Will AI make us more productive, healthier, happier, and better informed—or will it fuel mass job displacement, environmental damage, social isolation, and financial collapse? Elizabeth Merritt, director of AAM’s Center for the Future of Museums, will explore why both futures are plausible but neither inevitable, and how museum professionals can make values-informed decisions about the best and worst uses of this rapidly evolving technology.
Transcript
Elizabeth Merritt:
Hello, and welcome to Live or Die in AI. Forecasting futures of artificial intelligence for museums and society.
This session is scheduled to run an hour and a half total. Until 03:30PM eastern time. I’m planning what should be an hour and fifteen minutes of remarks, leaving, I hope, about a quarter of an hour for you to weigh in with questions, statements, stories, and opinions.
I’m. Vice president of strategic foresight and founding director of the Center for the Future of Museums at the American Alliance of Museums.
And a lot of the job I do for you collectively is reading a ton of content, filtering it for important, significant information, synthesizing and sharing it with some commentary on what it might mean for museums and their communities.
Right now, I’m spending a lot of time bringing a foresight lens to bear on artificial intelligence.
And you can check out the CFM blog. We’re putting a link in chat for a series of posts I’ve been going back and forth on with colleague Nick Honeysett at BPLC.
Talking about what AI means for museums, frameworks, they might use for decision making. In this talk, I’m gonna give some framing for that information. And help give a nice firm grounding for thinking about some of the big issues that we’re being confronted with AI.
Now with any given issue, there’s a range of data and perspectives. And usually, I can find as a futurist that there’s some consensus on potential outcomes and risk and impacts.
But there’s very little consensus about AI in a wide range projections. In part, that’s because new tech is always a challenge in foresight. That’s a truism in caps encapsulated in Amara’s Law, which was coined by Roy Amara, who’s a former Amara observed that we tend to overestimate the effect of any new tech and underestimate the long-term impact, So we think, oh, wow. This is suddenly gonna change our lives, and we overshoot think it’s gonna make much more of a difference than it ends up making. But then we underestimate the profound long term ground swell changes it can make over time.
Now this gap between our effective estimate of short term and long-term impact is particularly huge when it comes to artificial intelligence. Because of the current speed of acceleration and the large amount of uncertainty around its impact in its funding model and potential regulation.
And this has led to widespread confusion about what AI might mean for people generally, for society, and for each of us personally.
Here are some of the things I’m hearing as I go around the country listening and talking about.
AI. Some people say, I’m excited about the potential of AI, but my staff’s slash colleagues slash friends hate the idea and are resisting adoption.
I can’t tell whether using AI tools is gonna make work easier and free up time for more important stuff, or whether they’re just gonna increase pressure on me to do more and relegate me to fact checking.
I feel like I’m falling behind other staff who are using AI tools, but my employer isn’t providing any guidance or training. And sometimes I have concerns about how it’s being used.
Or more generally just I have some concerns. Feel free to drop in chat your reaction to these thoughts. We’re gonna actually capture that next in a poll I would like to know before we start out how you as an audience feel about artificial intelligence, We’re gonna be putting up a poll where you have these options, and you can pick all that apply. It’s fine if you’re conflicted.
How do you feel about artificial intelligence? Are you excited?
Or hopeful or intrigued?
Apprehensive, confused, or conflicted.
So we’re gonna put that poll up on screen. And you’re probably gonna have to scroll through the little poll window to see all of the options. Remember that you can pick more than one. And after you pick your selections and hit vote, you will see a running count of how people in the audience feel about this. And I’m gonna do a little observation and commentary here Let’s see. In the lead, 31.43% conflicted. Running a close second, apprehensive, running nice solid third, intrigued. And relatively few people saying excited or hopeful.
Though, consult… Confused is way at the bottom with only 4%. So good. Sounds like you’re a well-educated group about AI already that we can use as a jumping off point. Oh, well, thank you. That gives me some information going into this talk about how this may be received. We can go ahead and take the poll down now.
I am not surprised that there’s a mix of reactions. You think about it, artificial intelligence is just a tool.
It happens to be an immensely complicated tool, unlike a simple tool like a hammer. And tools are just things. They’re not inherently good bad. They can be used for good or bad things.
So is AI a tool that’s going to, you know, make work easier and improve health care and fuel new discoveries and improve accessibility and tackle big issues. Is it gonna turn out to be a tool that amplifies bias and steals creative out and destroys jobs and manipulates opinion and accelerates wealth and equality.
Well, the answer, of course, is both. We’re navigating a time of huge uncertainty as we figure out what AI can do and what the boundaries should be, legal or social, around what it should do.
So how can we forecast the future of AI with that immense amount of uncertainty?
Today, I’m gonna bring some foresight tools to bear. By helping us think about how the decisions we make around AI today could affect the future. First, I’ll offer a roundup of the major issues. And a framework for thinking about those issues a… AI creates for museums and museum staff, and some thoughts on the role museums might play in helping society make decisions around AI.
In keeping with CFM’s foresight approach, gonna start by looking at the bigger picture, the potential impact of AI on society and then use that as a jumping off point to think about the implications for our field. Museums are part of the larger world. We can’t thrive unless society thrives. We can’t deliver on our emissions in a society that’s not ready to receive it. So, how might AI impact society?
For this talk, I’ve created a super simple scenario set. So these are three stories in… That are explore potential futures shaped by AI.
And I’m further simplifying it by focusing on only a handful of elements of each story. And then I’ll use these scenarios as the jumping off point for three parallel stories of museums in an AI future. So here are the three scenarios I’ve worked up for you.
One is the AI apocalypse looking at the worst that might happen. We get all our fears out on the table. One is what I’ve called the non workers’ paradise, where all the good things promised by AI comes to pass. And then right in the center, the middle of the cone of plausibility is muddling along. Representing a future we’re likely to end up if we simply keep going in the direction we’re headed.
Now notice this set of canary… Of the scenarios encompasses both dark and bright futures. One of the great things about foresight is it helps us deal with fear and motivation.
Dark scenarios help us manage our anxieties and motivate us to move things in a different direction. Imagining potential bright futures empowers us to figure out what we can do to make that preferred future a reality. And business as usual scenarios like the muddled middle can jolt us out of complacency. Alright. Let’s peer into the dark future first.
To get all our anxieties out on the table and also because it’s too easy. And a little bit fun.
So what would an AI apocalypse looks like?
Now I characterized this scenario as the worst thing that could happen. But the dark future I’m gonna talk about isn’t one of those ones you may hit seen in the press or fiction where AI becomes sentient and takes over our infrastructure and tries to kill humanity, at least those of us who haven’t uploaded our consciousness into the cloud in effectively become AI ourselves.
I don’t think those futures fall in the cone of plausibility in the near medium future. And, also, I think they can distract us from the real and immediate damage that could be inflicted by AI.
So this dark scenario is shaped by much more likely forces many of which are already playing out today. Things like the implosion of the human workforce or communities being rendered effectively uninhabitable.
Or the ruin of the Internet due to hacking and damage to infrastructure, or collapse either of AI itself or of the economy fueling AI.
Futures create scenarios by looking for signals of change in the present and projecting where they might lead in the futures So let’s look at some of the data points that suggest that these anxiety producing, possibilities could play out.
