This is a recorded session from the 2024 AAM Annual Meeting & MuseumExpo. As museum professionals, understanding the basics of AI and its implications to our industry is crucial. If the museum community can approach AI with a balanced perspective, harnessing its potential while being mindful of its implications, ethical concerns, and informational biases, we have the opportunity to revolutionize how we work and foster innovation. This presentation provides an overview of AI and explores the various opportunities, challenges, and serious concerns that we must face together.

Additional resources:

Algorithms & Artifacts: Deciphering AI’s Role in Museums slides

Transcript

Jack Ludden: I thought it was going to be a bit lower, so it’s fantastic that we’re all diving in and getting involved. Okay, so for better for worse, what you’re going to find out over the next hour or so is you’re going to get inside our heads.

That’s what’s going to happen. Each of us are going to be up here for about 10 minutes or so to walk through and talk about how AI is packing the work that we’re doing. So that’s what’s going to happen. Shout out to CFM, AAM, if Beth is here, I want to give a shout out to Beth, the cone of possibilities, right? I’m not going to go into depth of it. Again, QR code, you can learn more about it. But the gist of it is we don’t know what’s going to happen yet with AI. And specifically, we don’t know how it’s going to impact all of our lives in the museum community. So, we have a lot of control over what that future cone is going to look like, how we can impact it. It’s a great concept to think about as you’re starting to investigate and use AI.

Another AAM resource. A lot of us quote this, a lot of times. Again, it’s helpful to think, when we’re thinking about AI the concerns and issues we all have. We are respected and we are known for providing  providing truth and providing trust to the community. That’s a powerful position to be in, again, as we’re thinking about artificial intelligence.

Perhaps you’ve seen the adoption curve before.

You don’t have to shout it out. All you have to do is have an inside voice and say, well, where am I on this curve, right? Because many of us might be innovators, some of us might be early adopters, but a majority of us, right, are going to be early majority, late majority, or even laggards, right? Where are you now? And as an industry, where might we be? Something to sort of keep in mind.

I love this point, right? It took three and a half years for Netflix to get a million people on the platform. Chat-GBT, it took five days. Right. Pretty remarkable

Again, QR code to sort of walk through that.

Ok, can’t become a classroom too much today. But again really useful to have a basic understanding of some of the key elements.

machine learning.

Machine learning is a method we use to teach computers how to learn patterns from data. Generative AI is a type of artificial intelligence that creates new content based on what it has learned. Very few people even say the full large language models.

You’ll hear LLM really quickly. You’ve probably heard that term, right? It’s a large language model is a type of artificial intelligence that can understand and generate human language. And hallucinations, which is one of my favorite terms in AI, it generates information that simply isn’t true or doesn’t make sense. And I’m hoping you all have seen that in some capacity. And if you didn’t know as a hallucination, that’s what it’s called.

This is an opinion. You can take it or leave it. But I see two things that we need as an industry.

One is some level of governance. Now you may pause when you’re like, wait a minute. Did he just post a link to the intelligence community governance about AI? I did. And whether you read it specifically and in detail, that’s fine if that’s not something you want to do. But looking at how it’s organized and how the sections that it’s created to help articulate what governance is required, it’s really useful. So as a framework, it’s really useful to have that.

Data management, if you do not remember anything regarding artificial intelligence, knowing how you’re managing your information is critical, right? You’ve got to collect it properly, you’ve got to manage it properly, you’ve got to be able to access it in a meaningful way, right? Good data going in helps provide good data going out.

One of the things that I have found is using journey maps to sort of help brainstorm challenges that you may have right. Thinking about that audience you’re thinking about your visitor we’re all thinking about it for internally from our various perspectives well where is AI possibly helping support the work we’re doing right. So, this is that a helpful way to kind of connect the dots between what’s that visitor experience what’s the work you’re doing behind the scenes. How can AI help you?

Okay, quick. So, one of the things, what each of us are doing is we use the SWAT analysis framework, and we’re each going to talk about some element of AI and walk through that framework, right, to kind of help you give a perspective, again, get inside our own heads as to how we’re tackling all this stuff. Right? It helps provide a perspective.

Okay, so for better or for worse, I’m first.

So, I’m going to go through the SWAT analysis. Maybe we’ll quickly figure it out, but I had a little help with a computer friend on the various palms.

Okay. If you know where you are, it will help you get where you want to go. I say this a lot to people in my family, to people I work with, right? And I’ve already sort of said it to you guys about a minute ago. So, the way I love this as a notion, right? And the translation of this in AI is, again, I said it earlier, good data going in helps with good data going out. Because for me, the thing that I have been obsessed with with AI has to do with prompts. Prompts are commands or instructions that let LLMs help you generate new content. That’s what its trying to do. And, the better your prompt the better the outcome.

Right, so take something fairly straightforward, so if you have a notion this is a quiz, you can kind of see what’s going on here. Okay, wait, I want to generate a quiz, well what do I need to tell – you can’t simply write I want to write a quiz. Clearly, half the room knows that that’s not something so providing a level of detail is really powerful. So think of it as a template. Prompts are a really powerful template.

