Samhälle
Kultur
OL Gräver
Underhållning
På Campus

On a Quest for Where and When

Matteo Aquila - Authormatteo.aquila@osqledaren.se

Maria Zelenika - Illustratormaria.zelenika@osqledaren.se

How do you know what time it is? How do you define where you are? Sync your clocks and get your compass, as we explore how difficult it is to communicate information, how AI is helping us solve this problem, and how harder everything becomes when there is a language barrier between you and your search engine.

When I enter a library or a bookshop, one of two opposite scenarios unfolds. Number one: I know exactly what I am looking for, will go straight to the correct shelf and will be out of the building in ten minutes. Or number two: I will get in without knowing how and when and if I will get out. It may be with a pile of ten books or with nothing at all. Sometimes I am a man on a mission and sometimes a wanderer with no care in the world. Sometimes I launch myself in explorations of a random topic and sometimes I just don’t find a book that speaks to me. In KTH Library, sometimes I don’t even need to go through the main door.

It happens, at unpredictable times, that the library decides to free itself from the burden of some books and leave them in a cart at the entrance. From there, anyone can take them and bring them home. “Varsågod”, says the sign; and sometimes I “tack” it indeed. That was the case a few months ago, when among the brittle encyclopaedias on the cart I found a book called Ambient Findability, written by Peter Morville. I guess it was fate that led me there. As the book’s subtitle says: “What we find changes what we become”. And so, from that book, here comes this article and my new hyperfixation on information architecture.

Peter Morville is the first information architect (because he invented the title). He works as a consultant to many of the biggest companies in the world to help them achieve one goal: make the information they produce as “findable” as possible. Peter didn’t properly coin this adjective himself, but he sure gave it the meaning it is most used for these days. In Ambient Findability, he defined findability for the first time as “the ability of users to identify an appropriate Web site and navigate the pages of the site to discover and retrieve relevant information resources.” This was 2005, the Internet was still a fairly new creature and many of today’s most impactful platforms were just newly born. Facebook was around one year old and only accessible to college students, YouTube was born in February, and Twitter was yet to come within another year or so. The amount of content available on the net was negligible compared to what is continuously generated nowadays. And yet, Peter had a vision: there will soon come a time when the information in the palm of your hand will be too abundant to be navigated through without a guide. Blogs and web pages were there already, and already so hard to find and distinguish and appreciate. As he held his primitive smartphone on a beach, he recounts, Peter envisioned a world where everyone could upload their own picture of a sunset on the sea in the blink of an eye. Peter knows that a way to organise this chaos is to be found, and fast, before it gets too impractical. He knows because he graduated in Library and Information Science and his studies tell him that this is a problem which humanity has solved already, many times.

***

This is not the article I originally wanted to write a few weeks ago when the OL editorial board agreed on the theme of “chaos”. The idea for an article on “findability” had wriggled into my brain already, because I had just finished reading Peter’s book. Still, I wanted to write about a different topic: time. No better way to discuss chaos than to talk about the story of how humankind decided to wrestle time into a legal framework and won. I was interested in investigating civil time, which is any time standard designated by civilian authorities: e.g., the time zone you live in, which in Stockholm’s case is UTC+1 (or UTC+2 during the summer). UTC (or Coordinated Universal Time) is the successor of GMT (or Greenwich Mean Time), which became obsolete with the introduction of TAI (or International Atomic Time), on which UTC is now based. I hope you’ll forgive the information dump. It is a very cool, if complicated, system – held with a surprisingly democratic process. TAI is determined as the weighted average of the time kept by over 450 atomic clocks in over 80 laboratories worldwide. It is then nationally implemented by State-backed institutions that spread a master time signal around their country. For Sweden, that would be RISE, keeping time with five atomic clocks in Borås.

