What happens when beauty meets AI? Key learnings from Roquette Beauté

Roquette Beauté recently hosted its annual Experts Days with prestigious clients of the beauty industry. The focus was on AI and beauty. I was kindly invited to host discussions with amazing speakers from Sociology, Law, and Fragrance, as we tried to dissect the implications, challenges, and opportunities arising from AI within Beauty, charting a course for the future of our industry. From regulatory implications to impact on human well-being, and challenges or opportunities created in our sector, R&D and product development. Are we to fear or to embrace AI?

- As a matter of example of AI capabilities, this article has been drafted using AI -

State of play

The beauty industry is rapidly adopting AI technology to enhance various aspects of its services. Brands like Sephora, L’Oréal, and Perfect Corp are using AI for advanced skin diagnostics, virtual try-ons, and generating complete hair and makeup looks. 

On the manufacturing side, companies like Schrödinger use AI-driven Physics-Based Simulations to create digital models of formulas, predicting ingredient interactions and product properties. On the packaging side AI is supporting ideation and maintenance and production forecasts for a smooth supply chain.

While AI in beauty raises concerns about unrealistic beauty standards, data privacy, and IP infringement, it also offers significant potential for innovation and creativity. If implemented responsibly, AI could bring numerous benefits to the industry and humanity.

AI and LAW

Matthieu Bourgeois, a lawyer and partner at Klein-Wenner, outlines the legal challenges and opportunities of AI in the beauty industry, focusing on intellectual property (IP) and data privacy. AI's integration raises issues related to patent, copyright, and GDPR regulations. Accurate data is crucial for AI outputs, which face risks like bias and discrimination.

For Intellectual Property (IP), AI-assisted works may be protected if they show human originality. France has no specific case law, but AI creations could be assessed individually. Protecting non-traditional IP, such as fragrances, highlights broader challenges. For example, “ looking at the protection of fragrances, with the Court of Cassation ruling that a perfume's fragrance, being a mere implementation of know-how, does not qualify as a protectable intellectual work. This principle, established in several rulings, underscores the broader challenge of defining and protecting non-traditional intellectual creations in the evolving landscape of AI and IP.” explained Matthieu.

In conclusion, AI's potential benefits must be balanced with robust regulation to mitigate risks, drawing lessons from past regulatory successes like GDPR (GDPR example is being followed in several countries outside Europe to support individual privacy rights).

AI and Mental Health

With Professor Rosalind Gill, a researcher in social studies at Goldsmiths, University of London, we understand how beauty and AI is impacting young generations following her recent publication “Perfect” and a separate report, on young women’s experiences on social media.

The young generation, and young women, in particular, feel immense pressure to appear perfect and experience a love-hate relationship with beauty content on their phones. While beauty content can be inspiring, it often makes them feel inadequate and anxious. Gen Z women also feel constantly scrutinised in public and online, developing a hyper-detailed focus on appearance that differs from older generations.

As the beauty industry is starting to use AI, it needs to protect young women from falling into this permanent surveillance fear. The industry needs to offer freedom and more support to women's mental wellbeing. 

As an industry, “ Beauty needs to ensure that ‘AI women’ does not create even greater pressures at a time when young women already suffer with the consequences of impossible beauty standards. She also emphasises the importance of how companies avoid algorithmic and other biases to reflect and create for a world of diverse body shapes, sizes, ethnicities, skin and hair types” expresses Rosalind. 

Down the line, it is about how to use AI to create ethical and sustainable futures.

AI and Formulating Beauty

 Thanks to the testimony of Nicolas Olczyk, creative consultant in the world of perfume, we discovered how AI supports the artistic creation of perfume and becomes a major ally.

The perfume industry has been leveraging AI for years reading the plethora of ingredients used in fragrances for molecule interactions learned from past formulas, historical success data and industry trends for instance.

