – 15%
• Secondary surfactant (Cocamidopropyl Betaine) – 5 – 10%
• Thickeners (Hydroxypropylmethyl cellulose) – 0.5 – 1%
• Conditioning agents (Polyquaternium-7) – 2-5%
• Humectants (Glycerin) – 2-5%
• Preservatives (Phenoxyethanol) – 0.5 – 1%
• pH Adjuster (citric acid) – qs
• Fragrance – 0.5 – 2%
• Special additives 1-5%

Now, this is by no means a finished formula and I disagree with some choices. For example, glycerin should not be used at a high level in shampoos! And Polyquaternium 7 at 2% is too high in my opinion. But these are all things that can be worked out at the bench. The point is that the AI can come up with a reasonable starting formula. And if you tell it specific ingredients in your formula, it can be even more useful.

If you are just starting out in the lab or are asked to formulate a type of product you don’t have a lot of familiarity with, this AI tool can be an amazing help.

Uses of AI for Cosmetic Chemists

Beyond getting starting formulas, some other specific tasks that AI can be used for include:

Learning and Research – AI tools like Elicit.com can simplify complex research papers, making them more accessible. Or you can just put the abstract in a chatbot and have it explain it in more simple terms.

Problem-Solving – AI can suggest potential solutions to formulation or stability issues. Whether it is new test methods, stability problems, or production challenges, AIs are able to brainstorm any number of potentially useful options.

Data Analysis – AI can help summarize and analyze large data sets, aiding in decision-making. You simply enter your data into the chatbot AI and it will find patterns for you.

New Product Ideas – Chatbots are really good at coming up with hundreds of ideas for products. Of course many of these are bad ideas and not workable, real-world products, but it can also come up with some interesting products that could be successful in the marketplace. The ideas it creates are certainly as good as you get from an all-day, off-site group brainstorming session.

The Limitations and Cautions of AI

Of course, AIs are not perfect yet. And while they offer numerous advantages, it is not without its problems. You should be aware of the following whenever you use an AI assistant.

AI Hallucinations – AI can often provide inaccurate or misleading information. It is trained on information from the Internet. And often, the Internet is just wrong. So, if you are trying to use it to learn new concepts, just realize some of the information it tells you is incorrect. Always double check facts. I would note that while this is a significant drawback in using AI, a similar thing happens when learning from people. Individuals can present misinformation as fact just as readily as an AI.

Data Privacy – AI tools use your data when responding to other users. Therefore, sensitive company information should not be shared with them. ChatGPT and other AI products don’t keep secrets!

Mathematical Limitations – The way that AI chatbot works is through a Large Language Model. And as such, they don’t really know what they are answering. They just give you a good prediction of what word should come after another. This means that it is not reliable for complex mathematical calculations.

The Evolving Role of Cosmetic Formulators

With the introduction of AI into society in general, the job of a cosmetic formulator is going to change significantly. I know some in my field will want to ignore and dismiss these tools but they do so at their peril. It’s imperative to adapt to these changes to avoid becoming obsolete. AI is not just an emerging trend; it’s a tool reshaping our approach to formulation and product development. It is not going away. This means that to keep up in this industry you need to learn to use AI to remain competitive in your career.

The Future of AI in Cosmetic Chemistry

At present, AI is not a replacement for human chemists. It still can’t reliably mix chemicals together and make emulsions. However, it is an important tool that can enhance their productivity. This may change over time, but at present the most important thing for formulators in all fields to do is to learn how to use these tools to improve speed and efficiency in the job. The people who do will best thrive in the inevitably AI powered future.

Contact and Further Resources

For more insights and a list of recommended AI tools, feel free to contact me at perry@chemistscorner.com.