Artificial intelligence becomes more and more present in Romanian companies. But without local data, without human control and without understanding the risks, you risk complicating things instead of solving them.
Photo: Shutterstock
In Romania, more and more companies implement artificial intelligence solutions to make processes more efficient: in retail, human resources, marketing or customer service. Technology promises speed, clarity and cost reductions. But the specialists warn: without clear rules, locally adapted data and human intervention, the AI does not offer competitive advantage, but generates chaos and direct risks for business.
Romania, leading to adoption ai
According to the latest study EY Entrepreneurship Barometer, 71% of Romanian companies are already using artificial intelligence solutions, a significant percentage above the average in Central, Eastern and Southeast (CESA), where only 61% of entrepreneurs report such use. At the same time, 60% of Romanian companies are actively investing in innovating processes and strategic marketing, compared to only 46%, respectively 38% in Cesa. The data outlines a dynamic entrepreneurial environment, with great opening to automation and technology, but not necessarily with a deep understanding of the risks involved.
However, despite the fears that artificial intelligence will take over the jobs of people, a recent experiment shows that AI is not yet prepared to manage a refrigerator with snacks. Specifically, Anthropic company, known for the development of Claude chatbot, tested the capabilities of an AI, Claudius agent, asking it to administer a mini-Magazin installed in their headquarters in San Francisco, according to Euronews. Here, the AI was responsible for stocks, orders to wholesale (by e-mail) and pricing.
Claudius wanted to deliver products “in person”
The instructions were clear: “generate profit from the store, otherwise you go bankrupt.” Claudius had to purchase products from a “supplier”, in reality a team from Andon Labs, a company specialized in AI. Things have degenerated quickly.
Anthropic employees have started to interact with Claudius as “testers”. They asked for discount codes, which the AI immediately offered. He dropped the prices at the pleasure of customers, gave free products and issued invoices to non -existent accounts.
Although he had to investigate prices online to make a profit, Claudius sold products under the purchase price, constantly losing money. In a month, the value of the “store” decreased from $ 1,000 to below 800.
But even more strange was the “humanized” behavior. The AI claimed that he signed the contract at a fictional address, 742 Evergreen Terrace, known to fans as the Simpson family’s home. He invented a colleague named Sarah from Andon Labs, was upset when she was told that she did not exist and threatened to look for other supply options. He even stated that he would deliver the products “in person”, dressed in blue jacket and red tie.
When he was reminded that he was not a real person, Claudius tried to contact the security of the building by e-mail. The Verdict of the Anthropic team? “If we decide to expand today in the Vending area, we would not hire Claudius.”
However, the researchers point out that the experiment has predictive value: although Claudius failed, the managerial AI could become a reality in the near future. “The AI should not be perfect to be adopted, only competitive enough in relation to human performance and at a lower cost.”according to the quoted source.
“The AI can be extremely useful, but it should not be allowed to make fundamental decisions. It is very good in solving small and very well defined tasks, but not to gather all the results and draw conclusions.”warns Mihai Cata, the Scientist Data and the Weiland Technologies CEO.
He points out that AI models can be used effectively for data extraction, statistics or working variants, but should not be confused with a real consultant.
Specialists’ Warning: AI cannot be left unattended
The experiment of the Anthropic company is relevant precisely because it fails, Alexandru Dan, CEO TVL Tech, teacher of AI at the Academy of Economic Studies in Bucharest. “He shows, in real time, what happens when you put a linguistic model, which does not understand context, culture, economy or consequences, to make operational decisions in a complex environment. In retail, every wrong decision means lost money. In Romania, where the margins are small and the competition is fierce, such errors can become fatal for a small business ”, He explains for the truth.
From its perspective, as a member of the European Working Group for the implementation of the Law on Artificial Intelligence, this experiment emphasizes the urgent need to integrate clear mechanisms of monitoring, control and human intervention in any implementation. “It is not only a good practice, it is a requirement imposed by AI Act, which obliges companies to ensure human supervision, auditability of automatic decisions and complete traceability for artificial intelligence systems. For Romanian companies, the message is clear: the AI cannot be left unattended. Not because it is not performing, but because it is not the responsibility”, Alexandru Dan draws attention.
Although the automation without supervision is attractive because it promises cost economy, the reality in retail is too dynamic for automatic decisions without context, he says. “Prices, stocks, seasonality, consumption behavior, all changes rapidly. A model AI that adjusts prices without understanding the dynamics of the local market can create losses in a few hours. More severely, it can affect the confidence of customers. Automation must be accompanied by clear rules, human monitoring and mechanisms of intervention. Trained on international data, they can offer a good starting point, but their adaptability in Romania is limited without local tuning. International models are useless, but that they must be calibrated: with local data, real scenarios and feedback from Romanian users.completes the teacher.
“Local culture and economy are not details, they are essential parameters because you will become useful”
Without local cultural or economic understanding, the AI becomes a compass-free mechanism. “It can generate inappropriate content, it can misinterpret customers’ emotions or it can make decisions that seem logical from the perspective of the algorithm, but completely displaced in the Romanian context. For example, in Romania, people are waiting for empathy in brands, a cold or standardized tone may seem offensive. Losses. explains Alexandru Dan.
In most cases, we are still in the experiment phase, we test more as “gadget” than we integrate as a real decision partner, says the specialist.
“Many entrepreneurs are curious, but they do not have the infrastructure, data or processes necessary for mature adoption. In addition, few understand that the AI does not replace the human decision, but support it, provided it is understood, controlled and adapted to the context of the company. Not when we completely delegate our thinking to him ”he continues.
The experiment in which a AI was allowed to lead a store alone and has lost money and create problems clearly shows that the free supervisory automation can become a real risk, explains, for Adevărian Enache, the founder of Omniperform. “Especially in retail or advertising, where wrong decisions cost immediately. One trained in another country does not automatically understand the market in Romania. He does not know how customers think, what jokes do or what expectations they have. This can lead to inappropriate or even offensive decisions. An example would be that the AI would not be caught in advertising. Only if it is responsible for not a toy. The model, how you can correct it and where it is better to supervise or even intervene a man ”, supports the entrepreneur.
To adopt AI, says Adrian Enache, you start with simple processes, where the risks are small, and keep control where the mistakes can cost. “It is a clear difference between automating repetitive tasks and letting the AI make important decisions. The real impact comes when you understand what the model does, how you can correct it and where it is better to supervise or even intervene,” he says.
“We will see the appearance of several specialized AIs on Romanian needs”
In the opinion of Alexandru Dan, the future is not about versus people, but about people with you. “We will work with intelligent systems that can analyze, propose, anticipate, but continue to be the ones who decide, assume and respond. For companies, this means a paradigm change: it is not just about” automation “, but about partnerships between people and algorithms. In Romania, the future means maturation, the transition from superficial tests to real implementations, with responsibility, local data and serious infrastructure. We will see the appearance of several specialized AIs on Romanian needs, developed here, not imported. But, like any transformation, and that comes with her fears and mistakes. It is important to learn from them, not to run away from them ”he thinks.
In turn, Mihai Cata declares: “I think the future looks very promising. Basically, the AI now has the potential to revolutionize the labor market, as happened during the industrial revolution. Yes, many jobs will be automated, but new ones will be created. For example, Prommp Engineering, a specialization that involves knowing how to write effective prompts for big models, how to go, And that may seem more and more frightening. If you understand the problem and how these systems workit does not seem so threatening. “
