Representational tethers are a merger of [[bridger-languages]] and [[representational-alignment]]. They are means of relating the internal representations of ML models with internal representations employed by humans. One part of this process is incentivizing the artificial representations to be cross-compatible with humans ones (the legacy of [[representational-alignment]]). Another aspect is enabling people to perceive and understand said artificial representations by making them more legible and cognitively ergonomic for humans (the legacy of [[bridger-languages]]). The two previous frames complement each other: it's easier to understand a language containing concepts you're familiar with than a more alien one (a shortcoming of [[differentiable-cosmogonies]]), and it's easier to keep a representation within reach if you make sure it's consciously accessible at the other end. Representational tethers are meant to further alignment by improving human oversight or automated versions thereof.