This is a weakness. This is the second thing, if you find yourself using AI and you’re simply copy and pasting it that is, something is wrong. There’s a loss in translation. That human connection should not be missed. That to my mind is a weakness and quite frankly with AI in general.

But man, the opportunity right, responsible prompting, right, this notion of thinking of where you are, where you want to go it gets you farther, faster. What ever type of department and expertise your in in the museum community there are elements you can work with to allow you to do it farther and faster. This is the quote right if you’re five minutes early you are already 10 minutes late I know it I hate it to say it but I’m gonna say it out loud I got a lot of people to hear it right if you’re not using it start using it today you gotta try using and and I mean that in AI but I also mean it with prompting if that’s if a prompt feels uncomfortable to you there again I I posted a few links there, you’re going to find great resources to help you provide a level of clarity and understanding. Those templates can be useful.

It’s again, I’m second, I’m going to, I told my impact is the opportunity. You gotta do, if you do this, the opportunity and the impact that you’re on your work and the work at your organization is going to be remarkable.

So, I already said it. But the fact of matter is, is the recommendation I have regarding prompts and what came out of my SWAT analysis is, what you really want to do is make sure you’re using this stuff, I try to use it every day. That sounds maybe a little outside the norm. And it’s not so much wanting to attach it to directly to the work I’m doing, but just to feel comfortable in understanding how it is as a tool.

Because the tool is changing regularly. It’s updating constantly, right? That adoption curve is changing pretty remarkably. So the more you can be in it and feel a bit more comfortable with it, the better you’re gonna be. Okay, I am now gonna step away from the mic. I get to introduce Jonathan, who by the way, is a brand new father. So Jonathan, thank you. Congratulations.

Jonathan Munar: Thank you very Thank you and yeah if thank you for the out if some of my words get a little cloudy they just jump on a train here this morning Jack I thought you wrote this so I thought you were very brilliant with words here but very lovely to be here today my name is Jonathan Munar I am on the arts team at Bloomberg Philanthropies based at in New York though I spent my entire career in and out of the art world of course the last two decades working both at a very large museum and a small non -profit also based at in New York. So, where I’m going to focus on today is um ai as an accelerator right so the point of view here for me humans of course are very much a part of AI from the beginning to the end right um we really do have to see ourselves as the ones who create these systems, to train these systems. These systems are coming off of everything that is inherently human, right? So always to keep that in perspective, what we like to think about here is where can AI start to really serve us and help us become and continue to be a little bit more human.

So I’m going to focus on accelerators and humanity just as a whole as part of my SWAT analysis. So a quick thing what do we mean by accelerator accelerator basically anything that’s you know is designed to help make our workflows a little bit more efficient right automation is typically something that’s involved there but it’s really thinking about the things that kind of take some steps away that you can then therefore get to your goal at a faster pace right so of course our entire civilization is built off of different types of accelerators ways to make things easier and I point to a very primitive example here of a ball floating ball valve the thing that you find in most common toilet bowls right something that detects a water level that it knows how to refill the water basin and stop refilling a water basin I don’t think anyone in this room may have lived a life where you used to manually have to put water into a bowl but this is a kind of a nice kind of, a very primitive type of technology that points to the idea of let’s figure out ways to take out some steps to get to the thing that we want to do, right?

So we’ll see this a lot, we’ve seen this a lot even, you know, in the early days of technology, OCR, being able to, and maybe perhaps a lot of people in this room have dealt with scanning great amounts of text to be able to just have something workable in a system and scannable and searchable and a system. So OCR, great, great accelerator. Speech recognition, of course, you know, foundation to a lot of the things that we do with AI or just assistance today. Google Translate, you know, where, how could we not, how could we communicate with folks in many different languages were it not for something like Google Translate? And even in the early days of internet things, right, generating a really cool Pandora Station off of, this is not my Pandora station, but off a death cab or some kind of death cab adjacent programming, TiVo, filling up your DVR with, again, not mine, Gray’s Anatomy type programming, or even a room by just, you know, remembering your room and designing itself to be able to make your life a little bit more efficient.

And if anyone just came from Elizabeth’s TrendsWatch session to know session to think about this idea of where I want AI to focus on my cleaning and not focus on you know doing the skillful things I want to do I love the fact I never really thought of a Rumba as that but yeah thanks Rumba doing those kinds of things so some of the strengths of course I mean it probably goes without saying where the strengths are with AI you know this is this is really less time spent, you know, performing routine tasks, right? There are many of us in this room that, you know, perhaps work for very small institutions where you’re forced to wear many different hats.

I myself, you know, I came up as a web developer at the Metropolitan Museum of Art earlier in my career and I would spend probably double amount, the double amount of time to do something so that I wouldn’t have to reproduce again, double the amount of time it would take to do it manually. I would spend time to generate a program to be able to do that. I think maybe any engineers in this room might be able to relate to that type of experience. But you know, it’s the idea of being able to create something once so that when the same situations come up again, you don’t have to deal with that. I already develop a system to process a spreadsheet to put calendar events into a website. I don’t have to manually enter it a second time.

