I had an interesting conversation with a peer this morning, which touched on where things might be going with the various MEMEX type devices, chatbots and so on.
Of course MEMEX is not a term of art in current shoptalks, but does come to us from the 1940s with the right connotations and expectations. Rather than clinging to LLMs + LMMs as the presumed architecture, or otherwise overfitting, we bring back an old bottle ready for new wine.
MEMEX was a term coined by Dr. Vannevar Bush, the original National Science Foundation chief, when describing his vision of today’s AI but in terms intelligible at the time. Microfilm was a big part of it.
Everyone knows the direction in computing is towards ever smaller (vs ever larger) leading Bucky Fuller to propose his UMC (his ultra-micro computer), whereby human subjects might become pinheads in a good way, as metaphysical components within some “hive mind” or “zeitgeist OS”.
The picture on many drawing boards is of massive centralization in the cloud (much talk about data centers), whereas many social engineers are planning for the individual school-level MEMEX, picturing both a physical campus and a virtual sphere in Cyberia (cyberspace), in user space.
The logical boundaries for a typical school would be inclusive of theater and sporting events, debating events, any number of extracurricular, as well as curriculum-related, activities (chess club, glee club… specifics are obviously per school).
Year books, course readings, pages for faculty (active and retired), pages for students, past research… all would be raw material for the MEMEX, the computer that makes sense from a bewildering variety of saved goodies, with more always being added.
My expectation is Princeton’s MEMEX will be usable by alumni and in aggregate alumni across classes will represent the bigger slice of the pie, based on demographics. Those currently enrolled are the pipeline into the alumni tiger tank (where we tigers hang out).
However, similar to what happened in the music industry, with barriers to entry falling, thanks to the technology, a smaller personal MEMEX is also forecast and indeed is what many are busily building right now today, after which they make public videos about it. To what degree these individual onboard models are given agency varies of course.
Clearly we’re in a transitional period with AI, where many small experiments provide the feedback the bigger solution providers will need, if their relatively large scale affordances are considered affordable, meaning worth their cost.
Many possible scenarios we simply cannot afford, only fantasize about, and maybe get people to pay for anyway, as they share the same fantasy without critical insights into the practical feasibility of whatever crazy plan. Groupthink is like that. People feel safe and secure as long as they’re not alone i.e. as long as they have a critical mass of true believer subscribers to whatever vision (one would hope for more than a mere vision).
Those with a stronger sense of the Reality Principle serve to help filter out the truly crazy schemes. Some do get through though, and those involved find out the hard way why they should never have gotten funded in the first place. By the time that becomes obvious, it’s too late to course correct, while meanwhile the crazy think tank professors who dreamed this stuff up, are off to other assignments. There’s no one to call.
When the game is to try many things, going with fast and small over slow and big, the consequences of a scenario failing or wiping out will be manageable as well as instructive.
Some plans look like they could work and one only realizes in hindsight about some of the obvious flaws.
Learning from one’s mistakes is of prime importance, in order to avoid one of those doom loops we always hear about, wherein a failure to realize what’s self-defeating is in itself a blind spot in the analysis.