While eagerly researching all of this, I found an unexpected historical fact: according to a blog post written for English-speaking students of the Swedish language, Sweden was the first country in the world to determine a nationwide civil time by law. I decided to look more into that but soon discovered how hard that would be. The blog post didn’t do me any favour, linking zero sources to its claim. I began my research process the classic way: a Google search. Imagine my confusion when I found that no other result came up other than the original blog post. I tried changing search terms – nothing. I tried searching English Wikipedia directly – nada. I even gave a shot to another search engine, DuckDuckGo – zero, nisba, niente. I don’t consider myself a beginner in researching, as a student for almost 20 years and a professional curious person. How could it be explained, then, that nowhere on the Internet I could find a source for a very basic historical claim? Could it be that maybe my searching skills were not in the wrong (I'd say so!), but rather the issue lies in the way knowledge is inherently built?

***

We live in a 3D world, which means that, from the second we are born, we need to learn how to navigate space. By 18 months, babies have usually developed basic spatial awareness, achieved by very slowly experiencing the environment around them. In time, they become masters of looking for stuff, moving following directions, and even, with no directions at all, forging new paths according to their goals. The ability to have complete mastery of space and move through it actively is often called “wayfinding”. It is an idea that is surely familiar to architects and urban planners: street signs, prominent landmarks, and compasses are all wayfinding tools that help us get on the right track. And of course, the most important of all wayfinding tools is the map – a visual representation of space that, combined with other tools (representing street names, prominent landmarks, and compass orientation…), gives us the best possible knowledge of space around us. Wayfinding has been a problem to be solved forever, both in the natural and artificial worlds we created. The concept itself is so powerful that it has been transposed to different fields and mediums: we can talk about wayfinding in virtual worlds such as video games, we use wayfinding metaphors in our everyday conversations, and it is a central topic of, you guessed it, information studies.

As a 1998 paper called “Metaphors We Surf The Web By” testifies, spatial metaphors very quickly became the main way to talk about navigating the Internet; the digital world was built as a reflection of the physical one. In some cases, there were even attempts at building web maps by creating a geographical view of it, and positioning pages on fictional islands to represent different websites and the crosslinks between them.

The company Antarcti.ca Systems (its product, shown in the picture) tried to make a business out of geographical representations of the Web and raised almost $4 million in 2002.

This approach, as beautiful as it is, has many failure points – first, a website is not an island. Geography was born to represent distances: knowing how far a place is from another is fundamental to planning how much time and resources you will need for your trip. The Internet, instead, was built exactly to fix this problem, to allow for quick connections between far distances. And if you thought that these experiments I was talking about were using physical distance as a metaphor for how hard it is to get to that piece of information (therefore applying the concept of findability), I am sorry to break it for you: they weren’t. Those nice maps were just for aesthetics and, therefore, useless. But the need for directions is real: the abundance of hyperlinks and its opposite condition (orphan pages that are not linked to anything) call for an orientation system. The problem, however, already presented itself once with the ancestor of Internet links: bibliographic citations. The solution to this problem can be found, as so often happens, in libraries.

Librarians have been dealing with information management for literal millennia: we have testimony of catalogues back in the Great Library of Alexandria, the biggest library of the ancient world, or in the even older library of Nineveh, with clay tablets provided with crude bibliographies. Different systems for classifying books have come and gone over the centuries; in library science, as in all fields, tools for wayfinding adapt based on the current needs. In the modern era, it was standard for libraries to have thousands of cards in drawers, one for each book, containing information like the title, the author, the year, and brief keywords regarding the content of the book. These small pieces of information are called “metadata”. Metadata on card catalogues gives an expert librarian all the important characteristics of a book at a glance. When a specific topic was requested, the librarian would go to the catalogue’s section corresponding to the topic, they would skim through all the cards, and with a mix of luck and experience would find a reference to some potentially useful material. It was a necessary tool to find useful sources among hundreds of shelves. When computers came, card catalogues evolved into digital solutions, and metadata changed their structure as well. The system worked great, so much so that, under heavy adjustments but keeping the same philosophy, when the Internet’s librarians arrived on the stage, they quite quickly understood that this was an idea they could steal.

