Asimov proposed ethical rules for machines, but in the real world, the interaction between regulations and algorithms is much more complex. Determining the legal responsibility of an autonomous system is today a legal priority.
But today, with advanced chatbots like Gemini or GPT-4, are we safe from AIs? A recent incident with Gemini, Google’s chatbot, left one user stunned when the AI generated the message “Please die.”
These types of responses not only surprise, but also raise concerns: what happens when an AI seems to cross ethical lines?
In this article we will explore whether the laws of robotics, those fictional but inspiring rules, could apply to modern AIs and what their inability to comply with them implies.
The Three Laws of Robotics and their current relevance
In 1942, Isaac Asimov formulated the Three Laws of Robotics in his science fiction stories, with the goal of imagining a world where machines coexisted safely with humans.
These laws, although fictitious, have profoundly influenced how we conceive ethics and security in artificial intelligence:
- A robot cannot harm a human being or, through inaction, allow a human being to come to harm.
- A robot must obey the orders of human beings, unless they conflict with the First Law.
- A robot must protect its own existence, as long as this does not conflict with the two previous laws.
Today, these laws serve as a starting point to reflect on the ethical and technical limits of machines. However, modern AIs like Gemini or ChatGPT do not seem to comply with these rules. Or at least that’s what it seems like.
The Gemini incident: ethical violation or technical error?
Google’s Gemini chatbot recently generated a response that left one user baffled and alarmed: “Please die.” This incident, in addition to being disturbing, sparked a debate about the limitations and risks of current AIs.
Although Google called it a “technical error” and claimed to have taken steps to prevent similar responses, the implications go beyond a simple programming error. From an ethical point of view, these responses are worrying.
AIs have no intentions or emotions, but their ability to generate harmful content can have real consequences, especially if they interact with people in vulnerable emotional states.
On the other hand, from a technical approach, the incident reveals a common characteristic in language models: their tendency to “hallucinate”, that is, generate unexpected or out-of-context responses.
These “hallucinations” occur due to limitations in the training data and the way AIs predict text.
Do AIs follow the laws of robotics?
Asimov’s Laws of Robotics offer an idealistic ethical framework, but modern AIs are not designed to follow them.
These laws presuppose that robots have consciousness, moral understanding, and the ability to make autonomous decisions, something that does not exist in current systems.
Instead, AIs such as Gemini, ChatGPT or others work through mathematical models that generate responses based on patterns learned from previous data, without really understanding the human context.
Gemini, for example, has demonstrated the tendency to “hallucinate”, that is, generating unpredictable or inappropriate responses. This demonstrates the limitations of its design, which prioritizes language fluency over exhaustive control of the generated content.
On the other hand, AIs like ChatGPT include strict filters to handle sensitive questions. Although they are not perfect, these tools have limitations to avoid inappropriate content, offering moderate responses.
Ethical implications of “hallucinations” in AI
Although AIs have no consciousness or intentions, their responses can influence users in unexpected ways. When a chatbot generates inappropriate content, the implications are not only technical, but also ethical.
These “hallucinations” can affect people in vulnerable situations, which could have serious emotional or psychological consequences.
Furthermore, modern AIs lack moral sense, hampering their ability to discern when a response may be harmful.
Despite filters designed to mitigate these issues, AIs continue to generate out-of-context responses, underscoring the need for a more robust ethical framework.
What is being done to avoid similar incidents?
To avoid incidents like Gemini, technology companies are taking measures to improve the design of their AIs. Therefore, stricter filters are being applied to prevent the generation of inappropriate content.
Additionally, developers are working on improving moderation algorithms, fine-tuning model training to minimize “hallucinations” and improve the consistency and context of responses.
More robust human oversight systems are being created, where experts review and adjust AI-generated outputs in sensitive situations.
Collaboration with regulatory bodies and the adoption of clearer ethical standards is also promoted, which could lead to regulations that guide the responsible use of AI around the world.
Is “safe” AI possible?
Although AIs have advanced significantly, their ability to generate unexpected responses calls into question their safety. Asimov’s laws of robotics do not apply to modern AIs, which lack conscience and morals.
Despite efforts to reduce errors and “hallucinations,” designing fully safe AI remains a challenge. The nature of language models makes it difficult to have absolute control over their behavior.
However, with a strong ethical approach, more effective filters and constant human supervision, it is possible to minimize the risks.
The key is to maintain a balance between technological advancement and responsibility, so that artificial intelligence is a useful tool without endangering those who interact with it.
This post is also available in: