AI in Language, Economics, and Law: What’s at Stake for the Future?

Artificial intelligence (AI) is profoundly transforming multiple sectors—from the automatic processing of human language to economics and law. Recent research explores these transformations in diverse areas such as machine translation, coreference resolution in language processing, and algorithmic pricing practices. These studies reveal both AI’s immense potential and the social, legal, and ethical challenges it poses.

By: Ashwin Ittoo

Artificial intelligence (AI) is profoundly transforming multiple sectors—from the automatic processing of human language to economics and law. Recent research explores these transformations in diverse areas such as machine translation, coreference resolution in language processing, and algorithmic pricing practices. These studies reveal both AI’s immense potential and the social, legal, and ethical challenges it poses.

Improving Machine Translation for Underrepresented Languages: Building a Digital Bridge

Machine translation (MT) has made remarkable progress in recent years. However, many so-called “low-resource” languages—such as Japanese, Malay, or Vietnamese—or even local dialects like Occitan or Walloon (in French-speaking Belgium) still struggle to achieve performance levels comparable to widely spoken languages like English or French. This remains a major obstacle for millions of speakers worldwide.

A recent article addresses this issue by proposing two innovative strategies. The first involves building bilingual corpora from comparable data sources, such as Wikipedia articles. This approach leverages existing resources to offset the lack of direct translation data. The second strategy relies on using “pivot languages,” such as English, to facilitate translation between two low-resource languages.

The results are striking: these techniques can improve BLEU scores by 2 to 7 points. In other words, they not only help break down language barriers but also offer speakers of these languages fairer access to information and digital technologies.

Note: BLEU (Bilingual Evaluation Understudy) is a metric for evaluating machine translation quality by comparing a machine-generated translation to one or more human reference translations. BLEU scores are numerical values often expressed as a percentage.

Embeddings: The Key to Better Text Understanding by AI

Natural language processing (NLP) relies on a key technology: word embeddings. These numerical representations allow machines to understand and analyze the semantic relationships between words in a text. They are particularly useful for tasks like coreference resolution—identifying which entities or events refer to the same thing in a text.

For example: Marie bought a house. It is beautiful. A good coreference resolution system should determine that It refers to house and not Marie, based on sentence context. A recent study compares different types of embeddings: static (such as Word2Vec), contextual (such as ELMo and BERT), and character-level embeddings.

The conclusion? These embeddings are learned through neural networks and can vary in size. Results show that larger embeddings can improve model accuracy, but the gains are often marginal relative to the resource cost. In other words, it’s often better to favor simpler, faster models that offer comparable results with greater efficiency. This balance between performance and computational cost is crucial for developing large-scale NLP systems. This study forms the basis of Ashwin Ittoo’s current research, which focuses on compressing large language models (LLMs, such as GPT) to reduce their carbon footprint.

When Algorithms Agree: Price Collusion and Price Discrimination

AI is not limited to linguistics. It also plays a major role in corporate pricing strategies, with potentially significant impacts on consumers. In one study, researchers examine two sensitive issues: price discrimination and tacit collusion (see also another article by the same authors).

Price discrimination involves adjusting prices based on consumer profiles. For example, a loyal platform customer might be shown higher prices than a new user. AI algorithms analyze purchasing behavior in real time and automatically adjust prices to maximize corporate profits. While legal, this practice raises fairness concerns—especially when it exploits consumer vulnerabilities. It can, of course, be harmful to the buyer.

Tacit collusion is even more concerning. Unlike explicit collusion, where companies directly agree to fix prices, tacit collusion occurs when algorithms monitor competitors’ prices and automatically adjust their own accordingly. For example, if an algorithm detects that all competitors raise prices after each drop, it may “understand” that it’s better to maintain high prices to avoid a price war—leading to a direct and costly impact on consumers.

While these practices are not yet widespread, researchers call for pragmatic regulation that takes into account the current capabilities of these technologies rather than focusing solely on futuristic scenarios. Protecting consumers while encouraging technological innovation is a delicate balancing act—but it is essential to prevent abuse.

Toward Ethical and Fair Artificial Intelligence

The research presented here demonstrates AI’s enormous potential but also underscores the ethical and social challenges we must address, in alignment with legislators. Whether in machine translation, coreference resolution, or economic practices, AI can enhance users’ lives—but it can also reinforce inequalities and have negative impacts on consumers.

For AI technologies to benefit everyone, researchers, lawmakers, and regulators must work together. It is crucial that innovations be accompanied by appropriate legislative reforms that take into account both technical realities and risks to consumer rights.

In conclusion, recent studies clearly illustrate one of our era’s central challenges: how to harness technological advances while ensuring they serve the collective good. AI has the potential to revolutionize many sectors, but its development must be guided to ensure its benefits are fairly distributed. Hence our interest in regulatory compliance and our research into modeling and implementing LLMs with a minimal carbon footprint

This article was written in collaboration with Dr. Arnaud Stiepen, expert in scientific communication.