***

My Internet librarians were proving themselves useless. Search engines, that is. I spent so much time throwing different keywords in the search bar and got no significant results. I was stuck and so was Google. I had to surrender to the fact that the issue here was language: keeping the hypothesis that it’s impossible for anything not to be on the Internet, I had to conclude that sources for this niche fact would exist in the Swedish language only. As I don’t know Swedish that well, this was a hard obstacle in my path. Not insurmountable, as I could have tried any online translator. But one also must consider the intricacies of language and its effect on web searching – not all keywords are the same. Traditional search engines are smart, but not that much: they look for the keyword you provide them with and will return specific answers to specific queries. If you lead them astray with the wrong keyword, that’s on you, the searcher. Pareto’s Law, familiar to all students, also applies to information studies: 20% of the dictionary will account for 80% of written language. Therefore, if your keyword is not precise enough, the engine will fail to return what you asked for and just spew generic and uncorrelated information. If you translate a keyword from one language to another, without knowing all the subtleties of said language, your search will most likely be inconclusive.

To prove my point that Internet findability is still an issue, I decided to keep this language limitation for the time being: no translation tools allowed. Then I started to think as a librarian: if I can’t access the direct keywords, I’ll try to broaden my research field and get back to my specific topic through a chain of bibliographic citations. I looked for “Sweden, time, law” and similar keywords on the KTH Library catalogue and Google Scholar. I found three books that gave me a strong starting point: The Global Transformation of Time: 1870–1950 by Vanessa Ogle, Greenwich Time and the Longitude by Derek Howse, and The Clocks Are Telling Lies by Scott Alan Johnston. These (very interesting) sources gave me the jumpstart I was looking for: a comprehensive history of how civil time came to be worldwide, not just in Sweden. All three of these books spoke about Sweden at least once, referring to it as a world leader in setting a law for a national time standard. But, unnervingly, none of them gave sources for those claims! How Sweden did that and when it happened were still details unknown to me. And I wouldn’t stop until I got them.

***

I know what all readers are thinking by now: why am I purposefully ignoring the easiest way to get to the bottom of this? Why didn’t I ask ChatGPT about it? I did. Did it solve the mystery? I will reveal that in due time. Before we get to that, allow me to briefly tell you why I think Large Language Models (LLMs) like GPT-4 are the reason why KTH Library decided to throw away that book that I took home.

While reading Ambient Findability, one sentence struck me as odd: “Despite the hype surrounding artificial intelligence, […] computers aren’t even close to extracting or understanding or visually representing meaning.” Remember that this is a book from 2005, 19 years ago. “For as long as humans use language to communicate, information retrieval will remain a messy, imperfect business,” Peter Morville stated while looking at the then state of the art in AI. We now know that this sentence is wrong; or at least, we are closer than ever in the history of humanity to a situation where this sentence can be considered wrong. Morville said what he said because he dedicates a whole chapter of his book to proving that context is fundamental to interpreting written information, that writing metadata needs thought and experience, and that computers were not (and never will be) able to properly do that by themselves. At the time, he was totally right. Search engines in the early 2000s relied on manually created metadata by the content’s author (who rarely were librarians) or tried to create them on their own (very generic and uneducated guesses). Machines did not get context, just clues, single words. Nowadays, instead, we have created software that has a notion of context: LLMs, such as ChatGPT. Generative AI defines context in mathematical, stochastic terms; we could almost say that Gen AI understands only context and ignores the single words it is using to build a sentence that, in its calculations, “makes sense”. This makes Morville’s case arguably false. Computers can now extract, understand, and visually represent meaning.

Half of Morville’s book is built on this assumption of computers’ inability – which makes it obsolete. This would explain why a technical library such as that of KTH would want to get rid of it, at least in its physical form (if you’re interested, the e-book is available to borrow). However, LLMs are not perfect, and many problems still need to be solved, through the concepts and past solutions that Peter Morville so eloquently describes. So, don’t worry, he isn’t going out of business.

The two biggest issues with LLMs in searching are probably relevance and reference management. Relevant results are those that are interesting or useful to the user. Since relevance, like beauty, is in the eye of the beholder, traditional search engines very often fail to grasp it. LLMs come closer to getting it right, but still heavily rely on the ability of the user to declare their interests through a precise prompt. Prompt engineering is a common buzzword among AI folks, and this is the reason why. When talking about reference management, we mean that LLMs need to be able to give sources for the claims they make. As all students know ChatGPT is terrible at this: Gen AI often suffers from hallucinations, for the exact reason that it doesn’t care for a sentence to be true– just that it makes sense from a mathematical point of view.

