Can AI be your formulator?

Before we start, I want to say I do use AI daily, and I love how it enables me to do things I couldn’t imagine myself doing, but we usually forget AI is a language model; it makes predictions about the next word and usually in a way that you would want to hear.

I think, like everyone else, I do get clients saying that they have “formulation” written in an AI chat with the ingredients they desire but that never exist because, unfortunately, fairy tears are not available in a commercial sense, and I find myself trying to explain what a formulation is, but let’s take this into its roots.

From a customer perspective, everyone is trying to cut costs and find a quicker way, because that is what is required of us at the moment. Do it now; it will either get expensive or AI will do it anyway.

So the unavoidable is prompting AI to make a cream that will work on melasma in 2 days, and believe it or not, AI will give you a “formula” without even citing any research on how skin or melasma works, with a DIY xanthan gum serum and all the most searched ingredients and ensure they are encapsulated, with a checklist of your formulator qualifications, such as whether they know the lamellar structure?

Ask for their CV and whether they have worked on any serums. Also, can they list every single product they made with similar ingredients without signing an NDA? Also, do they know the new regulations(which most of the time do not exist)?

In the days when everyone had a second brain, this became the norm, but really, what we are forgetting is that this is a language model that is just putting words together, and it will also act as well as you prompt it.

Formulating a cream vs. formulating a cream that has the same benefits as x and the same texture as x, utilising z and y as hero ingredients, is really different.

Even if the outcome might change, everyone is forgetting the open-source dataset that AI is learning from, so basically, the formula is really basic.

Also, I should give an honourable mention to Hallstar because Olivem 1000 is definitely AI's first choice to make a cream. I would assume a British pharmacopeia cold cream would be a more reliable source.

Unfortunately, this is not a second brain to me; this can be my brain in the first month on the job. I think overall we are forgetting what it means to learn and articulate something, rather than glazing over a summary that is meant to serve your view.

I am also sorry for all the other people, especially marketers and creators, who are going through the same thing because AI was targeted to replace them first. I guess the more unknown or tactical the job is, the easier it is to say I did get AI to make it for me for zero dollars, and realistically, in today's economy, it is not zero dollars, where Claude charges US$ 100 and ChatGPT at $40.

There is Gemini and perplexity, and then one is not enough because they are all good at different things. It is actually costing a lot, even without hitting the token limits. Also, there is the time on top of this to learn how to use and integrate all of them.

First Case
This is a part of an inquiry that came to me. I should also say it is one of the better ones; they should be prompting really well.

“Our primary technical requirement is the development of a proprietary lamellar gel-network or lamellar liquid-crystal emulsion system designed for ultra-premium skincare performance and sensory distinction.

Before proceeding to the NDA, we would appreciate insight into the following:

Named lamellar, liquid-crystal, or gel-network systems your R&D team has previously developed or worked with (by INCI, supplier technology, or platform reference), along with one or two anonymised case examples”

Well, basically, I can give them an INCI listing. Still, first, I don’t know what it will mean without their texture, or whether they can take it back, dissect it, and use it to reformulate on different platforms, but this is the reality. I think, especially in OEM and consulting, we will see this kind of request more and more, and basically whoever actually helped DIY communities to formulate will probably get mentioned more. If the client is AI-obsessed, we will end up with a single formula.

Second case
A client came to me for a formula review that they had got done on Fiverr.

The formula was poorly prompted in ChatGPT, and it also included ingredients that didn’t exist, either as registered INCI names or as commercially available. Even the one that exists was used in percentages; either it won’t form a viable product or it will be over the limits of what can be used. They have created hopes of building this brand and invested their time, only to find that it realistically cannot happen. This is a real case that shows how AI can make people sell false information and hope.

I might sound like I am against AI, but I am not.

AI is not going anywhere; the internet didn’t disappear because it had both positives and negatives, but our dependence on it, trust in it, and values have changed around it. I think we are in the scarcity phase of it because AI companies need to make money, and the more they scare all of us that AI will take our jobs and run countries.

We will be that fat person in Wall-E, which is a good tactic to subscribe to them, create their training data and basically be addicted to them, but I think this is another topic. There is a really good read [1] and a watch [2] on this topic.

Good things are happening with AI as well, like Potion AI or CM Studio+. Now, if you like a product or haven’t made it before, you can make a better guess about the chassis and see which ingredients might be key to stability.

Still, I think there is a long way to go for these systems to be able to provide a 100% formula, but overall, this is a good thing as well, because we are living in a world where IP has already become a real grey area.

Also, I think Google’s Notebook LM is really, if you are trying to learn something new to yourself or learn about a subject or an ingredient, to basically load lots of data to be summarised and cited in the short term, it is amazing, but that does not mean that you shouldn’t read that for yourself.

For example, I created an app for the New Zealand Society of Cosmetic Chemists innovation zone because it was hard to calculate the results manually, and we had no visibility into the products in there until the day before the event started. Believe me, without AI I wouldn’t have had any clue. However, if I didn’t have my lovely software engineer sister-in-law, it would still show the same picture for all the products because that was the most technical part to solve.

In short, we need to create better value propositions for the things we are doing and be more upfront when something seems like AI, and if the client still insists, maybe you really don’t align on the values and it wasn’t really your time.

[1] Examining human reliance on artificial intelligence in decision making - Nature.com

[2] There Is No AI, Really (It’s Just People), with Jaron Lanier - YouTube.com/startalk

Looking for something else?