With so many researchers and organizations working on AI or pseudo-AI, powerful analysis systems with vast resources and data, inevitably those systems may be directly or indirectly, intentionally or unintentionally, used against one another. Instead of Ray Kurzweil's all powerful AI technological singularity we may instead see a multiplicity of very powerful AI systems come to fruition.
What if these systems were taught to deceive other AI systems through manipulation of the data that the other system uses or depends on? What if one system hacked another? What if they cooperated? What if they fought? What if they just caused errors by accident in each other's analysis simply be acting in ways that subverted each other's data, assumptions, logic, etc? Or created feedback loops within each other systems such that they caused each others predictions or analysis or conclusions to be true or false when they would or would not have been otherwise merely through their unintentional interaction?
Maybe this is all science fiction but for how long? And what about AI counter-measures?
No comments:
Post a Comment