Could 'expiration dates' for AI systems help prevent bias?

Could AI models programmed to expire help to mitigate bias? Some experts believe so, but the technical barriers are significant.