So a clear strength of an accelerator. Buy you some time to focus on things that actually, you know, require your thought and let the accelerators handle the rest. The idea of advancing the starting line, basically, you know, that there is this great possibility where you are relying on an AI to build you a foundation to then kind of build on top of that. I didn’t spend the eight hours prepping the spreadsheet work the spreadsheet was already populated or if you need to think a step further I have a structure around a spreadsheet and I have a source of data and that source of data can then send it to a spreadsheet that will help me then continue on to think about you know how do I process this information so really you know advancing your starting line and you know there is a something out of this where you know if you are programming something, if you’re automating something with great predictability, then there’s a less chance, lesser chance of human error, you know, and you’re not copying and pasting out of a wrong field or pacing into the wrong field.

Weaknesses, of course, it probably should go without saying to anyone in this room that there is nuance to being human and nuanced to being emotional, and if I do need to explain that to you, you are your west world robot get out of here um but you know there is a lot to you know especially the field where a lot of us work and you know around art and creativity where um to understand the painting an emotional response to a painting and understand a work of art you could spend many lifetimes trying to design a program to tell you that this is some type of representation of a stadium right this julie meretu painting right um so there is a lot of nuance to humanity that is not easily programmable. Clear weakness to trying to rely on these systems to help us respond to things like this. Another indication that humans are really a big part of this.

Of course when your talking about a system that getting you past a start line or moving up your start line there is a degree of vetting or reviewing that one would have to do to anything that is generated from an AI system. That of course can a take a little more time than expected when you are dealing with things that are otherwise unpredictable or flat out wrong.

And then, subjectivity itself is something that just is not very easily predictable.

So, I want to talk a little bit about context here. So, I work on a project called Bloomberg Connects. On the Bloomberg Philanthropies team and I think that what we do in this space is very much operating in the opportunity’s quadrant of our SWAT analysis. I do want to give a shout to our head of product who’s in the room here David Harding as well as our data engineer Angie Burnett who’s helped put a little bit of perspective in the types of things we are producing in Bloomberg Connects in the AI space. So, for those who do not know what Bloomberg Connects is, this is a project that was born out of the Bloomberg Philanthropies Foundation. It is really meant to see through the vision of Mike Bloomberg himself, who really does try to grant access to art and culture through technology. So, this really roots back to his own experiences, going into a museum, being led around a museum by a museum director and thinking to himself more people should have experiences like these so this is really where this Bloomberg Connect’s project was born out of it was allowing technology offering technologies a philanthropical support to institutions to institutions that otherwise don’t have the resources or the skill sets to be able to to build a system to connect their audiences to to their knowledge.

So, AR and our use of AI is really an extension of this idea of, like, building an app is one thing, but thinking about how AI can impact a field now 400 plus institutions that we work with is really important for us, right? This is a great way for us to be able to also think about where, you know, Museum A might have some ethical issue about approaching AI in one way, where as museum be maybe not so much. This gives us a really good opportunity to address all the different types of use cases across institutions and come up with solutions that work, you know, not maybe across the board, but at least work to a way that can be tuned to get us closer to where we would like to be. So really it’s us about us employing AI as a way to continue to connect audiences to the culture within your institutions.

So I’ll point to two examples that our team has been experimenting with. One is the Bloomberg Connect system. This is basically a fielded, more or less object information repository. Right. So, say for example we’ve got this great Stuart Davis painting out of the Phillips Collection, one thing that we’ve experimented with is well what can we do if we just fed this image by itself into a system, train a system to look at different sources to gather information about this object and then at least give, say, a content producer at an institution a starting point to say, okay, this is everything that we know about Stuart Davis. Maybe it’s coming from the Phillips Collections Database. Maybe it’s coming from some other vetted sources. But this gives the institution a starting point to be able to generate something like this where our content management system is already populated with information. Known information, vetted information about an object. And it’s on a human, its on the person, the staff member to be able to go through this and say yes, correct, correct, correct, correct.

We can go a step further with this  and say to an institution that is unable to produce audio content that we’ve generated this great description, perhaps its from your own scholarship again, perhaps its from other sources. But we’ve created something that works as a very serviceable audio guide that you can then choose to distribute to your audiences as well. So this puts museums in a great place to advance forward with getting their collections out there without significantly, without much significant effort.

I had videos in here, but these are basically the ideas of using an AI system to be able to scrape knowledge that’s already entered by an institution and then build out itineraries based on that.

For example, you know, this is one about a teacher trying to take a class to the Intrepid Museum. How can you structure out a tour based on things that they’ve already input in there?

So threats, irresponsible hallucinations. I don’t think I have to convince anyone about what’s threatening about AI, but like I think, And again, anyone who’s come from the trendswatch session knows that, you know, human capital is always something to consider here.

You know, and, you know, it’s really about, you know, trying to make sure that there’s always a human consideration in the work that we do with AI. The impact, of course, is that there is a great opportunity for vast amounts of knowledge to just be surfaced and exposed and out of the world, things that, you know, otherwise are sitting in binders or storage cabinets within your institutions there’s a great opportunity to have I tap into AI to be able to surface a lot of that knowledge and get it out to the world so my recommendation AI is there to help you not embrace you to replace you. Embrace AI now or get left behind I think if any institution has seen folks kind of roam through their galleries scanning objects and using chat GPT to get information about those objects.