When I asked various LLMs my question on civil time in Sweden, not all of them were able to answer. Only ChatGPT 4 and Copilot by Microsoft (built on GPT-4) gave answers that somewhat made sense. What they really failed to do was give me proper sources, and I had to keep looking for those on my own. However, AI helped by giving me the most complete telling of this story I could find so far. By investigating some of the names it gave me, I was able to track down The Production and Distribution of Synchronized Time in Sweden, 1850–1914, an article in English written by Gustav Holmberg and Johan Kärnfelt, from the University of Gothenburg. And through that, I got to their Swedish-written book from 2019: Tid för enhetlighet: Astronomerna och standardiseringen av tid i Sverige. Guess what? Full of references to historical documents, many of which are consultable in Stockholm, this is exactly what I was looking for.

***

The first form of shared time measurement in society was, of course, the Sun: sundials gave a good point of reference, providing the same results to all in a local setting. Then, mechanical clocks came along– first, there were big tower clocks that regulated schedules for entire towns, but once it became easy to mass produce portable watches in the 19th century, time measurement was finally democratised. At the same time, technology allowed to move for long distances in very short periods and a new problem had to be solved: how to synchronise all those portable watches, and how to agree on which synchronisation source to use, as local syncing was achieved still by measuring the Sun, and sundials give different results as you move through different meridians? Each city and activity chose the time setting that was most convenient to them– you could say that this is an issue of relevance. Train companies were the first to raise this concern, in Sweden as well as all over the Western world (for the sake of conciseness, I have not covered time measurement in other geographical areas in this article.) As Gothenburg and Stockholm had a 24-minute time difference when measured astronomically, clocks in train stations signalled two different times; time schedules got very messy because of that. Various solutions were proposed and around 1865 a temporary standardised time for trains was fixed, taking as a reference a location halfway between Gothenburg and Stockholm.

The first law that indicated a Swedish Mean Time valid for all uses and territories was signed in 1878 and came into effect on 1st January 1879. The time signal originated and was governed from Stockholm Observatory, near Odenplan and behind Stadsbiblioteket, and then transmitted around the city, with the use of time balls, and all over Sweden, through telegraph lines. At the Observatory, time was measured through the position of the Sun, and then the result was actually expressed in relative terms, comparing the Swedish time zone with the Greenwich one. And this was before Greenwich Mean Time was even a standard! The United Kingdom itself had not yet written into any law that GMT was to be national civil time. But in 1884, 26 countries from every continent met for the International Meridian Conference in Washington, D.C. and decided on the Greenwich Meridian being a worldwide shared standard for geographical and time reference. Sweden was one of the most vocal countries pushing for Greenwich’s new role. 16 years later, in 1900, Stockholm Observatory stopped measuring the Sun's position for timekeeping and officially aligned Swedish Mean Time with GMT, moving all clocks back a handful of seconds in the process.

So, over 3200 words and a few days later, I have to finally concede that that blog post was right: Sweden was the first country in the world to codify a national civil time into law. What I expected to be just a brief research, proved to be quite the adventure. The reason for that is that information is extremely complicated to manage, diffuse and discover, even in the 21st century, in the age of artificial intelligence. The existence alone of a piece of information doesn’t imply that it will be readily “findable”: it needs to be packaged in the proper way, an operation that needs skill and appropriate knowledge. Often, that isn’t enough either if the tools aren’t up to the task.

At the end of this all, tired but satisfied, I finally decided to break the rule I had given to myself and ask my Swedish friends for help. “Hej, – I said on the group chat – how long would it take you to find out if Sweden truly was the first country with civil time?” Two minutes, apparently. And so this is the story of how I started studying on Duolingo again.

Publicerad: 2024-05-24

Ansvarig utgivare: Raquel Frescia
© 2008 - 2024 Osqledaren.