Workers who copy themselves. An engineer from the Chinese technology company Baidu trained his avatar to continue doing his job once he left the company. The story, published by the digital magazine Sixth Tone, raises an uncomfortable possibility for workers. In such a scenario, a company could preserve the knowledge and know-how of its employees to create digital replicas capable of assuming their functions and even replacing them. The idea can be attractive to companies because it promises to retain experience, reduce dependence on certain profiles and prevent knowledge from being lost. But would it be legally viable in Spain? The experts consulted agree that an avatar capable of imitating a specific worker would collide with multiple legal limits.
Juan Medina, social graduate and director of Ecualis Asesores, considers that the main red line appears when the company does not stick to preserving corporate knowledge, but rather replicates a specific person. “Automating tasks and preserving the knowledge generated in a company is legitimate; what a company cannot do is build and exploit a recognizable replica of a person,” he explains. The first regulation with which it would come into conflict is data protection. To train a system that imitates a worker it would be necessary to use information linked to him, from emails and messages to voice recordings, documents or activity histories. Blanca Liñán, partner at Ceca Magán Abogados, highlights that the main risks appear “when the system incorporates or reproduces data, identity or individualized traits of the worker.”
In these cases, the employee's consent alone would not resolve the issue. Cristina Prieto, senior associate at Abdón Pedrajas Littler, remembers that “consent can be compromised by the situation of dependency inherent to the employment relationship.” Consequently, the worker's authorization would not automatically legitimize the use of their data to create a digital replica of their person. Added to this would be conflicts related to the right to one's own image and the protection of other personal attributes. Leandro Núñez, partner at Audens, believes that these systems open a debate that goes beyond data protection. “A broader question is beginning to arise: whether a fundamental right to a person's digital identity should be recognized,” he says. And it is no small matter: technology already makes it possible to build replicas capable of reproducing the voice, image or behavioral patterns of an individual. Labor regulations also collide directly with the possibility of companies cloning their staff. Article 1.1 of the Workers' Statute defines the employment relationship as a personal provision of services. Hence, as Prieto points out, “a company cannot replace the worker's position in the contract with an AI system.”
Substitute machines
The scenario would be even more delicate if the company used this digital replica to eliminate the worker. Medina recalls that, although labor legislation does not prevent automating tasks, it does require that any contractual termination be duly justified by a technical or organizational cause. “Just because a machine can do a job does not, in itself, make the dismissal of the person who did it valid,” he warns. Furthermore, the European AI Regulation introduces new obligations for systems that may affect people in the workplace. Liñán emphasizes that “an avatar that only executes tasks is not automatically high risk.” However, he adds that the situation changes when the tool serves to make or support decisions about hiring, evaluating, promoting, supervising or dismissing workers. In these cases, the requirements for transparency, risk assessment and human supervision provided for in the regulations come into play.
Another fundamental question raised by these tools is who owns the professional knowledge. “What the worker produces in his work belongs to the company and what he knows how to do is his,” summarizes Medina. The problem arises when an AI mixes business assets with knowledge or traits of a specific person. In these cases, Liñán warns, the fact that the company can use the former does not imply that it “automatically acquires any rights” over the latter. Added to this is another risk linked to intellectual property. As Núñez points out, “copyright protects only human creativity.” For this reason, the lawyer points out, if the human contribution disappears from the process and the content is generated entirely by a machine, the results would be left unprotected and could be freely copied and reused by competitors, even to train their own systems.
These digital clones would be the extreme of a trend that already exists in Spanish companies. Human resources departments are increasingly turning to algorithms to hire, organize, evaluate and supervise workers. Prieto highlights the growing use of time control tools, geolocation, activity and productivity monitoring, as well as task assignment and performance evaluation. In their opinion, these practices entail risks related to discrimination, psychosocial health, the information rights of the workforce and excessive surveillance at work. Regarding this, Núñez considers that the main challenge for companies is to guarantee that workers understand what information is collected about them and how it is used. “Transparency requires being able to explain what the system measures, what it is for and what effects it can have on people,” he concludes.
The impossible erasure
The conflict does not end when the worker leaves. As Leandro Núñez, partner at Audens, explains, “withdrawing consent is easy, but getting an AI to unlearn what it learned with your data is not.” Unlike a database, knowledge is distributed among millions of parameters and there is not always a specific record that can be located and deleted. The debate has already reached the EU. Some proposals contemplate that, when deletion is disproportionate or technically unfeasible, measures be applied that prevent further use of that data or exploitation of the results obtained from them.
