A[I]s We Many Not Think (nor search, nor link)

Somewhere in recent web travels I read someone’s post who had just discovered the insight and inspiration of Vannevar Bush’s seminal 1945 essay published in the Atlantic in 1945 As We May Think (linking to the open ACM version as sadly the original one from the Atlantic is stuffed behind their foul paywall).

I have so many associative trails flying off of As We May Think, and trying to organize them linearly in writing is just problematic. But with that said, I will aim to tour through where the thinking started: (a) The problematic public whinging of the “brokenness of web search”; (b) the opening to me of the essay as part of Gardner Campbell’s innovated New Media Faculty Seminar (2010-11?) (c) how my own thinking/experiencing information flows is based on experiential associations, which are connections that are on the far side of the statistical probability hemmed in world people are gushing about with GenAI; and (d) My own loose concept of my own memex.

Are you ready for a long blog ride? Strap yourselves in.

Web Search is Broken

This refrain seems everywhere and ends up being a backdoor into warming the alleged powers (which I disavow) of AI Search. The gel started with Bryan Alexander’s Will AI Search Gut the Web? (for which I remembered reading, had not saved, but quickly located it via a search on bryan alexander substack google search – how can search be broken when this works so easily?)

Generative AI continues to develop new forms and applications. One of those is AI-generated search, which involves using an LLM-backed service instead of searching the world wide web through Google Search or others. Today I’d like to explore some ways this might play out. The scenario I have in mind is one where we’ll see AI search clobber the web, with LLMs replacing a big chunk of the web ecosystem in much the way tv succeeded radio.

https://aiandacademia.substack.com/p/will-ai-search-gut-the-web

The first kicker is this whole myth of “searching the web” that completely gets lots in the wash. There is nothing that actually searches the web. It’s too vast, sprawling, for anything to search it. It’s like grabbing jello. All search engines search indexes of the web, which are imperfect, incomplete. That to me, is a feature, not a bug, that no entity short of Stephen Downes as read and tracked everything on the internet.

Also, in the same way we conflate so many discrete concepts when we lump them all as “AI” the same goes for search. Pulling from my comment to Bryan, “Searching for specific facts or answers differs from searching when researching differers from what I most enjoy, just looking to explore. For the latter, I get more out of results beyond the results, things I would not even consider or just what pops up as curious. GenAI removes serendipity.”

Search is More than Getting Answers

There is this consistent focus on search being getting answers to questions. I had read something I could not find, then recall it was a podcast, yes the podcast from MIT Technology Review Narrated (which has no link of its own) on the article AI search could break the web (paywall freed link). Over and over it uses examples like looking for restaurants or sports scores or just fact look up stuff.

I absolutely do that, but more often than not, I am not looking for answers in my search, but more sources to look at, or more things to read, track that I do not know if or things that just raise my curiosity of interest. I am (to foreshadow Vannever Bush) seeking to augment my associative trails. Search and review is part of my thinking, not just looking up the score of the hockey match.

When Google burst on the scene it was almost like magic. Earlier searches like Altavista and the curated hierarchy of Yahoo were knocked off the perch. Google search worked so damn well when the web was smaller, and more interlinked (coming back to this), and more original content.

And gone from most people’s tool chest are the uses of search modifiers. So strangely enough, even if looking past the AI summaries and the crap of paid posts, with just the complex act of scrolling and using my brain (aka thinking) strangely enough I am, able for the most part to quickly locate either an answer or a source, but also interesting stuff.

But my brain is working through the information, doing lots of micro-analyzing, comparing to all the stuff I have seen/read in the last 60+ years (well what I can recall)- thinking is active work, and there’s almost nil of that when you take what pops out of a vending machine.

Go Upstream My Friends

These days what is most trafficked in google are all the repackaging of information as news sites or stuff that is easily shoved through social media, teriarty sources just designed to have you get blasted by ads or pay to not be blasted by ads.