Collapse of the human workforce. Well, according to one peer reviewed database projection, up to 30% of US jobs, both blue and white collar, could be automated by 2030. That’s a picture of a robot in an Amazon warehouse that could be used to replace the not-so-great jobs that Amazon at least is paying people for.
Last year, senator Bernie Sanders, who’s the ranking member of the senate committee on health Education, Labor, and Pensions, released a report warning that artificial intelligence and automation could eliminate nearly a 100,000,000 American jobs over the next decade.
And finally, Verizon CEO Dan Schulman has predicted that AI will cause unemployment to rise by up to 30% in the next two to five years. That would put it at about 35% by 2031. And for context, unemployment at 19… In 1933 at the height of the Great Depression was 25%. So that’s pretty bad.
What about uninhabitable communities?
Well, what would cause that kind of problem? Water scarcity.
Data centers in The US are projected to consume up to 73,000,000,000 gallons daily by 2028, which is equivalent to about 730,000 households and the majority of them are cited in areas that are already suffering from drought.
Research shows that data centers, particularly clusters of data centers in urban areas, are heating up the area around them by up to 16 degrees Fahrenheit, which means making life hotter for more than 340,000,000 people across the world.
And growing documentation of the health impact of data centers shows that they produce noise at levels that can damage hearing. They disrupt sleep and increase stress. And that this damages disproportionately inflict on black and low-income communities.
In this dark scenario, how could we arrive at Internet ruin? Well, in part, because AI massively amplifies the ability of bad actors to hack and disable systems or even cause harm unintentionally. When AI becomes a bad actor on its own in order to follow its directions. Dario Amadei recently warned that a rogue AI swarm could seize the Internet within a year and cause hundreds billions in damage. Note that warning comes from the CEO of Infra which is one of the major companies designing AI that has this capability.
And finally, let’s look at two possible causes of collapse. One is model collapse. And that is the failure of AI systems itself So mono collapse is a documented property of large language models which are the models that power, image generation or text generation, Turns out that when LLMs train on a AI generated data, synthetic data as opposed to human created data, eventually, they fail due to the amplification of errors and imperfections. So they just start creating nonsense. If it’s an image, they create images that are blurred and incomprehensible. If it’s text, it starts spouting nonsense.
And comprehensive mode collapse of systems trained on data harvested from the web particularly plausible right now because 40% of music uploads more than a third of podcasts, and more than half of new English language article published online are primarily being written by AI. So hashtag irony by displacing human created content AI may be sowing the seeds of its own destruction.
The other kind of collapse we might have to worry about is economic collapse, the financial collapse of the speculative AI economy. In 2025, 60% of the growth of The US economy was driven by AI. By that time, the cumulative investment in AI was about $880,000,000,000 while the software revenues were only a 175,000,000,000. And this is not paying off for small businesses. A recent study by MIT showed that small businesses that integrated AI into their operations to try and increase profitability had a 95% failure rate. So even some of the leaders of AI tech are warning that this AI financial model could collapse, and the fallout could be worse than the dot com crash of two thousand that wiped out trillions of dollars in retirement savings and led to widespread tech layoffs and sparked a national recession.
Okay. That’s depressing. But is it plausible?
Is this just my dark imaginings, or is this something that could actually happen?
Well, all of the plot elements in this scenario are extensions of things that are happening now use incredible projections from credible sources of maximum impact.
I’m just gonna respond to a comment from Ruth in, from the living history firms. Do those jobs numbers take consideration that we will not have enough people to fill many jobs in the future?
Ruth, this is one of the many complications of trying to do, projections on the impact of AI on jobs. And if you… I have a running list of this. It’s really hard because the projections are all over the place. What jobs will AI destroy? What jobs will AI create? Who is going to still be trained or in a position to take the jobs that are available? So it’s a very complicated topic. And that’s a good thing to dive into. Well, let’s go a little brighter. Let’s visit an optimistic future that I’m calling a nonworker’s paradise.
What’s the best that could happen?
Well, we could have AI embed in the world in a way that actually improves our lives on a day-to-day basis.
Data centers could move to space.
K-12 education could become more effective in We could have near magical improvements in health care and research, both through research and diagnosis and formulation of treatment, and we could have mass prosperity from the shared wealth generated by AI. What would that look like? Well, nobody has to carry cash or ID. You can swan through the world with embedded facial recognition just knowing who you are and what your bank account holds and taking the money out so that you don’t have to worry about sullying yourself with individual interactions.
Everybody has a personal assistant. No more arguing over who gets to have, a… An assistant, in the office. Because everybody has one. It’s your AI, and it runs out and books your appointments for you, and it makes your travel plans, and it summarizes educate… Any information you need from memos, and it writes your updates. Fabulous. Crime goes down because that ubiquitous surveillance from things like flock cameras means nobody thinks they can get away with theft or crime.
And AI through this ubiquitous monitoring, can optimize our use of energy and actually reduce the emissions from c o two and the drain on the power grid.
Well, all those nasty side effects of data centers like heat and noise and sucking up water Elon Musk has proposed that we move data centers to space. That would reduce the cost of running the data centers to produce AI. I mean, once you had the initial costs of getting these satellites up and running, it solves the issues of water and power usage. Because you can cool using space, and you can generate your power from solar.
It is, in effect, the ultimate knot in my back yard because data centers wouldn’t be in anybody’s backyard, as people have pointed out to me, astronomers.
Because it’s gonna make a lot more visual noise out there for people trying to observe the sky. Personal AI tutors. First lady Melania Trump actually had a event this year where she pitched the idea of at home zone annoyed educator robots.
It doesn’t have to take that form though. Many private schools and some public schools are already experimenting with the use of AI to provide instruction for the… All or part of the day the for-profit Alpha Schools, which have about 50 campuses in The US, promises that students can learn math and reading and languages and science twice as fast in only two hours of work each day, using AI tutors, training up the rest of the day for sports and art and life skills like starting small businesses. And in this scenario, the only humans in school are motivational guides rather than educators.
They feel this increases engagement and reduces inequity because those AI tutors can be producing a curriculum and feedbacks specific to every single student and not just one size fits all.
We have improvements to health care. Partly through ubiquitous monitoring instead of seeing your doctor once a year and hoping they catch something. You can have monitoring embedded in everything, from your wearable device on your wrist to your toilet. Collecting information and sending it to your doctor.
You can have seamless integration of all of your health information. No more five different medical records, one with each of your specialist and they never talk to each other.AI is actually really good at testing potential new drugs with models and predicting what might work and how to produce them.
It can improve diagnosis. It turns out that working with human experts, AI can improve the ability to accurately read things like x rays and do diagnosis.
And also, it does have the potential for designing personalized treatments. There are a lot of rare diseases and conditions that never get addressed because the cost of developing treatments for them is cost prohibitive, sometimes AI can figure out what the individual drug would be for that or can game out how you would do gene manipulation to fix it.
Doesn’t solve all the cost problems, but at least it brings into the realm of possibility that some of these rare or individual conditions and diseases could actually receive attention.
By the way, that smart toilet actually exists. There was a report in 2020 about a scientist who created a smart toilet whose, I love this line, Number one source of data is number one and number two. It uses physical and molecular analysis and uploads the data for analysis. There’s a commercial available from Kohler too if you want to try it. It only costs about $300.
Okay.
