show index hide index
In recent years, artificial intelligence (AI) has made significant progress which has made it possible to set up ever more efficient recommendation systems. These tools, supposed to facilitate our daily lives and help us in our choices, are they infallible? This article looks at the major technological advances in AI and explores their limits, particularly in terms of bias and ethics.
The feats of artificial intelligence
Artificial intelligence has steadily advanced in recent years through innovative techniques such as machine learning and deep neural networks. Examples such as ChatGPT or Midjourney perfectly illustrate this evolution, allowing a democratization of technology among the general public. The fields of application are vast: health, education, commerce, marketing, etc.
Personalized recommender systems
AI-based recommender systems leverage available user data to provide choices tailored to their preferences and needs. Several online platforms, such as Goodreads, Amazon or Tertulia, offer book suggestions based on users' tastes and previous readings. Similarly, social networks and streaming platforms use algorithms to recommend content that may be of interest to their members.
The limits of AI recommender systems
Despite these promising advances, AI-based recommender systems have certain limitations and can sometimes generate errors or biases. Several experts agree that it is essential to be aware of these pitfalls to better anticipate and correct them.
Prediction errors
Like any machine learning algorithm, recommender systems are susceptible to errors. While some results may turn out to be relevant, others may seem inappropriate or even completely meaningless. This may be due to insufficient, erroneous or biased data, or to imperfections in the learning models used.
- Insufficient data: if the system does not have enough information about a user, it will struggle to offer relevant recommendations.
- Erroneous data: an error in the collection or processing of data may lead to inappropriate recommendations.
- Biased data: if the training data reflects existing biases, the recommendations will also be biased.
Algorithmic and ethical biases
AI recommender systems may also be subject to algorithmic and ethical biases. For example, a recent study showed that users tend to follow the advice of a chatbot like ChatGPT even though it is supposed to remain neutral. The researchers question the impact of these tools on the formation of public opinion and decision-making, and warn against the risks of excessive influence or manipulation.
Towards a regulation of artificial intelligence
In order to prevent potential abuses related to AI recommendation systems and to guarantee their ethics, several bodies are looking into the issue of their regulation. The European Parliament is working in particular on legislation aimed at regulating the use of artificial intelligence. Once adopted, this law will better protect citizens against discrimination and invasions of privacy.
To read Personal Computer : Découvrez « Claude Cowork » de Perplexity, désormais ouvert à tous
AI recommender systems represent a major advance in various fields, but it is crucial to take into account their limits and their ethical challenges. By working on their continuous improvement and by establishing an appropriate legal framework, it should be possible to take full advantage of their potential while minimizing the risks for society.
Sources
- https://www.jeuxvideo.com/news/1751021/ia-y-a-t-il-des-limites-a-l-avancee-technologique.htm
- https://actualnewsmagazine.com/pourquoi-les-algorithmes-sont-ils-toujours-aussi-mauvais-pour-recommander-des-livres/
- https://www.ladn.eu/tech-a-suivre/chatgpt-peut-vous-faire-changer-davis-et-vous-ny-verrez-que-du-feu/
- https://www.euractiv.fr/section/economie/news/reglementation-de-lia-lue-franchit-une-nouvelle-etape-apres-un-vote-cle-au-parlement-europeen/
