View this issue in Smore.

---

In this week’s issue: industry news, dangerous advice, and coopetition

Hello The Full SLAAIT readers!

Yesterday’s zoom meeting was lively and featured some excellent conversation. Notably, our own Kim Silk offered commentary on the parallels between the early internet and our current AI landscape, noting that the voices of the private sector are very different than those from academia and public administration. Don Means said that “public AI is a complementary idea. Why should we rely on Google [and Microsoft, etc.] to provide infrastructure for the common well being? Their base is profit-based enterprise. Will society accept only private enterprise based AI? Can there be a public AI infrastructure analogous to public roads, parks and libraries?” We don’t have the answer to this, of course, but finding it will be crucially important. Indeed, Don gave us food for thought on why librarians are more important than ever in this dawning age of AI technology, as we are beholden to the public as opposed to shareholders or profits.

Session 101: Public Option AI?! will occur tomorrow morning, hosted by Gigabit Libraries Network. Register here.

The difference between retrieving information and retrieving the right information (something in which librarians excel) is becoming increasingly relevant, with accuracy and safety now being affected by AI retrieval. Read more below.

Real Voices, Episode 1: Interview with K-12 educator Cindy Hlywa

The first episode in a planned series of interviews regarding AI’s impact is viewable here. Thanks, Riley and Cindy!

AI news

This week, there were multiple news stories relating to AI.

1. First up, Google began offering AI overviews for search results earlier this month. Rather than pointing users to a list of potential resources to help answer their questions, AI Overview attempts to provide an easily consumable summary for the user. Sounds convenient, right?

If you’ve spent time on social media lately, you’ll know that the rollout of Google’s AI Overview isn’t going quite as planned. Users are reporting that the feature is giving nonsensical, potentially harmful answers to many of their questions. Stressed and need to unwind? Google AI Overview suggests a relaxing bath with a toaster! Cheese not sticking to your pizza? Try adding glue! This article from Search Engine Land highlights some of the most absurd hallucinations made by AI Overview, as well as media coverage of this obviously flawed new feature.

Google’s AI Overview is an excellent example of how tech companies are pushing out AI products and features that are at the very least flawed, and at most, nowhere near ready for general use (see Marques Brownlee’s review of the Rabbit R1 AI pin). These companies are racing to maximize profits by capturing market share, hence the rapid release of technology that just isn’t ready yet. Not only is this frustrating for consumers that have to deal with nonfunctional features, it also hints at potentially more dangerous issues happening out of sight. While so much attention has been focused on generative AI, it’s important to remember that artificial intelligence is more than just generative tech. AI is being used to analyze large datasets, and to make decisions based on its analysis. What happens when this behind-the-scenes AI technology hallucinates when making decisions about rent prices, medical care, or employment decisions? It’s much less obvious to the public when AI makes a mistake in these use cases, but the potential for harm is high.

2. Also notable and newsworthy, Microsoft announced a feature for Windows 11: Recall. Alarming to privacy activists, this feature will take “screenshots of everything happening on the computer every few seconds,” ostensibly to allow users to review previously viewed/accessed/created information.

3. In other Microsoft news, CoPilot will be integrating AI features into multiple aspects of its user experience, with messaging services allowing chatbot capabilities. It is important to note that generative AI is not the only game in town, though it’s the interesting and appealing to us right now. AI assistants and seamless integration like these features are making moves into users’ daily lives.

AI librarians and educational programs have arrived…

Rollins College in Florida has now posted a job opening for a tenure-track AI Librarian position.

The University of Texas at Austin is now offering a master’s degree in artificial intelligence.

David’s corner

Is 2024 1997?

There have been those who have compared AI in 2024 to the web in 1997 (me included). The argument goes that at the start of 1997 no one really knew much about the web, but by the end of that year it was clear that business and government were seeing the web as a new fundamental thing (information source, marketing venue, media outlet). So, by the end of this year AI will be seen not as a curiosity, but as a new reality.

While this is a theme I touched on last week with the hype cycle (maybe not so much a new reality for generative IA?), and something we’ll no doubt return to, there is one aspect of AI in 2024 that is absolutely different than the web in 1997. In 1997 anyone with an internet connection and a modest computer could launch their own website. If you didn’t have a persistent internet connection, you could create your own site hosted by services like Geocities. In 2024, if you want to create your own Large Language Model like ChatGPT you need millions of dollars in infrastructure (computing centers, bandwidth, servers, and programmers). You’ll also need access to a lot of electricity.

Now, sure you can create your own chatbots built on top of things like ChatGPT, but you can’t make your own trained generative AI system from scratch just by knowing the equivalent of HTML. This means while it seems like “everyone” is doing AI, if you look behind the scenes, you’ll find a very small number of players (like OpenAI, Google, Meta, China). Even Microsoft is building their work on OpenAI.

I want to be clear, here we are talking about Large Language Models and generative AI. AI systems built on other types of data using machine learning and embedded in things like smart watches and cars? They normally require some hefty computing power to train up a model, and then be deployed on modest computing platforms.

Of course, that’s another difference between 1997 and 2024 – the cloud. Instead of building your own massive data center you can now create virtual machines and run computing intensive applications on services available on the web. Cloud services like Amazon’s Amazon Web Services, and Microsoft’s Azure provide for fee per application use of someone else’s computing investment. Many people don’t know that services like Netflix don’t own a bunch of machines, they use Amazon. In fact, the SLAAIT Petting Zoo is being built using Azure Lab Services. We can stand up a full computing lab for 40 cents per hour per machine. Need a lab of 100? 1,000? Just add more virtual machines.

All of this leads to a paradox for libraries: “coopetition.” if we don’t build it (chatbot, search engine, lab) ourselves, we are dependent on other organizations that may not share our principles. However, libraries can’t afford to build our own systems (the valuation of Amazon Web Services is estimated at 1.5 to 2 trillion dollars).

Why does this matter? The bottom line is that state libraries, and indeed the whole of the library sector, simply can’t go it alone. SLAAIT members should look to projects like the National Science Foundation’s plans for National Artificial Intelligence Research Resource (NAIRR) pilot. Likewise, for building more modest library-oriented AI capability, partnership will be key. Partnerships within states, and across them.

To be clear, these shared partnerships will most likely be not for the very visible generative AI service, but for application driven AI systems such as item tagging, expanding the metadata of archived materials, and, eventually, cataloging support (systems that use AI to help human original cataloging).

It also means, we need to focus on what librarians and library staff need to know to be good partners on these shared AI projects. Do we all need to be specialists in cloud based shared computing? Python programmers? Or is there a model, just as in the rest of librarianship, where there are core shared competencies and areas of specialization?

That’s where we pick up next week: library competencies for AI.

AI funnies

In searching for “AI political cartoon” to finish out this newsletter, I came across a site that lets you use AI to create your own! Have some fun and feel free submit responses to the Google Group!