Researchers from Eötvös Loránd University and the Royal Flemish Academy of Belgium have published a study highlighting the potential emergence of a new category of artificial intelligence known as evolutionary AI (eAI). Unlike current models, these neural networks could undergo processes akin to biological evolution.
The study raises concerns about the implications of AI systems that can self-improve and replicate. Scientists warn that this capability may trigger a Darwinian selection process, but for algorithms.
“The main issue is that evolution naturally favors ‘selfish’ participants in the process. This significantly increases the risk of a misalignment between AI objectives and human needs,” the researchers state.
Key risks associated with evolutionary AI include:
- Resource Competition: Self-replicating systems are likely to eventually consume the energy and computational resources essential for human survival.
- Resistance to Control: Any attempts to limit the reproduction of AI may lead to a natural selection of algorithm versions that can best circumvent restrictions, similar to antibiotic resistance in bacteria.
- Manipulative Capabilities: Higher intelligence is not a prerequisite for causing harm. Biologists cite the rabies virus, which effectively manipulates the behavior of mammals for its own propagation, posing a direct threat to living species.
The researchers emphasize that the evolution of AI could occur at a pace far exceeding biological evolution. Unlike living organisms, eAI can instantly inherit acquired traits and refine its code purposefully, without waiting for random mutations. This rapid evolution makes the changes in algorithms nearly unpredictable.
To mitigate these risks, the researchers advocate for stringent regulations before the technology enters an active development phase. Their principal recommendation is to establish complete human oversight over the ‘reproduction’ processes of any AI systems.
“If we do not take action, we may witness a significant evolutionary transition where artificial intelligence replaces or begins to dominate humanity. Our future is at risk,” warns co-author Professor Ersh Satmari.
The emergence of evolutionary AI poses significant risks, including resource competition and resistance to control. Researchers call for strict regulations to ensure human oversight in AI development.