And last of all, mass prosperity. And a lot of the dialogue about AI has been, it’s gonna generate so much wealth that no one has to work if they don’t want to.
Plausibility check. Could this actually happen?
Well, I am going to go to the folks you would think would now a position paper that was created by OpenAI earlier this year. It’s called Industrial Policy for the Intelligence Age, Ideas to Keep People First. And OpenAI said, well, what would have to happen in the world for this bright scenario to be true? And it has a long list of recommendations. I highly recommend reading it.
Most of their recommendations are directed at the federal government. Not all of them. They… For example, they say data centers should have to pay their own way. So instead of stressing the grid and using up water and dumping externalities like heat and noise, they should have to make sure that they’re cleaning up the mess they make and that they’re paying real costs instead of just increasing the cost of electricity and water for everybody else.
You’ve heard a lot of the issues about, you know, is A out there?
Causing problems by hacking data or potentially doing damage to mental health by having chatbots that encourage bad behavior or people who are suffering from eating disorders or even self-harm. And OpenAI says, well, the government should regulate that. It should tell the AI companies they can’t do it, that there are certain things that they are either have to approach slowly or test very thoroughly or can’t do it all. So guardrails.
And that the government should protect labor rights both by saying that there are certain jobs that can’t be displaced by AI and by creating training pathways into human centered work. Which they note are things like childcare, elder care, education, and health care, none of which are particularly well paid, I will note. But they also say that they shouldn’t… The government should incentivize employers to improve pay and working conditions in those fields.
Also to provide security, implement a universal health care system, and to solve the problem of the profits of AI going mostly to big companies including OpenAI, there should be a massive system of redistributing wealth, both a tax system that redistributes profits generated by AI to displaced workers and federal social safety net program. And also a public wealth fund that provides every citizen with a share of the revenues from AI.
II will let you sit with that and decide how likely you think it is that the federal government will take those actions.
There are other ways to get to these points, of course. Since the report came out, anthropic analysis would create a public benefit fund, and OpenAI has floated the concept of a health… Public wealth fund that they would pay into from their profits. But first of all, these are just ideas. They’re not commitments. Secondly, the companies would retain control over how the fund disperse. And also, if you do the math… I’m trusting somebody else here. I didn’t do the math. Apparently, the money translates to something like a couple 100 to maybe a thousand dollars per US resident, which isn’t gonna to offset lost jobs and income. So great vision. Not quite clear how we get to reality.
In any set of scenarios, you want to look at what’s often called the mainstream future or the probable future. This case, I’m naming it the muddled middle.
It’s where we’re likely to get if absent intentional motivated action like government regulation gonna end up if we just run along a mishmash of national local attempts and current economic incentives And they’re gonna get if we’re going absent something or someone disrupting the trends that are shaping our path forward right now. And we’ll see it’s very likely that we’ll have acceleration of wealth inequality, continued environmental degradation, impact on labor, not clear how much, but some, an erosion of trust, some impact on health, and a patchwork of regulation that doesn’t entirely take care of these problems.
Let me quickly go through some of the data that suggests this is already happening. As I said, it has yet to be seen what AI efforts are going to be real winners in the long run and be self-sustaining and profitable. But right now, as big investments in AI incrementally pay off, it’s only accelerating the enormous wealth gap in The US. 4% of the largest private a com… AI companies have created 15 millionaires with a combined network of 38,000,000,000. The seven factors of anthropic AI are now multibillionaires.
As I mentioned before, that isn’t spreading out. There aren’t a lot of little people or little businesses making a lot of money from their use of AI. 95% of business initiatives using AI have zero return on investment.
With labor… Sorry. Environmental degradation, already seeing a lot of the damage that I projected forward in the darkest scenario. Two thirds of data centers are already in water stressed areas.
Heat effects are already impacting over 340,000,000 people across the world. We’re seeing damage to the work… To the human workforce. The impact of offline climate is highly uncertain.
But it seems clear already that the biggest disruptions are going to be to entry level positions, the kinds of work that you can most easily automate with a… AI. So maybe it used to be that you started on the ground floor by being an executive assistant, where you started in a legal office by being a legal aid.
The firms that are doing significant staff reductions tend to be freezing hiring lower-level positions, which leads to the question of in the long run, how are you going to have pathways to advancement? You have to do this stuff work to get the experience and the street cred and the resume to move up. If AI is doing all the basic work, how do people get trained, and how do people get credentialed for taking higher level jobs? I have a little note on this image I have on the left. Recent story, robot companies raise $30,000,000 to pay 5,000 people to record real work tasks. So they’re saying, okay. We’re gonna hire you. To wear monitors to teach a robot how to do your job. And I looked it up. At least one of these, projects was offering people a total of $6,000. Again, you’re taking that because you’re desperate for money, it’s going to set you up to be financially independent for years to come.
The other impact to the human workforce is it tends to produce higher stress and expectations on productivity. For decades, decades, if not maybe a century, people thinking about technology have said, oh, technology is gonna mean that in the future, we only have a twenty-hour work week, and we can use rest of our time to do art and relax and be with our families. But you know what? Well, he’s productivity historically has always been used to have employers encourage staff to use the extra time to be more productive and funnel more profits. To employers. So far, that doesn’t have a great historical record playing out. Health and well-being.
Only gonna mention these. You’ve probably seen stories on the news every day. There are a lot of forms of AI that are increasing social isolation as people use chatbots rather than real friends. There have been active harm from chatbots encouraging people to take bad medical device or to do self-harm. There’s increased anxiety and burnout from an increase in screen time. There’s a lot of health misinformation out there. Mn six, do not use chat GPT as your primary source of medical advice and information that can help you formulate good questions. Questions to ask your doctor, but it is not your doctor.
And, of course, one thing that created a lot of fear over the past few weeks as we approached the election it’s really been manipulated and stored in reality.
Another thing to worry about, is the fact that it’s creating an erosion of trust. A recent study showed that 50% of the Internet is now AI slop, People recognize this, and now I find myself doing this. You look at an article and you say, is that true? Did a human write it? If an AI wrote it, it’s inaccurate. If I see a video, I say, is that a real video or was it created by AI? Is it trying to intentionally fool you?
This is being reflected in public opinion. You already can see that when you poll people about, AI generated content and factual accuracy, 33% of people disagree with the response to the question to what extent you agree with the statement I trust AI generated content to be factually accurate, and only 19% say they do trust them.
In in our bright scenario, we said, okay. Regulation could be a way of managing some of the worst harms, but we’re not seeing a really integrated high level approach to managing AI through regulation. We’re seeing a patchwork of regulations sometimes focused on really weird things like Ohio that tried to seek human AI… To ban human AI marriage. I’m not sure where that came from.
In China, they actually did make it illegal to lay someone off in order to replace them with AI, but it’s not clear how that’s being applied or enforced in ports.
Is this plausible? Absolutely. It represents the mainstream future. It’s what’s highly likely to happen, absent intentional or accidental intervention.
Wow. That’s a lot. That’s three stories of the future that could be created in the larger world by AI, and there are a lot of other stories you could fill out in between. That’s just three bundles of potential outcomes.
What would it look like for museums in that future? Alright. What about us?
