Inspired by Stevan Harnad’s example (though without his genius, alas), I input the final few pages of Vannevar Bush’s famous 1945 article in The Atlantic (“As We May Think”) into ChatGPT just now. Bush’s piece is a seminal text in LIS, as his description of the Memex is often (though not always) considered to prefigure how scientific communication and, accordingly, the libraries involved, would develop during the latter part of the 20th century, so I was curious as to what ChatGPT might make of it.
The 1945 article itself is currently behind The Atlantic‘s paywall, something that Bush probably did not envision, but fortunately I still have a copy from my teaching days, along with his Science: The Endless Frontier, which served as a blueprint for much of America’s science policy during the post-war decades. Bush also “revisited” the Memex in a chapter for his 1967 book Science is Not Enough, but that piece focused primarily on the idea that technological advances were rapidly bringing Memex-like technology close to reality. (And, sadly, the book has now been removed from the Open Library section of the Internet Archive, but that’s a topic for another post.)
So here’s ChatGPT’s summary of those pages in which Bush discusses the past and future of scientific communication as seen in 1945:
The article reflects on the profound impact of science and technological advances on human life, exploring both their benefits and limitations, particularly in the context of research and the growing complexity of scientific knowledge.
This passage explores the growing challenges of managing and accessing the vast quantities of information generated by modern society. It highlights the increasing volume of recorded knowledge, which is difficult to navigate and utilize effectively. The primary issue is selection—the process of identifying relevant information from a massive database, a process that, despite mechanical improvements, still struggles to keep up with the speed and complexity of human thought and memory.
The author contrasts two types of selection systems:
- Simple selection, where each item is examined individually (e.g., sorting employee records or matching fingerprints).
- Efficient, associative selection, which resembles the automatic telephone exchange system that identifies a call recipient based on a few digits, speeding up the process considerably.
However, the real challenge is not just mechanical selection, but the indexing systems we use to store information. Current methods often involve rigid, hierarchical classifications that don’t mirror the way the human brain works—by association—which can limit both the speed and flexibility with which we access information. Human memory links items by associations, while current storage systems force users to follow predefined paths, which can be slow and cumbersome.
The author envisions a memex, a hypothetical future device that would serve as an extension of human memory. The memex would be a desk-like machine where an individual could store and rapidly access vast amounts of information, such as books, notes, and records. The system would allow the user to create associative trails, linking items together based on personal relevance or context, allowing for flexible and immediate recall. The memex would offer an enhanced, personalized way of navigating knowledge, far surpassing traditional indexing systems.
By tying different pieces of information together into trails, the memex would support more intuitive and efficient knowledge-building, making it easier to explore complex topics across multiple disciplines. The device would represent a powerful tool for scientific research, professional work, and personal study, helping users sift through the ever-growing record of human experience and knowledge.
The passage ends by pondering the potential for future innovations in information processing, including direct interfaces with the brain, which could further revolutionize how humans interact with and benefit from accumulated knowledge. It stresses the importance of mechanizing the record of human experience to enable progress, while recognizing that society must avoid becoming overwhelmed by the sheer complexity of the task. Ultimately, the goal is for science and technology to help humanity harness and build upon the vast record of knowledge, avoiding stagnation and conflict while encouraging growth and wisdom.
This seems a fair summation by ChatGPT of the points of interest for LIS in Bush’s vision (though, of course, it has removed much of Bush’s individual style and the illustration of the Memex at work that Bush provided in his piece, which is shown in this animation):
Ironically, 80 years after Bush’s essay, we’ve reached the point at which “the goal for science and technology” represented by such technological titans as OpenAI and its “partners” Google and Microsoft is now apparently to, as the author of this new essay puts it, “flatten the architecture of thought” by providing AI-summarized chunks of putatively factual text upon user demand (on Google, this appears as the “AI Overview“) but the organization of which minimizes Bush’s “associative trails” (currently represented by hyperlinks or other citations) by which the user could assess the provenance or probity of the source of the information. It is well-known in LIS that users tend not to scan more than the first few hyperlinks in a query response, so this may further detract from the potential accuracy of the results.
