Artificial Intelligence (AI) has been praised and hated in various measures by all sorts of industries; but its continuing involvement in our society is in some degree inevitable. Where does AI find its strengths and weaknesses when it comes to cosmetic formulation?
Since this is such a big but important topic, it has been split into two parts. The first instalment focused on the pros and cons of AI when it came to ingredient research and inputs.
This instalment will take a closer look at the cons of AI when it comes to cosmetic formulation, the misguided information about cosmetic ingredient safety, as well as the best opportunities for AI to help cosmetic chemists maximise efficiency and efficacy of lab samples.
AI con: it cannot be relied upon for cosmetic ingredient safety information.
AI currently works by finding the most common responses to any given topic using the search terms input by the user. This means it is impacted greatly by the search terms used, and the most common responses – not necessarily the most correct.
It is unfortunate that for far too long the cosmetics industry has been influenced by the statements made when marketing the ingredients of a cosmetic product – both well intentioned and purposely misguided. This means that researching the safety of cosmetic ingredients using AI is currently not very reliable at all.
For example, for the search question*: ‘how safe are parabens in cosmetics?’ the results incorrectly listed this information:
- Parabens are potentially unsafe in cosmetic products because of absorption and their impact on estrogen production and endocrine disruption.
- Long-term exposure to multiple paraben sources from cosmetics can accumulate in the body to potentially toxic levels.
- Skin irritation and allergic reactions are likely.
These above statements are not scientifically correct based on the input and exposure (even from multiple sources) of parabens from cosmetic products; but it does show how a large body of misinformation can bias AI’s search results.
The result continued with at least some correct information, when it quoted the regulatory sources (which has been based on intense scientific scrutiny and research):
- FDA and EU regulators consider parabens safe to use in cosmetic products within the permitted levels.
But then it went on with still more incorrect information, again obviously obtained from the majority of current internet information, although inaccurate:
- Suitable alternatives to using parabens include preservatives such as vitamin E, grapefruit seed extract or rosemary extract. (Author note: vitamin E and rosemary extract are antioxidants, not preservatives; and the efficacy of grapefruit seed extract is not suitable to replace a paraben as a preservative).
The results did not at any point identify just how many chemical substances are covered by the term ‘parabens’, nor which are currently permitted in cosmetic formulas and which are prohibited for safety reasons.
This example shows just how unreliable AI can be when trying to use it to obtain meaningful safety information about cosmetic ingredients, and the inaccuracy of AI for some ingredient research purposes in general.
Other cons of AI…
Here are some other cons of AI for cosmetic formulators based on its current abilities, summarised briefly:
- AI cannot predict the likely outcome of combinations of multiple cosmetic ingredients the way that a cosmetic formulator can learn from experience.
- AI cannot be used to tell the personal preferences of a specific target market, particularly when it comes to the true sensory experience of a combination of cosmetic ingredients as they appear in a multi-ingredient cosmetic formula.
- AI can be manipulated or at the very least impacted by common statements to spread misinformation about cosmetic ingredients (the above safety example shows this all too well!)
Opportunities where AI for formulators could really be put to good use.
While AI for cosmetic formulators currently has significant limitations, suitably designed AI systems could be used to increase efficiency of developments and predictions in the following areas:
- Colour matching by cosmetic formulators can take a lot of time and relies on the experience of the chemist. A suitably designed AI system could hasten this process significantly, however the specific ratios of colours used always depends on the base formula and grade of colourant used – so any AI system designed to aid with colour matching will need to be adaptable to individual base formulas and selected colourants.
- There are various SPF calculators out there already, but all have limitations – usually supplier specific. SPF calculations also rely heavily on the type of base formula created, for example if it is oil based, w/o, o/w or gel based. It can also be impacted by the use of known SPF boosters. If programmed to address these current limitations, specially created AI could greatly help improve the calculations of SPF predictions and inputs to save time. Samples would still need to be created and tested properly, however, to ensure safe consumer use and regulatory compliance.
- Data from early stability results could be used in specially created AI programs to extrapolate results with greater reliability for earlier market entry of new product concepts with appropriate shelf-life justification.
Current ways of manually conducting the above activities could be dramatically hastened with the use of AI, but that AI would have to take into consideration the individual nuances of specific ingredient combinations and formulation bases before its results could be applied in a majority (if not all) cosmetic formulation situations.
Another area for opportunity for AI would be if it were able to create multiple variations of samples based on minor adjustments of ingredients to speed up lab-based sample development. Imagine if a machine could make 5 versions with minor tweaks of viscosity and sensory modifiers at the same time – saving countless hours that would otherwise be needed at the bench to provide multiple variants of a formula for consumer evaluation.
Imagine if AI could…
Now this is one thing I get really exited about – how it can be achieved with true accuracy and reliability I am not sure, but imagine this if you will: what if there was an ingredient database created – and constantly updated by reliable sources – that provided regulatory limits, compatibilities to be considered, accurate safety information and formulation inputs?
While a qualified cosmetic chemist would still need to apply the information to suit specific ingredient and formulation considerations, such a database – where proven beyond doubt to be reliable – could save a lot of time in compiling and conducting essential preliminary safety, regulatory and compatibility checks. Further time would be saved by knowing it is 100% reliable and up to date. Currently, even the best attempts at capturing this information is subject to further checks by those with knowledge in these areas… but what if such a system existed that we could just type in a CAS number and get results we are certain about?
So yes, there has been plenty of talk about how AI is limited at this point in time; but I hope for a future time where we will see more of the opportunities harnessed and brought to market to assist cosmetic chemists in a useful and reliable way.
Happy formulating!
* Note: if you run similar AI checks of questions your results may vary depending on your search engine and location.
Feeling inspired?
Then why not visit one of the in-cosmetics events around the world?