Nicolas Olczyk also tested open-source AIs like ChatGPT and Google's Bard/Gemini to create perfume formulas, with promising results showcased at Roquette Expert Days. AI aids in marketing, concept development, and speeds up creation while adhering to industry regulations on ingredients and allergies. However, human intervention remains crucial for managing these aspects.

Partnering with AI could be advantageous if AI-generated creations are clearly labelled, maintaining ethics and authenticity. AI does not replace jobs but requires human creativity to progress. The partnership with AI could enhance innovation and diversification in perfume creation, as long as human oversight ensures originality and authenticity.

"Is it better to create a perfume that works or one that pleases? There is a certain naivety in the vision of AI about the perfume market " states Nicolas, "but it offers unprecedented possibilities for innovating and diversifying creations". 

Ultimately, a partnership with AI could be beneficial, provided that human intervention is always maintained to guarantee the originality and authenticity of the creations.

AI and BEAUTY

AI is revolutionising the beauty industry. Generative AI enhances creativity by analysing consumer insights to suggest products that meet legal and sustainability standards. It optimises supply chains with real-time demand forecasting and improves marketing through engaging visuals and copywriting. In retail, AI supports customer service with real-time insights and interactive experiences like smart mirrors. Overall, AI enhances efficiency, sustainability, and customer engagement across the industry.

If done well, and safe-guarded against bias and errors, and protecting IP, AI can support the future of our beautiful industry.

AI & BEAUTY @ ROQUETTE BEAUTE

At Roquette, Digital teams developed RoqGPT, a secure alternative to ChatGPT, leveraging Azure Open AI resources. They also created multiple solutions to leverage Generative AI, like automated Customer Feedback categorisation or large R&D corpus search and interpretation.

ABOUT ROQUETTE BEAUTE

Roquette Beauté offers a full range of multifunctional plant-based ingredients, and high-performing, plant-based concepts for personal care, providing amazing sensorial experience and claim-substantiated benefits, for skincare, haircare, oral care, colour cosmetics and fragrance formulations.

"Sustainable beauty is all about skin health and nutrition. " Roquette Beauté

FULL CONTENT FROM THE DAY.

Find here below a full summary of the conversations about AI led at Roquette Beauté Expert Days.

State of play

The beauty industry has embraced AI technology with a flow of innovations. Beauty brands are using AI as an assistant to beauty advisors for enhancing skin diagnosis to deliver meticulous beauty prescriptions or direct with consumers using imagery provided by the latter. Sephora with Smart Skin, L’Oréal with Modiface or even Perfect Corp using Augmented reality for virtual try-ons and AI to generate full hair and makeup looks. 

On the industry side, innovators like Schrödinger are using Physics-Based Simulations with AI to generate digital twins (sort of) of formulas knowing how each ingredient will interact with each other and behave in a formula, offering an idea of the sensory properties properties of the finished product (ingredient or emulsion stability, solubility or viscosity, etc). InFlowsAI is also developing an innovation to digitally model formulas in order to simulate the replacement of controversial ingredients with better alternatives and predict the physical properties of each switch.

These are only a few innovations that are currently available or in development, and a lot more will be birthed in the coming months or years.

In essence, AI may be feared today for the risk of dispersing unrealistic beauty standards, which the industry may have already been accused of and is trying to correct today. AI may also be associated with misleading depictions of real-life events and the loss of critical thinking or uniformisation if it were to be used everywhere. while IP Infringement and data privacy remain a general concern. But AI is not all negative and it brings a lot of potential to the creative or innovative process if done correctly. If AI is well designed, we can imagine the tremendous possibilities it will bring to humanity.

AI in Law

With the support of Matthieu Bourgeois, Lawyer and partner at Klein-Wenner - a Paris law firm - we explained the limits and opportunities of AI with intellectual property and data privacy and how this may impact creation and innovation in the beauty industry.