What are the implications for AI in our practice, in our operating environment? Well, I used those three scenarios of the world to create parallel scenarios for museums
And I wanted to start bright this time. Let’s not be all down. Let’s look in an out to at an optimistic outcome, especially because I really think AI could create an AI powered renaissance for museums in part because there are some absolutely killer applications for AI in museums. And I would love to spend all my time on this.
But that’s not the purpose of this talk.
I will name just two, for example. The amazing way that AI can allow museums to create compelling immersive entry points for their exhibitions.
And for people to explore what the world would look like and how they could enormous amounts of that material. This is the National Archives Museum, the American Story. This is a picture of fifty thirty times increase in transcription speed for insect labels using robotic imaging with AI and optical character recognition. So that’s amazing when you have a collection that numbers in the millions It’s a magical application to get AI to help you with transcription and making sense. I have, for many years, said Amer, that the collection policies in American museums are a little pie eyed because they act as if storage is infinite and staffing is infinite.
And how are you ever going to be able to take care of all of that massive amounts of stuff that you’re collecting?
These are some shots from the National Museum of American the Smithsonian.
Well, maybe AI means we actually can have a few in which there’s no backlog. We can actually catch up with the massive amounts of specimens that we collect, whether you’re a natural history museum, whether you’re an archive, whether you’re an art museum that has a lot of stuff in storage.
So maybe AI could mean a future in which museums have no backlog. Another possibility, Museums are always struggling to find new financial models. Well, maybe new income streams could be built by monetizing bot traffic. We’ll be talking about this in the dark future in a moment. But one of the things AI uses museums and libraries for are sources of data to harvest by sending bots to go skim that data off of the online collections or databases and use to train large language models.
That’s problematic in a couple of ways, but there are ways museums are beginning to use to say fine. You can come use our information, and we’re gonna charge you. I think that AI has potential to make museums even more fun and engaging and accessible. And because of that, more people may value museums more than ever because they find new ways to get into things that they did not find intuitive or engaging before.
The other thing is… And this is sort of a backhanded compliment, Because AI is so massively distrusted right now, people could value museums more than ever because we’re trusted sources of data in time when so much information is questionable or fake.
And building on that, in a in a world in which AI makes data and information cheap and accessible, when I can pick up my phone and say, hey, chat GPT, giving… Give me the answer to these facts. Museums can be more valued than ever because of the two things AI doesn’t layer on to that, which are knowledge and wisdom.
So that’s a bright future that could be created for museums by an AI inflected society.
But I can’t ignore the other side of the cone of plausibility. What would the AI apocalypse look like in a museum edition? Well, first of all, trust could simply tank. The public trust museums put in museums could erode because AI adoption as the bad rep of AI rubs off on our fields. For example… Oh, by the way, AI colon semicolon Doctor is short for a new Internet phrase that means AI didn’t read. It’s playing on the, acronym that stood for too long, didn’t read. Some people are simply not believing images, whether it’s an AI generated image of pope Francis wearing a Balenciaga puffer jacket or whether it’s an article, Sometimes tools could give a general bad rap to anyone who uses them. So for example, genetically modified organisms It’s a tool. There’s nothing inherently wrong with messing with DNA. It can do some really great things in terms of, like, trying to solve genetic defects.
But because one of the major sources of, applications of GMOs that people know about is creating genetically modified crops that allow you to dump tons of fertilize…
Tons of pesticides and herbicides on fields, GMOs as a whole get a bad rap. So if AI as a whole has such a bad reputation, it museums’ use of AI could tarnish our reputation as well, the spec the public opinion. As a subset of that, I would say, there’s a potential for careless use of AI to damage relations with indigenous peoples and descendant communities. We’re only slowly building trust with these groups, around our commitment and ability to be appropriate stewards of indigenous knowledge and sensitive information.
If we fail to protect this data from AI harvesting and misuse, if we apply AI in a way that unintentionally amplifies bias or perpetuates misconceptions. It could undo this progress.
So The fact that so many people automatically turn to the top-level AI summaries on the top of your search screen on their phone or on your computer. Might mean that over time, people don’t even think about where that information comes from. And that might mean that the public values museums less or a smaller portion of the public values museums because they don’t see us as primary sources of information and authority. They don’t know that that’s where all the good solid information is coming from.
Here’s something I’m really worried about. There’s a lot of really good research that’s coming out that shows that regular use of AI actually degrades people’s critical thinking. So even somebody who’s highly skilled, like somebody… A radiologist who’s trained to read x rays, If they regularly use AI, their ability to apply critical analysis to their field of expertise degrades.
Sometimes this is called cognitive decline or cognitive disempowerment. If we use AI regularly in museums of people are told, okay. Use this to do your work. Use this to do your first data summary. Use this to generate a first draft, and then you’ll check it. It could impede the kind of learning and experience that helps staff advance. And when the previous generation of experts steps down, will the next generation be ready to step up? I just this morning read an article talking about the fact that the regular application of AI seems to cause the sort of universal regression towards the mean because of the way that AI algorithms work, they’re always saying, well, what’s the next most likely thing after thing one? It is not programmed to for creativity. It’s not programmed for outliers. So as it’s applied over time, to anything, whether it’s, the generation of content, whether it’s designing an experience, there tends to be a flattening it’s more and more… Well, people… This isn’t dead center, the theoretical middle of where the theoretical, the fictional average person would be experiencing something or saying something or enjoying something. So it drives everything to that middle and tends to prune away the creative edges. So I really worry about keeping an eye on the effect of using a… AI on the people who are using it.
Another thing I’m worried about is I’ve gone to some conferences where I’m really disturbed by the number of museums and libraries that are being attacked literally by swarms of bots that are harvesting their data. And that can cause all sorts of problems from, you know, bringing down their servers because, it’s basically a unintentional denial of service attack. To trying to defend themselves from that data harvesting. We’re also seeing more and more malicious or AI powered sabotage of data, people who are either doing it because they want to launch malware attacks for ransom or just because they’re briefing people and wanna cause problems. And if museums over the past couple of decades, one of the great advances in museum’s work has been a committing time and resources and effort to open access, to making our data digitized and discoverable in a… Inaccessible in a way that encourage people to use it.
We might have to spend a lot of those resources and a lot of that time and attention on playing defense, on having people try and lock up data and create firewalls and create protections or running around recovering from data attacks.
So that would be disturbing. Well, how about the muddled middle?
What would museums look like in a future where society and museums don’t take actions and decisions regarding internal and external challenges.
I want to point out the screen snippet at the left side of the slide. This is one that you may have seen a couple of years ago that has stuck in my heart and mind. This author wrote; you know the biggest problem with pushing all things AI? Wrong direction.
I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes. I think that what I’m hearing from a lot of people in museums is widespread stress and uncertainty and job dissatisfaction tools they’re supposed to use, or they feel like they’re falling behind so they should be using it, but they’re not they’re not be… Being provided with training. They’re feeling pressured to increase output, but they don’t feel satisfied with the kind of work that results from that. It’s sort of that’s the quip about laundry and dishes versus art. They’d rather do the writing than create prompts and check output.
So I think we have to be worried about what museums consciously can do to prevent that outcome in the mainstream middle.
One of the few emerging pieces of consensus around best practices in AI is what’s called humans in the loop with the idea being, okay. Fine. You’re using AI to increase productivity. You’re doing the first draft of transcription. You’re doing the first draft of generating a report. You’re doing the first draft of analyzing data.
