Artificial intelligence has ceased to be a futuristic motif borrowed from films. Today it helps decide who gets a loan, how medical diagnoses are made, who is shortlisted in recruitment, how content is moderated, and even what we see on social media. As this acceleration continues, a simple question becomes increasingly urgent: how can we make sure that AI remains on our side?
This question is addressed in the article “3 Ways to Keep AI on Our Side: AI Researchers Can Draw Lessons from Cybersecurity, Robotics, and Astrobiology”, published by IEEE Spectrum. The authors include Bruce Schneier, one of the world’s best-known experts in digital security, Nathan E. Sanders, and Professor Dariusz Jemielniak from the Department of Management in Networked and Digital Societies at Kozminski University. Drawing on three perspectives, cybersecurity, robot ethics, and astrobiology, they propose a different way of thinking about AI safety.
AI makes different mistakes than humans
The first argument builds on an observation familiar to many users of generative AI. Systems based on machine learning make mistakes in ways that are fundamentally different from human errors. They can produce answers that sound highly convincing while being factually wrong, invent sources, or generate content without any genuine understanding of what they are producing. From a safety perspective, this has an important implication: we cannot simply transfer existing methods from IT security or human-centered risk management to artificial intelligence.
Schneier and Sanders argue that AI should be treated as a new kind of actor within the security ecosystem. On the one hand, it can amplify existing threats, for example by automating attacks. On the other, it generates entirely new and often unexpected forms of error that standard scenarios fail to anticipate. As a result, AI safety cannot be reduced to making machines more “human-like”. What is needed are mechanisms that explicitly account for the alien logic of AI systems and are designed to detect and mitigate the consequences of their distinctive failures.
A new “law of robotics” for the age of deepfakes
In his contribution, Professor Dariusz Jemielniak refers to a classic point of reference: Isaac Asimov’s famous three laws of robotics. In popular culture, these laws served as an intuitive ethical compass. A robot must not harm a human, must obey orders unless they conflict with the first law, and must protect itself as long as this does not violate the other rules. The problem is that contemporary AI no longer resembles the mechanical robots imagined by Asimov. Today, the greatest risk is not physical harm inflicted by machines, but the possibility that AI systems will mislead us subtly, at scale, and with great credibility.
Generative models for text, images, and video enable the creation of deepfakes, fabricated statements, false images, and messages that sound entirely authentic. In this context, Professor Jemielniak argues that robot ethics requires an update. Instead of focusing solely on the prohibition of physical harm, our norms must explicitly include bans on deception, requirements for machine recognisability, and transparency about the origin of content. The goal is not only to ensure that AI does not lie, but also to give users a real chance to recognise when they are interacting with a system rather than a human, and to establish mechanisms for verifying authenticity in the public sphere.
In practice, this means designing not only individual models, but entire ecosystems. These should include rules for labeling AI-generated content, controls over its further use, and clear accountability for platforms that integrate AI into their services. Asimov’s laws remain a valuable source of inspiration, but they are no longer sufficient as a practical code for the age of deepfakes.
Lessons from the search for intelligent life in space
The most surprising perspective emerges in the final part of the article, where the authors suggest that AI developers could learn from astrobiology, and more specifically from the search for intelligent life in the universe. In that field, researchers must define extremely rigorous criteria. What exactly counts as an intelligent signal? How can it be distinguished from background noise? How can false positives be avoided?
Similar questions should be asked in discussions about so-called strong AI or systems approaching human-level performance. Rather than relying on emotional declarations that “this is real intelligence”, the authors propose standards comparable to those used in space research. These include clear testing protocols, repeatable criteria, and independent verification. The aim is to move beyond intuition and narrative, and toward scientifically coherent methods for assessing what we are actually dealing with.
This analogy works on two levels. First, it tempers the imagination and reminds us that claims about breakthroughs should be treated with caution. Second, it shows that building a safe future with AI requires procedures and institutions, not just promises made by technology companies at conferences.