This should be you giving that information to your visitors, not chat GPT, especially when we’re living in an environment where things aren’t always exactly right. Embrace it now. Think about how we can harness this power to then keep our humans happy, and of course continue to be able to use AI responsibly. Sorry everyone over. New dad, right? Totally fun.

Jessica Herczeg Konecny: Hi everyone, my name is Jessica Konecny, I’m the lead technical analyst for digital asset management at the Metropolitan Museum of Art. So just a little bit about me, I’ve been practicing DAM for 12 years. Who I am as a DAM practitioner is I’m in a service role. I’m here to help my colleagues do their job better, and then help guests, visitors now and the future. I’m the custodian of these assets, you know, history of the institution, history of artwork objects, and it represents a significant investment of time and money into these resources over decades and decades. So I’m here to help preserve what’s next.

And usually these systems are seen as kind of set it and forget it. And I have spent my career trying to convince management that DAM doesn’t work without people. There’s a forthcoming book, DAM for Museums, that will be published next year, edited by yours truly, and Jack and Nick are participating.

And then I just want you all to remember that a digital file isn’t a digital asset unless it has some really good quality metadata attached to it. So I’m a custodian of this material and this trusted source of information. And so currently and into the future, human oversight of AI output in this context is essential.

Metadata is my jam. Metadata is magical if you put in the work. This is a picture of my ludicrously capacious bag. And I firmly believe in the phrase, I don’t know, right? Like, I’m not afraid to tell anybody on this panel. I don’t know.

So this is kind of what I’ve been thinking about over the last eight months. AI kind of broke my brain. I’m trying to wrap my brain around all of this. Right, as a DAM practitioner. So, last fall I attended this webinar given in part by Phillip Shearer, who is the deputy university librarian at Stanford. And I wrote to him afterward and said you broke my brain because he talked about how we’re not going to, we possibly won’t need this kind of storage of descriptive metadata in the future, that you’ll have some kind of discovery mechanism that queries the digital asset directly. And so he challenged us to be careful that we’re not just replicating the way we’ve been doing things for the past 100 years.

And I’m like, but I like the way we’ve been doing things for the past 100 years. Like I’m stuck in this mode of creating records, right? Now I need records for inventory control because I need to know where that object is in place. Well, I don’t, but people need to know where that object is in place in time. But anyway, using a chat bot to interrogate the data, bypass the cataloging process, and abviating the need to store the data. Like, that sounds good, but that sounds like it might be my job or part of my job. And I freaked out, and I immediately was trying to come up with the ways that Phil was wrong, right? That was my defensive reaction. I was like, no, this can’t be right. I have a bag.

But I really hate coming at these things from a place of fear. Like I’m trying, I’m trying. Okay, so my approach, like in this era of post -truth, suspicion, distrust of institutions, like Jack mentioned, museums are still seen as being trustworthy. Whether or not we always fulfill that is another story. Just remember that cultural repatriation should include a discussion of the data.

And, like, balancing this idea with, there are certain things that we still don’t know about our objects, like retitling a piece or if the artist is in, you know, there’s some disagreement about who the artist is. Research is always being done. Anyway, I love this quote from a podcast I listen to. It’s like, get work done, avoid anything controversial. I kind of took that as my like rally and cry. Like, how can I get work done and avoid something controversial? Incidentally, I got this quote, this is me in the flames. But I asked GPT if you could find me a painting from the collection of the Met that could help me represent the concept of stress.

And they brought back this exhibition, now on view at the Met, anxiety and hope in Japanese art. So, there’s anxiety, hope, but that’s me in the flames. So I started experimenting just with chat GPT to help me with SQL queries.

That was a simple, and I started getting used to, you know. And then I started thinking about like how we might be able to use this for my long list of backlog cataloging projects because for now I still need to catalog and store the data.

So, this is, so we have, I need image descriptions of thousands of digitized and born digital photographs in our collection not of the collection but of like archival material exhibitions the building as you see here and then i need to catalog and classify and alt text and i used the gpts through open a i went to our senior project manager for emerging tech Brett Remfer, and I said, hey, how do I do this? And he’s like, you should try this out. Bruce Wyman created these GPTs. And then, so this is my prompt. This is what I put in. It’s based on the Cooper Hewitt visual descriptions instructions.

And actually, people should talk to, I went to Mark from the Lowe Museum. He gave a really good case study on visual descriptions yesterday. So, you should all contact him. And now we’re looking at, like, can we train something to recognize our objects in the collection in these images, not these images, but in our images.

Okay, so what do we have going for us on the DAM team? Like trying to look at this holistically humans plus AI.

So using the AI as an intern for data cleanup and data creation. Is this intern providing a good enough draft for me to review and edit? Question mark. And scaling is a challenge, but at least we’re starting with something. We humans are good at context and interpretation aboutness.