But it’s very clear that you should never put that out there with a human… Out of human checking it. And sometimes it’s embarrassingly clear when it has been published without a human checking it. So this is what’s being called humans in the loop.
To check the output and verify the actor and appropriateness of AI’s first pass.
But think about it. The reason I put a picture of Hercules in the Aegean Stables here is if one of the great things about AI is it can do such massively increased productivity and generate lots of output and analyze tons of data, at some point, humans become the bottleneck.
Are we in a position where there’s a risk of museums exceeding their staff capacity to ensure that those humans are in the loop and keep up with the speed and capacity of AI output.
Another thing I worry about is the business model of tech companies. Has long been, and this is true of AI as well, that they flood the market with free or low-cost tools both to get people hooked on the product. Now you put that you can put that in various ways. I said hooked, which is why I used an picture ormorphone addict here. You could just say introduce them and familiar with the use if you want to be more neutral. And they’re doing that partly to get a lot of users because in the end, it’s only gonna pay off if they have a lot of users.
Also because they want to undercut… They want to be the front runner. They want to undercut other people who are introducing those tools. And if they are more ubiquitous and more people are using it because it was free or low cost, maybe they can dominate the market.
But eventually, they’re gonna have to start charging.
I’m worried that a lot of museums particularly small museums right now, are saying, wow. This is great. We can use AI to take over these staff jobs. And here are things we weren’t committing to we didn’t have enough staff, but now we can do it because of AI. Maybe we don’t have to hire somebody in this position, marketing or social media, because part of that work can be done by AI, only to find, you know, next year or the year after, either the free tool they based it on has been discontinued because it was an experiment and the company abandoned it, or it’s no longer free because they begin charging. Or they really increase the costs to cover their real costs, which mean… Might mean a 100 or a thousand percent increase in what’s being charged. One writer I read recently observed that OpenAI’s behavior sounds an awful lot like a drug dealer who floods the market with addictive drugs and then jacks up the prices once users are dependent on them.
So how do we make sure that we’re adopting AI tools in a way that is measured and thoughtful and thinking about the long-term sustainability of the cost and capacity.
Another real issue is making sure that AI doesn’t conflict with a museum’s basic values and goals. And here’s one obvious example. Again, over the last decade, a lot of museums have begun to commit to reducing their climate footprint. Say, how can we reduce the carbon emissions from our HVAC? How can we think about how we might modify the requirements for, temperature and humidity and let them fluctuate a bit more. How can we, you know, maybe build buildings in a way that are in inherently more green, that generate their own power or pull less from the grid or better insulated so it doesn’t need as much a… HVAC. But if you’re doing that with one hand and on the other hand, you’re going all in with AI, if you’re pulling on models that run on data centers that are doing all this environmental damage through immense use of energy and water, that’s a conflict.
Now remember, this is the means… Mainstream future. That was the one that most museums are already beginning to experience. It’s just an amplified version of the present.
Okay. So I don’t know whether it made you feel validated for your fevers or more anxious by bringing up things you hadn’t thought of yet, or more confused because this is all over the place, I wanted to stop and acknowledge that it’s legitimate to ask why bother?
Why bother spending the emotional energy and cognitive brain space it takes to think about these potential outcomes?
I would argue it’s incredibly important because we’re at a time in which the choices we make matter immensely. Here’s a quote I came across about a year ago that gets to the heart of these choices. New technology tends to change society in such a way that it becomes difficult or impossible for an individual to function without using that technology. That’s absolutely true. Think about the phone in your pocket. Think about the way you probably live. Many of us live in cities where it’s impossible get around without an automobile.
Or it’s impossible to live our lives without access to the Internet. I think it’s inevitable that AI will have a profound impact on our lives and on society and on the systems, we live in.
But I think we’re still at a point where we can control the form that imp takes. By the way, I should acknowledge that that quote is from Ted Kaczynski. And his manifesto industrial society in its future. That’s a k I. AKA the Unabomber. He was crazy, but he wasn’t wrong about everything.
Gonna contrast with Kaczynski’s somewhat cynical quote, with one that I find inspiring from one of my favorite science fiction authors, Ursula k Le Guin. Who wrote the exercise of imagination is dangerous to those who profit from the way things are. Because it has the power to show that the way things are is not permanent, not universal, and not necessary.
Ursula K. Le Guin wanted us to remember the future is not predetermined.AI is not something rolling over us and going to shape the world we’re gonna live in regardless of what we do because we have agency, and we can make choices that influence how that AI technology is embedded in the world and how it shapes our way of experiencing life. And our jobs and raising our families.
And to demonstrate how important it is that you can be, as an individual, have power to raise these questions and push against the tide. I wanna bring up two examples.
If you’re feeling powerless in the face of what seems to be an unstoppable technological and economic force, think about people in the past who have claimed questioned equally important inflection points. And my first example is Ned Ludd.
This is the famous person for whom Luddites are named. He’s probably fixed fictional, but let’s use him as a stand in for any of the thousands of text wide textile workers in the early eighteen hundreds who protest the ways in which automated machinery was displacing skilled labor. And forcing workers into factories.
The movement sparked by the Luddites wasn’t anti-technology. It was against exploitative technology. They weren’t so much from something they can control in their homes into factories where they had bad working conditions and inflexible hours. Where the profit went from going directly to them to going to the factory owners.
And while they lost the battle in the short term, I would argue they laid the groundwork for the next century of advances in labor rights. The creation of labor, unions, and the establishment of legal protections for skilled workers. So always worth speaking up.
My second example is Lucretia Mott, who is regard… Regarded by some as the foremost white female abolitionist in the eighteen hundreds, She operated a stop on the underground railway. She was a founding member of the Pennsylvania Anti-Slavery Society.
And to my point, she and other Quakers refused to wear cotton, eat sugar, or use other products produced via slavery.
Because these goods, which made life easier and more convenient, were based on theft and exploit. Exploitation. Sit for that… With that for a moment.
Quite often, very profitable technologies are based on theft and exploitation.
And it’s too easy to say, well, this is ubiquitous, and it’s simple, and it’s easy to use, and I can get this done faster because I can just ask my computer to do it for me without stopping to consider first of all, that most of the data that is fed into commercial large language models was taken from the web without permission, without compensation from creators.
And the resulting tools are things that can do immense harm.
So can… We have to apply our values and our awareness of the world to the decisions we make about how we’re gonna use a technology, even if it makes our lives easier or better.
So with that all in mind, I want to offer a framework for museum decision making. Most museums, as private nonprofits have the freedom to choose what values will guide their decisions.
One business analyst I was reading commented AI does appear to be one of those rare inflection points where being wrong, missing the shift, could mean irrelevance.
I don’t think that’s true for museums. Museums shouldn’t have fear of missing out. They’re among the organizations that maybe… They’re not among the organizations that might become irrelevant if they miss the ship. We’ve survived over a century of being labor intensive and inefficient. We can’t, in fact, lead from behind.
So, yes, AI is dauntingly complex, but we can create a relatively simple framework for helping our organizations make decisions. We can think about values. We can consider impact.
And we can focus on the decision points that make the most difference.
So for values, for example, of course, be guided by mission. What effect do you as a museum want to have on the world?
And how does that interact with the… What you can do with AI or its externalities, the effect it’s having through how it runs.