While we can bemoan AI Slop muddying and spreading poop across the web, it is much a mess from the proliferation of sites that are not original content, but packaging of content from elsewhere.

Here’s a small example, someone in Mastodon shares a link for “Bounce” which sounds like an interesting, possible, approach for connecting Bluesky and Mastodon. But the link they share is to a TechCrunch “article” (clear the pop up) It’s just a rehash. Why not go one link upstream to the source itself to share, the Bounce site? Is that not a better source? Free from ad surveillance?

I might be whinging, but this is the forgotten fundamental of Mike Caulfield’s SIFT or originally the Four Moves, number 4 to trace things back to their source.

But there’s such a rush in social media to quickly share, who has time for that?

Me.

Here’s my trick. I do sift my RSS Feedreader and my social feeds for interesting resources, but largely, the more interesting stuff happens 1, 2, sometimes 5 links from the original shared one. Often you find a reference that sparks interest. I see a link deep in a domain, and I often go to the root. Who is this person? Organization? Look they have other resources, projects of interest.

GenAI would never lead me to these things off of the original search (I hear your “but they might get you started”).

As We May Not Link

Do you remember why Google’s early search was so innovative? It was based on what sites linked to what. Who here remembers hyperlinks? You know the very fabric of the connected world wide web?

How many durable hyperlinks have you added to the web this week? With so many giving up writing their own blogs, web sites where the act of linking was just natural, we are collectively at some responsibility for the decay of the web. I work so often with colleagues who just do not bother to add links when writing content in web sites. It galls me that folks do not naturally think of linking.

And of course there’s sites like Newsweek who only internally link to their own stories.

And I doubt that link sharing in social media helps, those links are fragile or if you are in foul places like LinkedIn ro that other single letter place, always obfuscated in their own links appended with tracking crap.

You can’t lay all the blame on on big tech for search going sour if you are not doing your own little bits to weave the web with hyperlinks.

GenAI is Just The Evolution of I Feel Lucky Button

So in the old days, you could “feel lucky” with just plopping 3 keywords in and searching. And to me that’s almost the glossy thin fake gold paint veneer or GenAI, it’s dropping some text into a box, and trusting the results of the “I Feel Lucky” button.

Because its almost free from effort.

From thinking.

The Original As We May Think

I give all the credit for my first expoure to the genius of Vannevar Bush to Gardner Campbell. Back in 2010, 2011, maybe beyond, he organized something called the New Media Faculty Seminar first at Baylor University then when he took it when he moved on to Virginia Tech. I got involved more when we did things together through the New Media Consortium (NMC) (RIP).

It was his effort to open minds, generate discussion, of yes, thinking, to the foundational writings and people in the New Media field (is there such a thing). He can explain it better, but I took a stab in my own post:

Maybe Not Massively, But Definitely Open, Faculty Seminar

A few months ago and an NMC Board meeting, Gardner Campbell conjectured the idea to try and create some sort of online reading group among our community

…

Over in Texas where he directs the Academy of Teaching and Learning and Baylor University, Gardner was already planning to host a New Media Seminar for faculty, more or less, applying the syllabus he has done so successfully for years as an undergraduate course From Memex to YouTube: An Introduction to New Media Studies.

I cannot give justice to a description, but if you ever had heard Gardner do a presentation, you can imagine how electric he would be for an entire semester– he leads students through the history of ideas and innovations that gave rise to modern computers and this internet landscape that seems familiar to many of us, focusing on the ideas and the people who made this happen. I’ve heard him describe more than once the joy of seeing his students light up in amazement after reading what might be seen as “old” essays on this New Media history.

NMFS was framed around The New Media Reader edited by Noah Wardrip-Fruin and Nick Montfort. You can even see on the cover the importance of Vannevar Bush, that’s the photo from As We May Think.