While Google does warn users that the “AI Overview” may be inaccurate and should be checked, another issue is the so-called “hallucination” phenomenon of ChatGPT disseminating very obviously inaccurate information. (And the term “hallucination” gives unwarranted credence to the notion that ChatGPT is actually conscious.) LLM researchers are far from solving this critical problem (and, indeed, I think it is unsolvable, given that the possible questions and answers are limitless, unlike even LLMs.) I had touched on this briefly in Chapter 10 of my book, as such a turn to corporate-constructed “artifici-facts” is clearly disruptive (and probably disastrous) to LIS research (and to research in general, for that matter) if people become habituated to their use.
Ten years ago, as libraries were building discovery layers into their catalogs, hoping to emulate the ease of Google’s one-box searching for their users, librarian Barbara Fister famously wrote that “The Library Isn’t Flat,” because “knowledge isn’t flat. It’s a set of ongoing conversations among human beings who share and build on and deconstruct and quibble over what to make of the world. We consider it ethical to trace those conversations in our scholarship and are pleased when our contributions are acknowledged and built on. It’s a shame that when libraries devise discovery systems they’re adapting to Google but not to the memex. Discovery layers are great if you’re approaching research as a consumer – I need five peer-reviewed articles and two books on the subject of neoliberalism. Done in five minutes! But flattening the world that way makes it look as if nothing is connected, that sources are things produced through some mysterious process that are selected and put in your shopping basket for checkout.”
Today, the research process itself, inside or outside the library, appears to be in danger of being further flattened. Fister wrote in a 2023 opinion piece in Inside Higher Ed (also paywalled):
“As the novelty of ChatGPT drives a wave of hype, are we repeating the same mistake with worrying about how it will change what we do in the classroom, when instead we should be thinking about the unintended consequences of AI hubris? … Unlike Wikipedia, ChatGPT and similar products under development are proprietary, not the product of a nonprofit powered by volunteers and small gifts. It ingests data from the web (including the entirety of Wikipedia) but unlike with Wikipedia’s history and talk tabs, there’s no visible information about how the chat bot arrives at its answers or how it “decides” what sources to draw on. In fact, it invents sources when mysteriously moved to do so… . ChatGPT is geared toward customer satisfaction rather than facts. The company is planning to build in options for customizing results to match one’s political and cultural values. By avoiding taking a position on truth, ChatGPT can be fine-tuned to give users the answers they prefer. . .. And all this is happening without any kind of oversight. Once again, Big Tech is moving fast, and even its leaders have no idea what they might break next.”
What Bush’s view on all this would have been were he alive today, I am not sure, given his belief in science and technology as a dominant force in society. I would hope that he would be in agreement with the views of Archibald MacLeish, the Librarian of Congress, who wrote “The Librarian and the Democratic Process” in 1940, a passionate defense of the importance of freedom of information to democracy.
Although “As We May Think” was apparently written in 1939 prior to Bush’s appointment as head of what was to become the U.S. government’s Office of Scientific Research and Development, the essay was actually published in July 1945; the last German city, Berlin, fell to Allied forces in May 1945, while the atomic bomb targeted Hiroshima in August 1945, forcing Japan’s surrender, so by then the power of science and technology was very apparent to all.
No, on further reflection, I believe that Bush (and, of course, MacLeish) would be opposed to flattening the scientist’s Memex (and also that of the citizen and student), which all represent individual contributions to collective knowledge. He saw the results of such flattening during the war, on scientists and civilians alike, in both allied and enemy nations, where information without context and context without information fueled the spread of propaganda. I’m also certain he would be against either the further politicization or proposed defunding of the National Science Foundation, which he was instrumental in creating in the aftermath of World War 2.
And now, following my own associative trails, I’m off to find Collin Jennings’ book, Enlightenment Links, which I hope to soon be reading in conjunction with Martin Mulsow’s Knowledge Lost: A New View of Early Modern Intellectual History!
Keywords In and Out of Context 