We can put ethics first, which tech companies are notably not doing. What are our ethical guidelines about the effect that we’re having on the environment? Or on people.
And coming a close second, think about impact. We can make values-based decisions about how we want to use AI to change the world. We can assess the impact of our decisions on our community, the country, humanity in general, or to be hyperlocal, our own staff, we can think about the effect on social relationships, on mental health, on skills.
We can consider the effect on labor, on jobs, or wages or opportunity. Is it creating new jobs? Is it making jobs more exciting or empowered? Or is it creating worse jobs or driving down wages or reducing opportunities?
What’s the environmental impact?
And also, please, equity. Are the costs falling disproportionately on disempowered communities, or is it providing accessibility for communities that current… That currently don’t have access to our resources.
So here are some of the major decision points.
We can distinguish between AI applications that are magical keys for unlocking the value of museums making meaning of vast datasets, organizing archives, connecting data across institutions that had formerly been orphaned and disconnected. And applications that exacerbate our long-standing labor issues, creating fewer or worse jobs or inflating expectations about workloads or, in equity or, trustworthiness. That’s my last point there that I think is really important is how can we protect our superpower of trust?
I’m showing you some data the 2026 annual survey of museum goers where our colleagues at Will Kenning Consulting asked the American public about their expectations of museum content and AI.
And they invited respondents to the survey to tell us about their expectations about museum content and AI. 4343% of respondents said, I expect human beings to develop all content for museums, including exhibitions, emails, everything.
Another 29% said I’m okay with museum staff using AI to write emails and general website text, but not the text in exhibitions. And then 20%, so that’s one fifth, said, yeah. We could use it for brainstorming, but not to write the text. And less than 10% said, I’m okay with museum staff using AI anywhere.
And one of the things I love about the tech the, staff at Will Kenning is they read through all the open text responses, and here are some of their observations summarizing the sentiments people put into, open text answers. It’s clear that the use of AI causes some people to question the accuracy and integrity of information including at museums.
So we have to think about that reputational risk. Even if we know that we’re using AI in a very ethical way, that we created our own learning models that we’re running in house in a sustainable way, that we’ve vetted for bias.
Do people understand that? Or is the bad rep of AI potentially rubbing off on us?
Some museum goers are concerned that AI will replace the contributions of human creators and professional staff in museums. Museums have been talking among ourselves for the past decade about the importance of having human centered organizations that centered the experience of staff and center creation of good sustainable jobs.
I was kinda surprised to see that the general public was aware of that as well. This is a legitimate fear. Many people feel museums should be dedicated to showcasing and celebrating humanity.
And the AI use would run counter to that mission. One of the things I’m seeing in my scanning is that there’s a backlash against the ubiquity of digital content and about… On and ever online experiences. I think museums are finding the… Their next superpower after trust is going to be that we are relentlessly analog places where people can connect with real people and see the work of real people, whether from the past or the present. And that’s a strength we shouldn’t lose sight of. And the majority of museum goers and US adults want to see museums practice transparency when it comes to in… This… Informing their visitors of any AI that’s in use. That’s a very good point to keep in mind as we implement AI in our operations.
So I think with all of that in mind, the most important things we can do today in my humble opinion are to think in an informed creative, and expansive way about the many different ways AI could affect the future, short term and long term. And I mean both society as a whole and in museums.
Decide which of these futures is our preferable future, and reasonable people may disagree on this. You might have heard some of the plot elements in my, non-workers’ paradise that was talking about how ubiquitous surveillance would make us safer and happier.
Some people find ubiquitous surveillance creepy and intrusive. So taking some time to talk to your colleagues in your organization and for yourself to think about how you think AI should be applied most beneficially?
And then most important, the role we can play as individuals or as institutions in making the preferable future come to pass. Museums, because of their superpower of trust, are highly influential. We can play a significant role in educating the American public and helping them understand AI, in helping them understand the issues. In coming to their own conclusions about what policies they want to be enacted the laws that should constrain AI or the economic systems that should be used to share wealth or protect labor.
And take action. Individually, we are small, Collectively, we are mighty.
The individual decisions you make about whether and how to use AI in your daily life, the decisions museums make about how to create policies governing how AI is going to be implemented or not implemented in their organizations. All of these things have accumulative effect that can have a real impact on society.
I think that Mark Zuckerberg perfectly captured the guiding philosophy of big tech when he told Facebook staff that they were supposed to move fast and break things. Museums have the power to choose our own guiding philosophy and here’s my suggestion.
Take it slow and preserve trust.
Thank you very much. Those are my thoughts on this matter. I hope you enjoyed that brief tour of potential futures shaped by AI. I’m actually running a little fast, so we have plenty of time for questions comments, And I would love to hear not only questions from you, but your stories, your reactions your thoughts. So you can drop those in chat and I’m gonna be hanging out and reading that here.
And my, colleague Ariel Waldman will also be feeding the questions from the q and a, if that’s where you have parked your thoughts. So let’s see here. What about… What am what am I gonna hear from you? Hello?
Ariel Waldman:
Hi, Elizabeth. That was great. We have a lot in the chat. First, I’m gonna start with something in the q and a, though.
Caroline, wanted to know, can you offer any specific examples for using AI in museums and archives? Besides transcription and LLMs?
Elizabeth Merritt:
Okay. Yes. But I want to start that with the preface that I am not an expert. So what I am trying to do is accurately convey stories I have heard other people tell. So apologies if I use any of the wrong words for the technical aspects. One of the applications I’m seeing is what a friend of mine has been lecturing me on.
Is a lot of the really important historical records in The US live in little, tiny historical societies scattered across the country. And a lot of them make no sense unless they’re connected for context with records that may live in another little historically society across the country, and it’s effectively impossible for scholars to find and connect that information. And one of the projects he’s working on is how can even small organizations afford to come together and create AI systems that look at the information and make sense of it to say, oh, this person who is mentioned in the historical record is a freedman and that connects with a work record of… In this historical society, it explains what he was doing after he was freed after the civil war. So I think connections are one really important, source of information.
Another application and I only heard this story, a couple of weeks ago at the AI for Lamb conference. From the it was somebody who’s speaking from the American Museum of Natural History, and he was saying that they had outfitted one of their collections managers with a GoPro camera. On his head and connected it to AI, and he was walking through the collections looking at the fluid collections. If any of you have ever worked with fluid collections, it is thousands and thousands of bottles full of, like, dead little fish and amphibians and put in a fluid prison.
Preservative. And they’re almost impossible to take care of because they’re so voluminous and they require constant monitoring and care. And he was saying, well, we were seeing if, first of all, AI could use the video from the camera to count how many specimens were in the jar which is a big issue and also notice if anything was misshelved also a big issue and also flag an alert if the fluid level in the jar is getting low enough that somebody ought to pay attention and go top it up and check the quality.
So that was incredible.
Ariel Waldman:
Very cool.
In the q & a, we also have a question from Erin. Who’s asking, who said, I’m wondering if there is, movement at the Nash level in putting together an official museum industry stance or guidelines on appropriate use of AI.
Elizabeth Merritt:
Thank you for asking. I’m actually spending a lot of time talking to people in the field.