In effect, the beauty sector's adoption of AI raises complex legal issues, from intellectual property to data privacy, leading back to patent and copyright laws as well as GDPR regulations. 

re-sources: Matthieu, has AI got a regulatory framework today?

Matthieu Bourgeois: Let me start with an example. In 1973, France was the first European country to impose seat belts on the front of cars. People protested in the name of loss of individual freedom. However, 50 years later, its effectiveness and usefulness has proven itself well and would no longer be questioned. Indeed, as with AI, it is sometimes necessary to supervise and coerce to protect the community, society. We have to see this as an opportunity to be in the only continent today that has the will to see beyond the sole purpose of innovation. Today we see that GDPR has proven itself and is imitated elsewhere in data and privacy rights in the realm of digital technology such as AI.

Legally, AI systems are defined by their ability to operate autonomously, adapt post-deployment, and generate results that influence environments. This definition includes general-purpose models, a broad category within AI. 

Conceptually, AI can be divided into strong AI, which implies true machine self-awareness, and weak AI, which follows user instructions, or weak AI, which follows simple application of user instructions. Technically, AI is classified into symbolic AI which is based on an inference engine using predefined rules, or Connectionist AI which uses inductive logic.

re-sources: Has AI got a regulatory framework today?

Matthieu Bourgeois: First of all, we need to look at the importance of data accuracy. The quality of AI outputs is heavily dependent on the data it is trained on, mirroring any inherent biases. This probabilistic nature of AI results means they reflect probabilities rather than absolute truths, underscoring the critical role of data quality.

AI poses several risks, including bias and discrimination, as seen in Amazon's deactivated AI recruitment tool, which replicated human prejudices. Explainability is another issue, particularly with connectionist AI that can generalise without providing intelligible rules. The health impacts of AI usage are also concerning, with links to brain shrinkage, anxiety, and depression. In workplaces, increased AI reliance may reduce social interactions, affecting employee well-being. Additionally, AI's growing energy consumption, especially from data centres, contributes significantly to environmental degradation.

AI is categorised by risk levels, starting with prohibited AI, which includes unacceptable risks like subliminal AI techniques, social scoring, and unauthorised biometric identification. High-risk AI encompasses systems with safety components and applications in critical fields such as education, employment, infrastructure, and justice. General-purpose AI, which carries moderate risk, includes technologies like chatbots and generative AI, and has additional compliance obligations. Moderate-risk AI interacts directly with people or generates content, including deepfakes and conversational agents. Low-risk AI, the majority of current applications, includes spam filters, video games, and fraud detection systems.

re-sources: How should AI be managed / organised to mitigate risks?

In managing AI, different actors have specific roles. Importers market AI systems in the EU, while distributors are part of the supply chain without being the supplier or importer. Deployers or users apply AI for professional purposes, and providers develop and market AI systems. High-risk AI providers face extensive compliance requirements, including risk management, data quality assurance, technical documentation, and post-market surveillance. Users must follow usage instructions, ensure human oversight, and maintain relevant logs. Importers and distributors must ensure conformity and alert authorities if any risks are identified.

Companies should start by mapping their AI usage, identifying and qualifying each AI system as high, moderate, or low risk. Organising compliance involves assigning a project manager and training employees. Contractual considerations should define responsibilities and obligations clearly.

re-sources: And what about Intellectual property or Industrial Property? Can an innovation created under AI be protected?

Matthieu Bourgeois: The protection of AI-generated works raises important questions. It is crucial to distinguish between autonomously generated works and those created with AI assistance. AI-assisted works may be protected if they demonstrate human intervention. Potential protections include personalist copyright, neighbouring rights, or sui generis rights for AI-generated creations.

Under what conditions can creative human intervention be considered sufficient for the work resulting from AI to be original? There is no case law in France on this subject, but it is possible to imagine that AI-assisted creations could be assessed on a case-by-case basis. 

In any case, if an AI-assisted work takes up the original characteristics of a pre-existing work, it will be considered a derivative work and subject to the authorization of the author of the original work. 