The texts are from computer scientists, artists, architects, literary writers, interface designers, cultural critics, and individuals working across disciplines. They were originally published between World War II (when digital computing, cybernetic feedback, and early notions of hypertext and the Internet first appeared) and the emergence of the World Wide Web (when these concepts entered the mainstream of public life).

https://www.newmediareader.com/about.html

In the era of the original networked open ideas and the original connectivist style MOOs, the idea was Gardner leading a seminar group at his institution, they blogged ideas he syndicated to a “mother blog” but through our NMC collaboration and his contacts, we struck up seminars elsewhere, like when Tom Haymes was at Houston Community College, and more I forget. It was like a loose federation of seminars happening locally. There’s a raft of lost technologies- a WetPaint Wiki, netvibes, and more moldy web oldies.

For the 2012 edition of NMFS Gardner brought in the MOOS metaphor:

And now my new unofficial category for this endeavor, now in its sixth iteration for me in this form, is a MOOS: a massively open online seminar. (Apologies to Northern Voices and its mooseology–I won’t say branding–and I hope they will not be angry with this petty theft by a friend.) I think the “massively” is important, in that it modifies “open,” not “seminar.” That said, it’s also important for me that this experience scale somehow, across institutional boundaries both internal and external.

https://www.gardnercampbell.net/blog1/nmfs-fall-2012/

When I aimed to do for NMFS through NMC was doing a weekly podcast with Gardner foreshadowing the next week’s activity/readings. I find shreds in the Internet Archive of the NMC web site (which sadly most was tossed aside when EDUCAUSE bought the assets at a yard sale), but damnit, all of the audio was munged behind some kind of flash player.

But hold the bus! I kept my own copies of the audio! All of the recordings from Spring 2010 and Fall 2010. Here is the opening podcast as an introduction.

It’s All About the Trails

I won’t be giving a in depth summary of As We May Think. There’s a lot there with some of the mechanics, the storing information in microfilm, organizing it. The biggest takeaway was and always is the idea of associative trails (look I just made one)

An associative trail as conceived by Bush would be a way to create a new linear sequence of microfilm frames across any arbitrary sequence of microfilm frames by creating a chained sequence of links in the way just described, along with personal comments and side trails. At the time, Bush saw the current ways of indexing information as limiting and instead proposed a way to store information that was analogous to the mental association of the human brain: storing information with the capability of easy access at a later time using certain cues (in this case, a series of numbers as a code to retrieve data)

https://en.wikipedia.org/wiki/Memex#Associative_trails

It’s “a way to store information that was analogous to the mental association of the human brain.” My brain does not summon information by analyzing frequency of adjacent words. Nor does yours.

And whether we should dream of a magic machine to do this, a memex. Bush creates a narrative of a person interacting with the machine he imagined. There have been a long history of people trying to build a simulator of it, one I enjoyed is this animated version created in 1995 (with some nostaligia as the captions says it was build in Macromedia Director, those where the days… maybe)

Yet, I do not think of the use, creation, making of associative trails as a machined process, maybe one which machines can play a part.

On a regular basis, I hear something, read something and it immediately triggers a vague memory of ane experience, a reading, a video, an old show, a song, a presentation, something that has come across my senses (maybe that camera strapped on to man in the the As We May Think article is more like a third eye). I then grab for my search tools, using some very key self references, some organized by me, others organized by others — this blog, my flickr photos, my tagged bookmarks in Pinboard, email history, browser history, web search, Wikipedia, The Internet Archive… it’s all quite scattered, but somewhere in my brain are connections, fragments, senses. All these things make up what I think of as my memex, not one magic machine, but many resources loosely tied together when needed.

The web, the open hyperlinked web plays a big part, but its not the whole memex echinlada.

This is my way of thinking, all the time, by creating/following, connections, not relying on structured databases of facts, nor following the route of statistically generated probabilities of information.