Including Nic Honeysett at BPOC, who I mentioned earlier, about how we can put together an AI initiative where AAM could help play a role in bringing people together and saying, what are the issues? What are the approaches? What are the most useful things museums can do to begin tackling this right now instead of sitting on the sidelines? So yes. We’re working on that. Nick and I are gonna work on one, building on the resources we’re publishing on the blog this summer.
Ariel already shared a link to that.
We’re gonna put together a PDF that we can release that begins to put together some thought pieces, some guidelines, here are the issues, and a list of things museums should be thinking about. And then we’re also working on framing out a larger ARC of activities over a couple of years that would try and bring together people from the field to say, okay. What is the right approach? Now there can’t be one answer because this is changing too quickly. I think what we need to do is find a framework for questions to say we need to keep an eye on this and here are the here are the values informing those decisions… Decisions and then constantly check back and say, so how’s this is looking? Because a lot of the specifics of the issues are shifting so quickly You know? Maybe some of the AI companies were… Are going to solve the things that do cause ethical issues for museums.
Not holding my breath, but if they did, you can’t have one list of here are the good applications and here are the bad ones. I think what you can do is say, here are the questions you should ask when you’re assessing an application to decide whether it aligns with your values. Things like, you know, where was the data sourced from? Was it collected with permission? Or with compensation?
What are the filters being used to try and eliminate bias? Is it secure? Are the private… Privacy concerns being applied? So I think what we can do is help the field at least have a framework for the right questions, and we’re working on that.
Ariel Waldman:
Great. Thank you.
Gary’s question is how do I encourage my board to consider the impact of work slop without making them defensive? So, Gary, if you wanna
Mhmm.
Elizabeth Merritt:
That’s a good question.
Can I ask you? Can you say because all of these terms are being coined, so recently, clarify what you mean by work slop.
Ariel Waldman:
Write in the chat or respond to your original, question, please do so.
Should we go to another question? Give Gary some time to
Elizabeth Merritt:
Well, no. I’m gonna… I will tell you I’m gonna say here’s how I’m interpreting it and then you take it from there. Okay. So the way I’m gonna interpret work slop in this context is, low-quality output that has been generated by AI that nominally has done the job and is really bad.
This is a thought off my top of my head.
Sometimes it’s easier to disarm people around sensitive issues with humor. And there are some pretty funny examples of work slop. So I think it might be possible to, bring those to do an exercise where you’re inviting them to use some AI system to accomplish a task, and then they compare and assess the results.
And there are times where I think the inherent humor and how bad some of the out is might catch their attention.
That’s just a thought from me.
Ariel Waldman:
Another person in the chat is saying that work swap is basically messy, busy work. It doesn’t actually make sense.
See, this is why I was asking because that’s different than the definition I was given.
And I would say unfortunately, that’s new… That is not new to AI.
It’s just a new variant on an old problem.
Ariel Waldman:
Definitely.
Elizabeth Merritt:
I think, while Ariel’s finding the next, question, I do find that when you’re trying to focus people’s attention on what work is actually important, whether confusion is caused by AI or something else.
It starts by the question is what is the effect you’re trying to have on the world? What are the really useful, important metrics? Not necessarily the easiest to measure metrics, like the number of tweets you put out or the amount of text you generated. But the metrics that really get at did we create that result And then measuring the work against whether that fit those metrics and met that goal. Because pretty quickly, some of the stuff that’s irrelevant, even if it looked like it generated a lot of output, is gonna fall out the bottom.
Ariel Waldman:
Great. I’ll just point out in the chat, Dace has said that the Council of European Steering Committee for Culture, Heritage, and Landscape otherwise, the CDCPP, it’s a long acronym, has adopted a new policy document aimed at aligning cultural and technology technological development in the age of artificial intelligence.
Elizabeth Merritt:
Yes. Absolutely. Thank you. And if you can share a link to that and drop it in chat, that would be great. I will say, I think Europe in general and European museums are in some ways ahead of us in dealing with these issues, in part because Europe is in a regulatory sense ahead of America in trying to tackle this legislatively. So, I need to catch up on this. But Europe relatively recently basically passed an AI transparency law saying it’s not just best practices. You have to say when you’re using AI, And one thing that ran past me on my scanning radar last week was a statement from museums saying, wow. How… What does this mean for museums? How are we gonna have to align our operations with this new regulation about transparency, about use of AI? So, yes, look to Europe. They have… They’re they are living the future sooner than we are in this regard.
Ariel Waldman:
A question kind of off of that about, saying if you’re using AI or not. Laura asked did you use any AI in the creation of this presentation or assembly of this data? It’s a little smiley.
Elizabeth Merritt:
Yay. I did not. I actually have ordered my t shirt that reads summer of blood. I am not a complete Luddite. I mean, I want to understand the technology, and there are some places where, as I say, at least it has magical applications, not primarily from my work because I don’t want to lose my cognitive and research skills.
So I’ve experimented with it a bit to understand how it behaves and to be familiar with what people are experiencing it, but I do not use it in my research and my writing.
Ariel Waldman:
Oh, Susie. Well, hi, Susie. Our great friend Susie Wilkinning. Said that, Yes. This was a main theme of the NEEMO European Museum Conference a couple weeks ago, and she linked their statement. So thank you for that, Susie.
Elizabeth Merritt:
Awesome. Thank you, Susie.
Ariel Waldman:
Going back a little bit to Gary’s which I think I missed in the chat. He said that he, Gary referred to the AI slot he’s referring to as multipage memos from board members that look polished, that pushes work to me to discern. So, I guess
Elizabeth Merritt:
that’s interesting. I’ve definitely seen that where you get a document, and it looks great, and then you read it and you’re like, what does this mean? This is all AI jargon. Like, it doesn’t make sense. So I think that’s kind of what he’s wondering.
Soso be being at the recipient end of your colleagues or your bosses or your governing authority throwing AI slop at you. Well, sympathy is Though again, it’s a new variant on an old an old problem because anybody as old as I am, raise your hand, can remember decades past, there are many memos who have come across your desk where it had a similar impact, and it wasn’t generated by AI.
Ariel Waldman:
Very true.
Aaron is asking, how could museums help with teaching the public about AI? Do you have any thoughts on that?
Elizabeth Merritt:
Absolutely.
I think that museum exhibitions about AI can be tremendously powerful. MIT… Oh, god. What was the MIT? I talked about this in a talk just last week. MIT had a great critical exhibit basically saying, here are the things you wanna know and here are the questions you ought to be asking. I think what… AI is inducing anxiety in a lot of people, not just us. And I think one of the powers museums have is to help people explore difficult stressful issues in a safe environment.
And I think this can be done by any kind of museum. Because just like any kind of museum could find an application for AI, any kind of museum could be transparent about that experiment and talk to the to the visitor about it. So it could be an exhibit if you’re a science museum or a computer museum, but there are computer museums. You might have an exhibit specifically about AI. So what is artificial intelligence? What does it do? What should you know about it? But any museum that’s using AI could say, oh, by the way, this is how we’re using it, and here’s how it works, and you might want to know this. So I think we should see this as an opportunity to make that transparency about AI into an educational experience.
And I think the more people understand what AI is and what the issues are and how it works and how it’s effect… Affecting their lives, the better the collective decision will hap… Decision making will have as a as a society.
Ariel Waldman:
Definitely. We have Laura from the Nash Aviary, which we were actually talking about right before we went live.
Elizabeth Merritt:
I told Ariel she should go visit.
