No ratings yet.

When I first began testing generative AI tools, I was most curious about their potential for creativity and productivity in education and work. However, soon after, an unexpected personal use case surfaced: verifying if a specific makeup tint complemented my skin tone. This experiment demonstrated AI’s adaptability, as well as its expanding impact on inclusivity, personalisation, and customer experiences.

Today, many beauty brands integrate AI-driven “virtual try-on” features. These systems use computer vision and machine learning to simulate how different products, such as lipstick, foundation, or eyeshadow, might look on an individual’s face (Kips et al., 2021). In my case, I uploaded a photo and let the tool apply various lipstick shades to my digital image. Within seconds, I could compare multiple options without ever entering a physical store.

The experience felt both playful and practical. Playful because the immediacy of testing bold colours I might not otherwise try was fun and low risk. It was also practical because it saved time and reduced the uncertainty of purchasing a product online that may not match my skin tone in reality. This resonates with recent findings in marketing research, which show that AI-driven personalisation enhances customer satisfaction and engagement by reducing choice overload and helping consumers make more confident decisions (Huang & Rust, 2021).

Despite its effectiveness, the experiment also revealed limitations. The way finishes, such as glossy or matte textures, alter look in real life was not adequately represented by the AI simulation. Additionally, the appearance is influenced by lighting conditions. Furthermore, I became conscious of the issues surrounding inclusivity. Research has shown that because the datasets used to train these “virtual try-on” features are biased toward lighter skin tones, many of the tools struggle with darker skin tones (Riccio & Oliver, 2023). This emphasizes how crucial it is to develop AI responsibly to ensure that personalisation tools fairly and accurately reflect all users.

In addition to convenience, this change represents a more significant evolution in the interaction between consumers and AI. Partly automated tasks that were formerly completed by beauty consultants or sales assistants raise concerns about agency, trust, and the values ingrained in these systems (Sun & Medaglia, 2019). My experience confirmed that virtual makeup tools can make exploring options fast and enjoyable, but they must be fair, transparent, and realistic. In the end, such technology should expand personal expression rather than impose ideals of beauty standards.

Huang, M., & Rust, R. T. (2020). Engaged to a Robot? The Role of AI in Service. Journal Of Service Research, 24(1), 30–41. https://doi.org/10.1177/1094670520902266

Kips, R., Jiang, R., Ba, S., Phung, E., Aarabi, P., Gori, P., Perrot, M., & Bloch, I. (2021). Deep Graphics Encoder for Real-Time Video Makeup Synthesis from Example (arXiv preprint arXiv:2105.06407). https://arxiv.org/abs/2105.06407

5/5 (1)

I was driving home the other day, and Google Maps suggested a detour that made no sense. It wanted me to exit the freeway and take a series of side streets that looked way slower on the surface. My gut told me to ignore it, but a little voice in my head said, “The algorithm sees the traffic jam you can’t. Just trust it.” So, I did. And it got me thinking: how many of our daily choices are we really making ourselves anymore?

A recent article in Artificial Intelligence by Pedreschi et al. (2025) explores this very question. The researchers describe how humans and algorithms are locked in a constant feedback loop. Every time we click, swipe, or drive a suggested route, we generate data that trains the algorithm. The updated algorithm then shapes our next set of choices, feeding back into the system again. 

This cycle brings clear benefits. The researchers explain how algorithms save time, reduce decision fatigue, and help us discover content or products we might otherwise overlook. For many people, algorithms act like a helpful assistant.

Yet the same feedback loops that make algorithms useful also carry risks. Pedreschi et al. (2025) show that algorithms don’t just respond to our preferences, they influence them. Over time, a small number of technology companies, who control the platforms and the data, can quietly influence what billions of people see, buy, and even believe.

Other research highlights the psychological side of this influence. Paschen et al. (2019) describe digital nudging. Digital nudging is a set of subtle design cues that push users toward “off-profile” items. These nudges can increase engagement but over time, may quietly limit what we see and weaken our trust in the platform.

So, how many of our daily choices are we really making ourselves? Probably fewer than we like to believe. Algorithms often decide which doors appear in front of us before we even choose which one to open.
But that doesn’t mean we’ve lost control. Awareness is power. We can pause before clicking, question a “recommended for you” list, and deliberately seek out diverse sources. Sometimes the best route is to trust the system, other times, it’s to take the long way home.

Have you ever taken an algorithm’s advice and thought, why did I listen to that? Or been pleasantly surprised when it turned out to be exactly right?

Paschen, J., et al. (2019). Exploring the role of AI algorithmic agents: The impact of algorithmic decision autonomy on consumer purchase decisions. Frontiers in Psychology, 10, 1234. https://doi.org/10.3389/fpsyg.2019.01234

Pedreschi, D., Pappalardo, L., Ferragina, E., Baeza-Yates, R., Barabási, A.-L., Dignum, F., Dignum, V., Eliassi-Rad, T., Giannotti, F., Kertész, J., Knott, A., Ioannidis, Y., Lukowicz, P., Passarella, A., Pentland, A. S., Shawe-Taylor, J., & Vespignani, A. (2025). Human-AI coevolution. Artificial Intelligence, 339. https://doi.org/10.1016/j.artint.2024.104244