Looking at the protection of fragrances by IP illustrates the nuances in IP law, with the Court of Cassation ruling that a perfume's fragrance, being a mere implementation of know-how, does not qualify as a protectable intellectual work. This principle, established in several rulings, underscores the broader challenge of defining and protecting non-traditional intellectual creations in the evolving landscape of AI and IP.

In conclusion, while AI holds immense potential, it also brings significant risks that require robust regulation and management. The historical lessons of seat belt mandates and the GDPR highlight the necessity of proactive regulation to safeguard society from the unintended consequences of technological advancements.

AI, Beauty and mental health

With Professor Rosalind Gill, a researcher in social studies at Goldsmiths, University of London, we understand how beauty and AI are impacting young generations following her recent publication “Perfect” and a separate report, on young women’s experiences on social media.

Current thinking about appearance pressures suggests an intensification of these concerns over time. Despite decades of feminist movements, women continue to face significant pressures to conform to beauty standards. In her 1990 book, Naomi Wolf highlighted this issue, stating that women endure what she described as "a dark vein of self-hatred, physical obsessions, terror of ageing, and dread of lost control." She referenced research indicating that American women often identify losing weight as their primary life goal. Wolf argued that these pressures contribute to widespread feelings of shame and body dysmorphia among women on a national scale.

In society we see an intensification of appearance pressures. Rates of distress and mental ill-health relating to body image are increasing dramatically – for boys and young men this includes taking steroids or skipping meals, for girls and young women eating disorders, anxiety and depression are at their highest levels ever.

Young women feel under strong pressure to appear Perfect.

There is also a love-hate relationship with beauty content. Young girls and young women feel that they constantly see content related to appearance on their phone. They explained that it can be inspiring, motivating, and amazing to see what a difference something makes (from makeup to cosmetic dentistry), but it can also make them feel they will never be good enough and cause them to feel low, anxious and ‘drained’.

Gen Z women are also feeling watched across all aspects of lives - public transport, bars, clubs, etc. They look at themselves and others forensically, with a magnifying sensibility – both on their phones and off. And compared with older generations they see in much more detail and are developing historically new visual literacies of the face, partly because of beauty content on their phones: they zoom, they focus, they itemise. They practise looking through a ‘pedagogy of defect’- a focus on what is wrong- and they know they are different in this from older generations.

As the beauty industry is starting to use AI, it needs to protect young women from falling into this permanent surveillance fear. The industry needs to offer freedom and more support to women's mental wellbeing. 

With all this information, Rosalind recommended we, as an industry, ensure that ‘AI women’ does not create even greater pressures at a time when young women already suffer the consequences of impossible beauty standards. She also emphasises the importance of how companies avoid algorithmic and other biases to reflect and create for a world of diverse body shapes, sizes, ethnicities, skin and hair types etc with the likes of campaigns led by Dove for instance. Down the line, it is about how to use AI to create ethical and sustainable futures.

AI and the beauty formulator

Thanks to the testimony of Nicolas Olczyk, creative consultant in the world of perfume, we discovered how AI supports the artistic creation of perfume and becomes a major ally.

The perfume industry is very advanced when it comes to artificial intelligence and has been using it for years. With solutions which include AI system that can learn about formulas, raw materials, historical success data and industry trends. A system that uses advanced machine learning algorithms to sift through hundreds of thousands of formulas and thousands of raw materials, helping identify patterns and novel combinations.

The industry is even creating fragrances assisted by AI and neuroscience. Charlotte Tilbury and Moschino are using this approach to link scents to mood. Firmenich and AMOI have also turned to neuroscience. The idea of well-being and aromachology, once used by the Body Shop and Yves Rocher, is now supported by concrete data. Today, the combination of AI and neuroscience is used to validate claims and add a scientific aspect. The selection of raw materials is based on their scent and emotional impact.  At MANE, questions were raised about the relationship between structures and odours. For quality control, the interest lies in the speed compared to the mood. However, there are limits to Symrise's Philyra technology.