One of my favorite tech writer/thinkers for a long time, Clive Thompson, has this well covered in an archive post on his old Collision Detection blog from 2011– “How did you find my site?” and Vannevar Bush’s memex. I vaguely recall reading it, but only found it again searching the web (that thing that is so broken) for Vannevar Bush references.

Ironically, I went to bookmark the article in my Pinboard and head slap! I can see I had done so January 2012.

But let his words do the talking!

while I love finding cool things online, it’s often incredibly hard to reconstruct precisely how I stumbled across them. This is why I’ve always thought the corporate name for StumbleUpon is so brilliant: The process of finding something really does feel like stumbling. You careen about digressively, following intuitions, doing a couple of searches, getting distracted (destructively and productively), when suddenly — wham — you discover you’re reading something that is zomg awesome. You follow a link in page A that mentions post B that ports to tweet C that is from person D who posted on their personal site about E.

But later on, it’s damn hard to recall precisely how A led to E. You could look at your web history, but it’s an imprecise tool. If you happened to have a lot of tabs open and were multitasking — checking a bit of web mail, poking around intermittently on Wikipedia — then the chronological structure of a web “history” doesn’t work. That’s because there’ll be lots of noise: You’ll also have visited sites G, M, R, L, and Y while doing your A to E march, and those will get inserted inside the chronology. (Your history will look like A-G-B-M-R-C-D-L-Y-E.) Worse, often it’s not until days or weeks after I’ve found a site that I’ll wonder precisely how I found it … at which point the forensic trail in a web history is awfully old, if not deleted.

But hey: Why does this matter? Apart from pecuniary interest, why would anyone care about the process by which you found a cool site?

Because there’s a ton of interesting cognitive value in knowing the pathway.

http://www.collisiondetection.net/mt/archives/2011/09/how_did_you_fin.php

The most interesting ways which my associative trails play out are typically following a most un probabilistic path, or relating things which on the surface are several standard deviations outward form the mean.

One Small Associative Trail

I am not sure if this will make my point (if there is one) any clearer. But a few clicks back maybe two weeks ago I spotted in Mastodon a post by Brian Bennett, an ed tech person I got to know maybe back in the old DS106 days. Brian is currently a high school chemistry teacher who blogs and posts much about the ideas he has for teaching chemistry. When I read this Brian’s post referening dimensional analysis, the trails kicked in.

Almost before I could realize it, my memories kicked into my memorable 10th grade chemistry teacher, the quixotic Blooma Friedman. We were learning all that stuff about moles and Avogrado’s number and we. had to convert things from one bizarre unit of measurement to another. She told us this method we would use called dimensional analysis would be something we would use a long time and for tasks way beyond Chemistry. And the basic premise was multiplying numbers several times by fractional something over somegthing that was really just 1.

That might be a terrible explanation.

But the association was almost instantaneous. And the fuzzy tingle was remembering that I recall posting an example of it in action as a flickr photo. In fact, from a quick search of my own photos (I quickly search of my own photos with the power gift of time trick, typing in my browser bar “f dimensional analysis”)

I actually had two examples! I forget why I chose the second one.

Look an image of an associative trails point. But then the example that I was working out? The caption brings it across the improbability space:

Dimensional analysis, thank you Blooma Friedman… calculating an amount mentioned in Vannevar Bush’s “As We May Think” –if someone puts 5000 pages of information a day in the memes how much storage is used if done for “hundreds” of years.

It’s an associative trail related to Vannevar Bush referencing my own blog post from 2014 where I was re-reading As We May Think. The post it self winds and wends in many directions.

As We May Nugget

But the relevant part, and the /q

reason for the photo of dimensional analysis was Bush speculating the data storage for a regular memex-er:

Yet if the user inserted 5000 pages of material a day it would take him hundreds of years to fill the repository, so he can be profligate and enter material freely.

The nuggety of curiosity was wondering how much storage this would be, some back of the envelope stuff I did on real paper, with some thought out logic, and putting to use the lesson my 10th grade chemistry teacher taught me.

