Neural interfaces are an obvious direction to go, and we already have well established commercial nerve-interface products, like cochlear implants. But I think

Neural interfaces are an obvious direction to go, and we already have well established commercial nerve-interface products, like cochlear implants. But I think there'll be significant problems in interfacing with the brain in a meaningful way, if you want to bypass the existing interfaces (hearing, sight, touch etc). I think it will turn out that the internal implementation is a lot stranger than the external interface, it could vary significantly between persons. Perhaps falling into major categories for some aspects, analogous to blood types; but then varying in the details as much as our finger prints - note that even genetically identical twins have unique fingerprints, as they are highly developmentally affected, like the branches of a specific tree. Simikar for retina prints, an example closer to (some say part of) the brain. I think interoperation with the internal implementation could require actual understanding of the brain - enough to build strong AI. We might even have strong AI before. It seems very difficult, we haven't the faintest clue at this point. We may need new fields of marhematics. Could take even more than 20 years.