So let my colleague Stephanie Post do the investigative work on the assets that really need it. And create artisanal descriptions for those. Relying less on data as text can we break up to link the data and reduce maintenance on the data. And currently we have a lot of really good rights metadata or at least information on who to go to to talk about it. And then our, the way that the system is set up the permissions and access information provides us with guidance on what’s restricted.

Here’s some internal weaknesses. AI is not very good at nuances. Humans are good at context and interpretation from a certain point of view and in a way we’re trying to rehuman the machine.

We have infrastructure issues. You know, we’re trapped in systems that don’t allow for a lot of flexibility, resources. Do I put it, this is a side thing that I just started like kind of poking around with. This isn’t my, Douglas is here. This is in my main job. Do I put in time now to save time later? Like, do I have time to figure this all out? It is feeling a little existential dread slash threat to me at the moment.

So, I really should try to figure this out. So, I have potentially tens of thousands of maybe useful first drafts that me or some human, I don’t have it now, but like in the future that some human will have to QC. And then for some of our digital assets, we don’t have a lot of descriptive cataloging.

As usual, you hear this a lot, garbage in, garbage out. The METs still working on an institutional AI policy. And then, like Jack talked about before, you know, we need a data management policy. And I just referenced two groups that you should look up and just see what they’re working on.

And then also just, like in the past, what we’ve all faced is, I’m just trying to caution us against using AI just to use AI. Really think about what you’re trying to accomplish.

Linking our data within the institution. But also with others. Linking with existing knowledge, graphs, like this is an opportunity. And unlocking our collection I could see these images being part of a whole ecosystem of exhibition history or donor history donors love that kind of stuff you know and then we can have AI make metadata dynamic as vocabulary changes so I don’t have to manually go in and change something once language shifts or Library of Congress changes the subject headings, you know, as their terms are deprecated, it could on the fly be updated.

And like we get out of some of the maintenance of this data and shift to value -added cataloging, uncataloged material. Again, this discovery mechanism is all TBD. And breaking the data part in different ways.

Thank you. Okay. But what I really want to talk about is ethics. Just really, it’s hard for me to be using products I fundamentally consider unethical. They’re taking everything and are selling it back to us. And you know I’m really like conflicted that I might be replacing an actual human intern like I was. I learned how to catalog, I want someone else to be in my role in the future.

So, then like, copyright, privacy, environmental impact, the money that’s being put in to this, training bias’, harmful language. And, we’ve all heard the stories about people have done these remediation efforts already and then AI comes in and strips it all out and reinserts the harmful language. Hallucinations with flat out wrong information. Authenticity of records. And I think there is a misunderstanding of the work that we do. So, we do need to be involved and I get it some of this feels really grifty and you’re dealing with smug tech bros and like nothing that came out of Monday and Tuesday, you know, at Google and Open AI really made me feel good about what’s coming, to be perfectly honest.

Okay, so what’s going to happen to digital asset management anyway, thinking about is this the right tool? Review your contracts regarding the use of your data. In the absence of an AI policy. I started with materials that I know are they’re owned by the Met or have no known copyright.

And then we need to talk about how this materials flagged as I generated. Please, please contact me. I really want to talk. I’ve been talking to everybody about this because I just need to wrap my brain around it.

Here are some ideas. I just started with basic, super basic podcasts, like New York Times, The Verge, you know, and then these newsletters. And this, I’m just real quick. Okay, this is, you guys, I asked it to give me something regarding ethics.

And it came back with an article from the timeline of art history, Met’s website. It’s really good. Taoism and Daoist art, and I just randomly selected this piece because I really liked it. It’s called Cloudy Mountains. I thought it was really, like, peaceful. Selected the work, went to the object page, found a link to a Met video called How This Chinese Scroll confronts us with our own mortality.

And in the video, the narrator says, we’ve gone from the solid world into this confrontation with ultimate emptiness of life.

Uma Nair: So much for tech. Thank you.

Can everyone hear me? Hi, everybody. Good afternoon. My name is Oma Nair, and I am an organizational strategist and the founder of the Strategic Museum, a consulting firm. My Expertise is an organizational management, and especially now it is in helping museums optimize their workloads, reduce stress, being a burnout museum burnout victim, so it’s all very close to my heart. And the reason I focus on optimizing workload and improving internal communications is because it’s important for teams to focus on their high value work that they were hired to do instead of on inefficient tasks.

So, this quote struck a chord with me as I was doing my research for this presentation, particularly because while my work uses technology it consistently always starts with tapping into these underlying paleolithic emotions that rule all of us, and no other technology I think in recent times has brought out our reptilian brain’s safety alert like AI has, and rightfully so. There are so many ways that we’ve heard today that AI could upend our lives but again AI is not something that will just pass us by if we decide to close our eyes and ignore it. So, the focus of my work and on this talk, it’s not going to be technical but I’m going to be talking about why we should be prepared to embrace AI, because it’s not going to go away. I want to cast a vision of what could happen if AI is implemented well, and what could happen if we don’t. So I will cover a few tools, but I’m going to be talking about channeling these paleolithic emotions that we have.