I got an estimate of 1.35 TB of data for Bush’s example of regular daily associative trail work, that’s like a portable hard drive’s worth.

It was this un predictable route of trails, from Brian’s post ricocheting from my high school chemistry teacher and jumping to As We May Think that actually started the thinking for this post. This trail is so far from the probabilistic trajectory of auto completion on steroids. How would any probabilistic system go from dimensional analysis as taught in high school chemistry through my whacky teacher to As We May Think?

In my 2014 post I had pondered the very meaning of Bush’s title:

Lastly, I have always been curious about the title — “As We May Think”. I do not see where he really even hints at what that means. If you fly by it fast, you get a sense that he is writing about how we think, but that would make the title “As We Think”.

Is “May” hedging the bets? leaving room for error?

Or, is it how I heard Gardner mention this week, part of that phrase, like “it may not be as easy/bad/ as we may think”?

https://cogdogblog.com/2014/06/as-we-may-nugget/

What was Bush thinking about thinking? I am smiling and wondering myself at the animated GIF I created for that post with a quote from Bush, and I am guessing some clips from a video of him talking, superimposed on maybe a different video of someone playing a keyoard (?). Regardless this quote brings it all home:

As We May Not Think

It’s the process of navigating, constructing, re-constructing those pathways that is thinking, that has “interesting cognitive value.” I have no machines of measurement, but my body energy is electric when I am in the pathway making mode.

And this is the rub I am having with the embrace of AI. If it just hands you the answer, where is the pathway making, the actions that actually reinfonce the associative trails?

That quote from Bush above suggests he was suggesting the possibility of a brain like machine? It’s even part of the image for As We May Think I used for this post, found in Wikipedia Commons.

Was the point of As We May Think really “A Top U.S. Scientist forsees a possible future world in which man-mad machines will start to think”?

You only get a speck of this in the second to last paragraph:

In the outside world, all forms of intelligence, whether of sound or sight, have been reduced to the form of varying currents in an electric circuit in order that they may be transmitted. Inside the human frame exactly the same sort of process occurs. Must we always transform to mechanical movements in order to proceed from one electrical phenomenon to another? It is a suggestive thought, but it hardly warrants prediction without losing touch with reality and immediateness.

Vannevar Bush, As We May Think

He makes the comparison loosely between digitization of information for being able to share/send it. What stands out here is hos worry of “losing touch with reality and immediateness” — rather eerily foreshadowing the every growing hype of AI.

And he was “thinking” too much more broadly than machines that think, and remember the framing of his ideas having played a scientific military role during World War II and on the Manhattan Project.

Presumably man’s spirit should be elevated if he can better review his shady past and analyze more completely and objectively his present problems. He has built a civilization so complex that he needs to mechanize his record more fully if he is to push his experiment to its logical conclusion and not merely become bogged down part way there by overtaxing his limited memory. His excursion may be more enjoyable if he can reacquire the privilege of forgetting the manifold things he does not need to have immediately at hand, with some assurance that he can find them again if they prove important.

The applications of science have built man a well-supplied house, and are teaching him to live healthily therein. They have enabled him to throw masses of people against another with cruel weapons. They may yet allow him truly to encompass the great record and to grow in the wisdom of race experience. He may perish in conflict before he learns to wield that record for his true good.

Vannevar Bush, end paragraph of As We May Think

The machines he envisioned were not ones that thought or displayed some kind of “intelligence” but ones that could augment (not do) the thinking of humans– where such you see the direct associative trail to Doug Engelbart.

As I may think is by an un-statisitical improbable array of far flung bits of information but ones I relate through my lived experiences. I’ll be damned if my thinking is hemmed in by probability and frequency of word associations. I will think and I will link, and string together my trails to share in these meandering blog posts.

---

Featured Image: My own remix of The Memex (3002477109).jpg public domain image from Wikimedia Commons with additions of the OPenAI logo and a screenshot of a session with ChatGPT.