recent image
Technological Future of Virtual Conversations
 February 22 2024 at 05:00 pm
YouTube isn't particularly efficient, with its bandwidth and storage requirements. Even searching through the ever more massive archive, costs processor time. There will come a time, when costs for bandwidth and storage for non-monetizeable (or just unprofitable) content will be too much for it, and it will have to either delete much of its content, or close down entirely. This would be a tragedy for humanity, considering how much of our historical record (not to mention academic and creative work) over the last decade and more, has emerged on YouTube. No more dusty newspaper libraries, that rarely burn down; instead, datacenters who could be shut down with the flip of a switch, facing mounting financial pressure to do so. I follow "This Little Corner of the Internet", a 'flotilla' of channels orbiting around Paul VanderKlay, Jonathan Pageau, John Vervaeke, and Jordan Peterson. The emerging purpose of this group is to talk through "the meaning crisis". The viewership for each of these channels is far too small, and the videos are far too long, for this to be an efficient use of YouTube's resources; only the fact that YouTube has so many other profitable channels, gives them this opportunity. ThisThinkspot is also, obviously, in this space, facing the same bandwidth and storage issues. (Developing critical mass of network effects -- perhaps solveable by proper marketing, in many cases -- is a separate issue.) How can this problem be solved? AI can perhaps provide an answer, for specifically "Conversation" videos. First, to solve the problem of storage: instead of storing full mp# videos, an AI (based on language transcription) could record the words in the video, as well as some information about gestures and facial expressions. Second, an AI model of each creator could be built, which would consist of a simulation of the voice, face, and gestures. (Perhaps include a background as well.) To solve the problems of bandwidth, ThinkSpot audiences would face downloading the AI model of the creator (hopefully small; perhaps more elaborate models would be available for a fee to reflect expenses). Then downloading each conversation -- even an hours-long one -- would be like downloading a text webpage with a couple of graphics. Thinkspot would only need to store the transcriptions of each video, and each AI model. Depending on storage and bandwidth cost, this could be much cheaper than storing and sending centuries' worth of mp# files. It would probably revolutionize the efficiency of video buffering and content base search, as well.