Nicolas Olczyk tested several open source AIs available on the market, to share his experience and the results obtained. Nicolas tested formulas with ChatGPT and Google's Bard/Gemini. The AI came up with some daring formulations, sometimes with quite high scores. With the help of Cosmo Fragrances, Nicolas formulated these fragrances, which we were able to test at Roquette Expert Days. The results were astonishing. In his approach, Nicolas also called on AI for marketing, concepts and ideas.

The perfume industry is regulated, as is the beauty and cosmetics industry, particularly when it comes to ingredient percentages, regulations and allergies. Does AI improve development time while taking these restrictions into account? Does it help the creative process? The profession of perfumer is not without its dangers, and human intervention is essential to manage regulations and speed up development and creation time.

In France, perfume houses (such as Alexandrine) collect information on ingredients and have databases of formulas. The interactions of raw materials with the mind, allergies and neuroscience are also taken into account. AI, while not new, offers a real-time picture of companies and data. Data warehousing supports formulation and creation, while the effect of raw materials is studied through neuroscience.

Data science goes beyond simple artificial intelligence. It exploits existing data to generate new ideas and creations. Today, people accept that AI can write articles and books, and it is just as capable of creating a series of recipes with perfumery raw materials. However, whether these creations smell good is another matter. There are also stability and technical considerations that AI can learn through data learning. It would be wrong to say that AI cannot create perfume.

Why wouldn't AI use the dictionary of existing formulas to generate something new? AI could, for example, categorise smells and create a perfume in the style of Jean-Claude Ellena, like a painting in the spirit of Klimt, or even inspired by a concerto. However, this raises the question of the risk of standardising creation. The exploitation of existing data could lead to creations that lack diversity and originality, but it also offers unprecedented possibilities in terms of speed and precision in the creative process.

So should we partner with AI? 

Nicolas explains that those who know how to use this technology and how to decipher and verify information will have a clear advantage. AI is inevitable, but it should always be stated that a creation is "generated by AI" to avoid any deception. Comparing a perfume created by a human perfumer and by an AI on the same footing can raise questions of ethics and authenticity.

Fragrances created by AI could be very good. However, we need to continue to feed the tool to improve it through human intervention. AI does not eliminate jobs, but requires human intervention with "out of the box" thinking to progress.

"Is it better to create a perfume that works or one that pleases? There is a certain naivety in the vision of AI," says Nicolas, "but it offers unprecedented possibilities for innovating and diversifying creations". 

Ultimately, a partnership with AI could be beneficial, provided that human intervention is always maintained to guarantee the originality and authenticity of the creations.

AI IN THE BEAUTY INDUSTRY

At Roquette, Digital teams developed RoqGPT, a secure alternative to ChatGPT, leveraging Azure Open AI resources. They also created multiple solutions to leverage Generative AI, like automated Customer Feedback categorization or large R&D corpus search and interpretation.

Overall, in our industry, generative AI can be used to fast-track the creative process of product development by synthesising consumer data and suggesting adapted products, with added legal or sustainability criteria. On the other hand, the supply chain can create real-time demand forecasting to optimise manufacturing.

It can also accelerate or support the marketing process through imaging and copywriting for increased engagement on social networks. It will also help the product development process by capturing global data on consumer behaviour. It can also generate hyper-personalised loyalty programs thanks to available information and communication channels.

In store, AI will help store associates understand real-time customer needs, and offer support to beauty advice through smart mirrors for instance, while serving as a fun experience for consumers and deepening the brand-consumer relationship.

On the supply side, AI can bring greater transparency and insights into emissions and ingredient origins, which information is already available in the industry with the likes of Clarins for instance and can now be utilised by AI. On the supply chain or in retail, AI can analyse measures in real time for product logistics to ensure prompt stocking of items.