So Why talk about all these tedious internal tasks, you know, that we have to do instead of talking about how AI could be used for visitor experiences and creating amazing exhibitions? All of that is absolutely valid.

But I found that we generally tend to emphasize intellectual pursuits over managerial and administrative tasks. And here’s the thing.

While we are chasing all these shiny carrots and brand new ideas, our internal foundations are falling apart. We spent way too much time on work around work.

Our tools and our systems, they do not talk to each other. We have 15 Google drives that do to talk to each other. And it is important to find a news technology that will make our lives easier so that we can focus our times on the high value work that are, you know, externally focused. So that’s my, that’s my passion. How boring, but yeah. So let’s jump in. So how can AI help us internally?

In almost all of our departments, especially the mission -centric teams, we follow certain patterns in our work. There’s a dynamic rhythm to how we create exhibitions, how we document collections, how we plan our programs. We have specific methods for creating documents and even our review processes have a mechanism, have a certain cadence to it, and even conducting research, right? So whether they’re working well or not, that’s a separate issue. It’s these established patterns and methods where AI tools can revolutionize how we do our core jobs.

It can help us become more efficient in our jobs. It can help us reduce stress and workload. For example, there are predictive analysis AI tools that can learn from historical data to make predictions and suggestions about our work. Could we potentially use these to speed up review processes, reduce approval times? There are document analysis AI tools that can automatically extract text and data from documents with incredible accuracy. Could we potentially use these to streamline curatorial or research work?

Oh my gosh, the reptilian brain safety alert just like went boom in this room right now. AI -powered reporting tools can generate reports and interactive dashboards from various data sources using simple queries.

Imagine being able to ask a simple question like how did exhibition teams impact attendance last year and getting a report that tells you, gives you the answer without having to manually extract the data.

Could this free up your time so that you could potentially focus on generating new ideas to attract visitors just a thought and automating our project management systems and our processes can create a seamless experience across the museum so everyone knows what needs to happen and then so this is a vision guys this is a vision of what could happen right and you may have noticed that some of these tools are ones that you’re already using in your museum so AI is already there it’s up to us to figure out how to use it most efficiently. So using AI in our internal tools has one major benefit it saves time so think about it how much time do we spend on sending reminders searching for meeting notes documents. All of those adds unnecessary mental strain when AI tools could handle these tasks very efficiently.

Let me share a real -life example. I am currently working with a large museum who’s one of the departments hired me to create and implement a new strategy for their department.

So as any good strategist would do, if I may say so, I started by assessing the lay of the land before creating a strategy and, you know, jumping into implementing a strategy.

So when I did that, I did a capacity audit, and I discovered that the team was working at 520 % over capacity. That’s five times over capacity. Okay? They are not going to have any room to implement any kind of strategy in the future.

So by analyzing processes and optimizing their workflows, I was able to bring that down to 80%, but that was the easy part. That was easy. But what’s harder and more important is that it is more important is to ensure that the workload does not go back up to 500 % in the future.

And so using the AI features in the tools that the department is already using, I’m able to create an easy -to -use framework, a dashboard that the department head can use to easily monitor the workload and raise a flag, you know when those ad hoc requests come in, yeah, so she can raise a flag when requests come in and she’s seeing, she’s able to see that this is going to go about their capacity. So instead of constantly being in firefighting mode, she’s able to work on the next strategy, she’s able to innovate, She’s able to mentor her team, her staff, and she’s able to build skills for her staff.

So this approach isn’t just about saving time. It’s about creating a sustainable, scalable team and a more efficient and effective future. But she was able to do this because she was ready to rein in her paleolithic emotions about the fear of using tools. She was ready to adapt to using newer technology, and that is very important.

That said, implementation can be very resource -heavy because staff is going to need to be trained, to be upskilled, to use this. And before even all that, staff needs to be convinced to use better technology, and that itself can be pretty heavy resource-wise.

And many museums already use multiple tools that don’t communicate with each other. I was just talking about all of these 15, you know, 100 Google drives that we have. Even though everybody, curators, researchers, librarians, everybody needs to learn, needs to be ready to learn how to use new technology. And the biggest challenge is the daunting task of rethinking everything.

The need to pause and think is not what our reptilian brains want. Our brains want to go, go, go. And so, and it’s also important to set up those governance measures and ethical guidelines to oversee everything that AI touches.

And that is a really big heavy lift. Even with all these challenges, I think it is worth it to consider how to use technology and AI. There are lots of opportunities here. So if you look at these stats about the future of the workforce, it’s kind of worrisome, you know why? Because, multi-generational workforces are key to institutional growth. Because of something called institutional knowledge. Even with all of the systemic DEIA issues we have and the colonial practices, the hierarchies that are in some of our systems, institutional knowledge is invaluable. And yet at risk of losing this institutional knowledge if you don’t find a way to systematically synthesize and transfer this knowledge to future generations.