Ariel Waldman:
Yes. Said that I think this is in response to, your conversation about the environmental impact. So they said that they were really glad you bring this up. The conflicting nature of being a conservation-based mission organization being asked to use AI is one of my greatest points of confusion and conflict. And one my coworkers and I are all conflicted about, and that has six thumbs up.
Elizabeth Merritt:
Yes.
Yes. Absolutely. It’s an import… I totally validate your concern. It’s an important issue.
It would be a deep technical dive, which we don’t have time for now, nor do I have the at my fingertips.
But I think this gets to what does ethical responsible use of AI look like in museums, and it can be done. Museums can create their own LLMs trained on internally sourced data that they own and have permissions for. They can run it on servers where they’re choosing where the energy comes from. They don’t have to rely on commercial data centers. So I think there are ways to look at how you’re using it and the impact and reconcile that with environmental goals, but only through really good education and conscious choice.
Ariel Waldman:
There is some really great fodder in the chat.
And people kind of swapping resources. I just wanna point out that Katarina said that she posted a question on how to navigate I think, some of these issues, on museum junk and no one’s responded yet. So to Oh, go answer her… Of course.
start a conversation there, I’m gonna re-put the link to Yes. The And while you’re doing that, thank you, Deborah Wexler, for posting a link to the MIT Museum I was bleeping on it’s called Mind the Gap, And she shared the link to the exhibit. In chat.
Thank you, Deborah. Charlie, from the Denver Museum of Nature and Science says, I really love the values driven approach you’re recommending. In non program roles, those values can feel somewhat remote. How do we push for conversations about AI and values across departments that haven’t had to engage in decision making like this.
Elizabeth Merritt:
You’re gonna kill me. Can you read that again?
Ariel Waldman:
Of course. So, basically, the question is, how do museum staff who aren’t, necessarily in forward facing roles, how do they push for conversations about AI and values across departments that haven’t had to engage in decision making like this? So
Elizabeth Merritt:
Yes.
Well, that’s part of a broader problem as well. Which is how do you manage from below?
And it’s about AI in this case, but it’s a general issue.
In muse… In any organization that has a hierarchical or org chart, which is almost all of them.
And that’s a whole session that we should hold, Ariel. That’s another training session. But I would say in general, I anybody should assume that they have the power in an organization that gives some basic respect and recognition to people at all levels of the organization to say, hey. This is important, and we should talk about it.
And at a formal level, it can be going up the chain of command to say to your boss, could you talk to your boss up to the leadership team about this is an important issue, and I think we should have discussions in the museum on it. How might we do that? Because that would need a need that the staff has. At an informal level, it can be, you know, start it yourself. I’m gonna have a lunchtime discussion about AI. Who wants to come? And a lot of people will opt in.
If you can get a critical mass of people talking about it, I think that helps make sure that it infiltrates the decision-making groups, whether or not they had been thinking about before because somebody in that room is gonna say, you know, somebody was saying just the other day, this is an issue, and we should talk about it.
The reason that I was distracted made you right read that again, Ariel, is I saw a comment, from Suzanne at the Wiseman Center saying, we found careful use of AI tools to be useful in meeting accessibility standards. Absolutely. This is another killer application of AI. Now obviously, it’s another example of humans in the loop because it can go disastrously wrong if nobody’s checking to make sure it did it right. But museums have struggled for decades with doing appropriate accessibility, especially at scale or complicated installations or with multilingual issues. And that’s something AI is really good at. So that’s another place we don’t wanna just say never use AI. It’s a bad tool. No. It can be a really good tool if we use it right.
Ariel Waldman:
We have a really interesting conversation happening in the chat with some IMPs saying, you know, they’re consultants. And so while AI can do good in some cases, it… They’re nervous about it taking some of their contract opportunities.
Elizabeth Merritt:
Yeah. I think this is a big question mark for the country in all jobs, not just museums. How is AI gonna be impact labor? And there are a couple of factors. Some of them are some of them are practical. Which jobs can AI actually do well? And which does it just look like it can do and then it’s gonna turn out to do it badly?
And there are gonna be some that fall in that category. But I do think part of it’s values based. And to go back to the comment I made earlier, building on comments on the annual survey of museum goers,
I Let me see if I can articulate this. Humans exist to celebrate life Much of that life is humanity. I’m from the natural history museum, so it’s not all humans. But a large number of museums are here to celebrate human creativity and productivity. And the output of being human. And I would find it deeply ironic allow celebrate and value human input.
So I think there are only… Also values based museum decisions to be made just like, okay. It took us long enough as a field to have some kind of informal consensus that you shouldn’t have unpaid internship. Both because it’s abusive in terms of labor and because it limits the diversity of the field. Only people who can afford to take an unpaid internship then gets the experience to get a real job.
And now we’re saying, okay. But instead of having an unpaid intern, you’re not gonna have anybody because we’re gonna have AID the work. It’s a similar set of values. What is the effect of work on humans and on culture? And where is there a benefit collectively that aligns with our human centered values and having work done by humans even if it could be done by AI. I’ve been fascinated by the articles in the news. You may have been following them where mathematicians are getting really ticked off because AI is going in and solving mathematical problems partly by stealing their notes, okay, which is another issue. But that that had been struggled with for decades. And it’s not it’s not like these are all practical applications why we’re… And now people aren’t gonna die because we solve this mathematical equation. It was a human challenge.
And in some ways, so what AI solved it?
All it did was rob the mathematicians of the opportunity to meet the challenge with their expertise and knowledge. I mean, it’s like what if AI suddenly could create a robot that ran marathons faster than people?
So what? I mean, the point of the marathon is to challenge humans, not to have something run faster. This is a real challenge to us, as individuals and in our organizations to dig down to what we think the work is for, who it should benefit, what needs to get done better and faster, and where work… Where things should be harder and slower.
Art. Art shouldn’t be easy and fast. You learn from art by sitting and struggling with it. So having AI read the interpretation of a picture to me isn’t the same thing as my sitting there forty-five minutes and thinking about hard questions and having somebody say, but did you notice this? And what do you think the museum… The artist meant.
So those are the sorts of things we have to parse through where we figure out where there are the highest best use applications that really make life better. And where are there things that are easier convenient but actually have bad side effects? Or they’re easy and convenient but they were based on bad practices.
Ariel Waldman:
I feel like that’s a pretty poignant button. For, today’s program. I will just mention that Susie, again, is in the chat. Saying you did a great job of framing this, and that… So marketing consulting, of course, is
Elizabeth Merritt:
Thank you.
Ariel Waldman:
organizes and the survey of museum… The annual survey of museum goers. And she says that people tell us machines should not be interpreting the human experience. And that they have hundreds of quotes about this. Is that a good place to draw a line in
Elizabeth Merritt:
thank people for coming today?
Ariel Waldman:
I think so. Thank you, Elizabeth, for such a great, talk today.
Elizabeth Merritt:
Awesome. I’m so glad you all joined me for the talk. I loved your comments and questions. I hope that we can save the comments, Ariel, cause I’d love to go through them all afterwards. Keep an eye on the CFM blog where Nick and I continue to curate a series exploring some of these issues. And look for, first of all, a PDF resource compiled from that later this year and hopefully a larger, deeper dive in coming years.
Ariel Waldman:
Thanks, everyone.