And that’s where technology, particularly AI comes in. This is our opportunity to connect with future generations through technology. Gen Z have really strong feelings about hybrid work, mental health and what efficiency looks like. They’re not gonna use Google Docs the way we use Google Docs. You don’t have to redo the way the museum works, for just one generation but understanding what their needs and experiences are will help you to continue to evolve.

This is also our opportunity to do DEI right. Sounds counterintuitive, I know. You’ve heard many reasons why this is not ethical, but if we can use AI implementation as a chance to embed true and authentic DEI values into everything we do and not just make token gestures, it will work. Because we have to clean up our tools, like in the project that I mentioned earlier, since we are starting from scratch, since we’re re -hauling the way that team is working, yes, cut it. Team is working, you’re able to establish new guidelines, criteria, and automation that strive to minimize bias and inequity.

Speaking of bias, yes, All of these, the bias, ethical considerations, privacy, all of these are absolutely valid critical threats that we are being faced with. But remember, technology, and specifically AI, is only as unbiased and ethical as a person or team behind its design, development, and implementation.

So it all starts with us as humans. It’s up to all of us to decide how we want to move forward in addressing these challenges. Speed of technological advancements, it’s moving at lightning speed as we know it, how do we stay ahead and compete with other institutions and technologies that are actually a threat to museum visitation.

And changing landscape of learning methods, this is a project that I’m really excited to delve into in the fall. Gen Z is learning in new ways and that’s something we need to pay attention to.

What does this mean for our field? Which often relies on traditional methods of learning. How will AI impact teaching in universities? How will AI impact the content that museums put out? So, those are all my threats. And, yeah, I’m almost done. How is all of this going to impact museums?

Well, AI may not take away our jobs, but a person who knows how to use AI might, right? But I also believe that humans have a dynamism that algorithms cannot compete against, especially us humans in the museum field.

Our work, probably more than any other industry, thrives on human judgment, intuition, and emotional intelligence. So we have the opportunity to do the right thing.

But we need to prepare. And these are just my recommendations. Look inward. It’s a mindset shift we need to think about. Think about internal staff. Think about the work that you’re doing, how can you make it more efficient? No matter how paleolithic our emotions are, the uniqueness of human resilience is going to be our most important leverage as we move forward. Then start with what work needs to be done to make our lives easier. Then ask how AI can help. Not the other way around. While you’re at it consolidate your tech platforms get those Google drives to talk to each other and most importantly upskill your team your humans no matter what AI brings for the future your staff will continue to be the most important assets in your museum if you do it right, thank you.

Nik Honeysett: Hi everyone, my name is Nik. If you’re wondering what I was presenting to the other presenters, I have this card, it’s like a business card, it just says stop talking. It’s brutal.

OK, this is me, this is not an understatement. This is probably one of the most significant things I’ve experienced. I’ve been the museum tech world for 35 years now. This is truly revolutionary. This isn’t my quote. If you are smart, you will get smarter. If you’re stupid, you should be worried. I think we need to learn how to use this thing.

And I think my overview is our work, you know, we’re all kind of information workers, our work is going to be less creation and more validation. So I’ll talk a little bit more about that.

If this is you, kind of where you are, and you did a fantastic job of some actual data around how overworked and understaffed we are. Admissions are down. AI can help with all of this, and we are tasked with doing more, and AI is a tool that can help us do that. If you’re interested in the state of the world, I don’t know if you’ve seen this is a list of the top 100 art museums post -pandemic around the world and where they are with attendance and a lot of that that right -hand column is how bad this thing is 20%, 26%, minus 36 % in some cases. We’ve talked about this. AI is already in our lives, right? We are interacting with it on our social lives and in our business lives.

We’re just confronted with it in a much more kind of visceral way. Now it’s right in our faces. And we are tasked with doing more and using this as part of our work. I love this quote, from a strength standpoint you kind of need to know what the strengths of the tool are particularly for example ChatGPT, how is it trained, how does  it work, how do I leverage those strengths to get the advantage from what I need.

What’s important to know is it is a trained system on data it’s pre -trained to do something and it’s just going to regurgitate stuff back to you in a way that you need to control and we’ve seen all these capabilities it is both a tool enhancement and productivity enhancement for the internal kind of back of house work we do a lot of the kind of day -to -day kind of repetitive tasks we can train it to do that but it also has some unbelievable opportunities to engage with our audiences in a way that we have never engaged with them before.

I’m going to share with you a this is kind of the best advice I’ve seen about how to truly interact with with a large language model.

It’s called RISEN, R -I -S -E -N, role, instructions, steps, end goal, and narrowing. This is a very good way to think about how you interact with it.

So, chat GPT, for example, it’s trained to understand different roles. So you need to tell it when you interact, when you want something from it. How do you want to act in what you’re going to give me back?

Educator, a moderator, a scientist, a curator, give it a role. What do you want it to do? Tell it the steps that you want it to do whatever those set of instructions are.

What is the end result? What does it look like? Is it a matrix? Is it a narrative? Is it some bullet points? And then narrowing. Constrain it down.

I want 500 words. I want it in French. I want it for so a 10 year old can see it. This is a very strong framework with which to get real, A to stop ChatGPT hallucinating, cause largely it’s hallucinating because you just say, like Jack said, you know give me a quiz. Well it’s like I don’t know, here you go, it’s not surprising that it’s going to hallucinate.

It’s important to know that it is A, but it’s not I, right? This is a revolution in computational statistics, right? It’s just figuring out what the first word to give back to you based on what you asked it, and based on that first word, what is the second word? It’s just going through this statistical analysis of what the information that it has based on the instructions and the questions that you gave it it’s not conscious it doesn’t have beliefs despite what that google engineer said it is not intelligent um it will produce inaccurate results um it will produce bias results um but you need to correct it you need to tell it you know and you need to understand that it is fallible it doesn’t have access to real -time data uh yet i’m not sure what the timeline for that is but when you interact with it, you can give it real -time data, right? You can say, here are some current information around this topic, use this to give me your answers back.

We’re into, many of us are going to chat GPT right now as a product, right? It’s a destination, we go there, we interrogate it, we get things back. It’s going to kind of move into relief. It’s going to become part of a service. There are already plug -ins to allow you to do things. write grant proposals, for example, right? That’s chat GPT in the background, and it has a kind of a grant proposal writing front end. That’s the way things are going.

Remember when we should require people to have Microsoft Word skills and Microsoft Office and Excel and PowerPoint, this is what prompt engineering is going to be, right? On resumes, you’re going to require people to have a level of expertise in interacting with these types of systems. It’s a tool.

Use it to aid in the work that you’re doing, but you do have to monitor it, right? It’s getting its source from a range of highly dubious sources and some very smart and intelligent and relevant sources.

The best kind of framework I think is I treat it like it’s a graduate intern right you’ve got to you got to understand what it knows and you’ve got to teach it right it’s an interaction it’s called chat GPT for a reason right it’s not just this magical thing you’ve got to interact with it and have a conversation with it. I treat it like an assistant I ask it what it knows before I’m gonna ask it about something so I don’t assume I give it all the information I have remember that whatever you give it is is not confidential right it’s going into that pool of data that is that is out there there are things coming down the pike where you can and you can build your own large language models and you can say I’m going to give you this information and only respond with things that is in this bucket of information.

Details and specifics are important and always give it feedback right if it does something wrong say that was wrong and and correct it and it will be extremely apologetic and then it will give you the the right reason opportunities if you haven’t so we’re all many of us I assume are in the world of collections discovery and access right the National Museum of Norway go to this website see their semantic search it is phenomenal there’s a backhand chat GPT interface to search or three docks or women or trees. That stuff is not in our metadata now.

Apologies to Jessica. That stuff is just generated on the fly from the images. Sorry, Jessica. Just an example, I should maybe give myself my own stock talking card.

From the audience experience interaction the opportunities are staggering right to create we used to talk about personalization right and then we figured out that maybe segmentation is something we can get away with we can truly do personalization right now to the individual an individual can interact with us and it can be different every time literally different every time we have a project and I’m just going to give you a heads up shout out to guru who are in the in the room We have this phenomenal project coming that is, actually we’re going to talk about it on museum next. It’s an AI -based interaction with collection objects that we are very excited about.

Biggest threat is authority, right? This thing with the red circle, it’s a GPT, right? So you can go to chat GPT, you can create your own kind of application in the same way that the reason Apple took off was because of the developer network.

Any of us in this room could create a GPT, which is a tailored experience designed to give us information back based on queries that we have. This guy, Charles Tanowitz, created the Met Museum Explorer, right? He’s right now cornered the market in GPTs about the Metropolitan Museum, right? If that is an argument for you guys to get familiar with this stuff, I don’t know what is.

And then Joseph Santa Ford cornered the market in AP art history.

If you’re an art curator and you’ve never seen this, and you want to be really freaked out, uh QR code then The Next Rembrandt, massively, from a technology standpoint it’s phenomenal from an art curator or an art enthusiast it’s deeply concerning and worrying.

I’ll use this quote. We tend to overestimate the use of technology in the short run and underestimate the effect in the long run. We have these massive expectations of ChatGPT, which were overoptimistic but it will get there.

I’m going to share with you my other quote about technology which I love which is taken from the movie the terminator, of course, listen and understand that AI is out there it can’t be bargained with, it can’t be reasoned with, it doesn’t feel pity or remorse or fear and it absolutely will not stop until you are dead.

I’m gonna leave you with this because I know we’re at time we work in the world of kind of figuring out the maturity of technologies this is a kind of a five -level view of maturity of way where AI is on the right is revolutionary there are very few examples of it on the left is a very conservative approach and you’ll see it’s kind of a wait and see get to get to the middle right get to this bold kind of using it for specific functions enhancing visitor services chat bots some of us are maybe already using that um right now so we’re going to post um Alex um we uh if you’re familiar with BPOC we have some free webinars there’s that we have a bunch about AI there are some more coming.

Alex Cron is going to post out a link to this slide and so check that. I want to thank Jack for pulling this together, Jonathan for leaving his secondborn child for a day and for Jessica HK and for Uma.

Thank you for turning up.

---

This recording is generously supported by The Wallace Foundation.