As AI grows more powerful, who will shape how it is used—and how will it shape our lives? Dr. Mila Gascó-Hernández and Dr. Hany Elgala, associate directors of the University at Albany’s AI & Society College, share how they are preparing students for a workplace transformed by AI. They also explore the rise of autonomous systems, concerns about jobs, cybersecurity, data centers and the concentration of power—and the important role higher education can play in helping ensure that AI is developed and used responsibly to benefit people and society.
Part One of this podcast: Developed by Humans, for Humans: Understanding This Moment in AI's Evolution
Mila Gascó-Hernandez, Professor, Department of Public Administration & Policy, Rockefeller College of Public Affairs & Policy; Research Director, Center for Technology in Government (CTG); Associate Director, AI & Society College
Hany Elgala, Associate Professor, Department of Electrical & Computer Engineering, College of Nanotechnology, Science, and Engineering (CNSE); and Associate Director, AI & Society College
AI & Society College at UAlbany
Center for Technology in Government (CTG)
From the UAlbany News Center: UAlbany Launches New AI & Society College, Research Center
Definition of AI from the OECD:
“An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment.”
OECD stands for the Organisation for Economic Co-operation and Development. It is an international organization through which governments cooperate, study public-policy challenges and develop shared standards and recommendations.
It currently has 38 member countries, including the United States. Its work covers areas such as education, economics, employment, taxation, technology and artificial intelligence.
Special thanks to our recording engineer Scott Freedman of the University at Albany Digital Media Services team.
The Engagement Ring, Episode 45:
Developed by Humans, for Humans, Part Two: Living, Learning and Working with AI
[Lively, upbeat theme music plays as program host Mary Hunt introduces the program and plays excerpts from the program.]
ANNOUNCER/MARY HUNT:
Welcome to The Engagement Ring, your connection to an ever-widening network of higher education professionals, scholars, and community partners working to make the world a better place. I'm Mary Hunt. Today on the podcast…
HANY ELGALA:
There is definitely potential to solve many human problems, whether it's the climate change or curing disease or discovering new materials.
ANNOUNCER/MARY HUNT:
Ah, the promise of artificial intelligence. But at what price?
MILA GASCÓ-HERNÁNDEZ:
Do we want in X years from now to wake up and be in an environment in a society where we cannot control what we produce several years back?
ANNOUNCER/MARY HUNT:
In part two of my conversation with Dr. Mila Gascó-Hernández and Dr. Hany Elgala, associate directors of the University at Albany's AI and Society College, we'll explore how AI is being used to address community challenges, as well as growing concerns about autonomous agents, cybersecurity, data centers, workforce disruption, and the concentration of power. Mila and Hany also explain why universities have a vital role in helping future leaders use and design AI responsibly, thoughtfully, and for the public good.
MILA GASCÓ-HERNÁNDEZ:
These people are going to be running industries, governments, non-profits. They're going to be citizens. Whatever it is that we can help them with today, it's going to impact the future of our society.
ANNOUNCER/MARY HUNT:
Here's the conclusion of my conversation on artificial intelligence with Mila Gasco-Hernandez and Hany Elgala.
[Music fades out]
MARY HUNT:
How are your students reacting to the emphasis on AI in the curriculum?
MILA GASCÓ-HERNÁNDEZ:
I think that when it comes to teaching about AI, students are happy. They want to know more about AI. They acknowledge that AI is changing more or less the fields that they want to work in later on. I think that the main concerns for students are when we teach with AI, or when we discuss the use of AI by them and by faculty. Ah, you probably have read some articles of students suing their professors because they were misusing AI actually, because they were using AI and then they were not checking for hallucinations, they were not checking for accuracy…
MARY HUNT:
The students were suing the faculty?
MILA GASCÓ-HERNÁNDEZ:
Suing the faculty.
MARY HUNT:
Because usually you hear it's the other way around.
MILA GASCÓ-HERNÁNDEZ:
Exactly.
MARY HUNT:
The faculty are wondering if the students are using it inappropriately.
MILA GASCÓ-HERNÁNDEZ:
No, students suing the faculty and then saying we are paying all this money to be in the university, and then we have this professor who is misusing…
MARY HUNT:
Certainly not here. That hasn't happened here.
MILA GASCÓ-HERNÁNDEZ:
No, it hasn't happened here.
[Laughter]
MILA GASCÓ-HERNÁNDEZ:
So students are concerned about our faculty's misuse of AI. On the other hand, there are some students, particularly at the graduate level…I haven't found this much at the undergraduate level, but at the graduate level, some students that do not want to use AI mainly for environmental concerns. So, there is this discussion that we don't highlight as much that shows that, of course, there is a lot of water, energy consumption, location of data centers in marginalized communities, etc. etc. and some of the graduate students that I have discussed these issues with, they do not want to use AI, and they are happy to learn about AI, but they are skeptical about how faculty and students may be using AI. So, I think that we see several reactions. We have talked about the booing in the graduation ceremonies about AI. There is fear that AI is going to take jobs away from students when they graduate. The current market situation is not helping either in this respect. Although, there are some articles that say that that situation doesn't have anything to do with AI, but of course it's easy to link what's happening in the market with the increasing use of AI. So, I think that we are seeing a variety of reactions, differences between undergrads, grads, and different perspectives of how to approach AI. And clearly for faculty it is very, very important that students are not cheating with AI, and we are actually doing a lot and discussing a lot of that in our AI academies.
MARY HUNT:
Tell me about a project where students used AI to solve a real community problem. Has there been any examples of that on campus?
MILA GASCÓ-HERNÁNDEZ:
Yes. So, we at CTG, we have a long history collaboration with the city of Schenectady. We have been working with them for many years, and we have a project we that started as a code enforcement project, where the city of Schenectady needed support to identify if the grass was very long, if there is a window which is broken, and so they can address those. And CTG built…
MARY HUNT:
The Center for Technology and Government.
MILA GASCÓ-HERNÁNDEZ:
Yes, the Center for Technology in Government (CTG) and in particular, our director of innovation and technology, Derek Werthmuller, and his team developed, created this device that is based on machine learning and computer vision that we place on the top of municipal vehicles, for example, a bus, a trash truck, and these vehicles go throughout the city, and they capture this data, this information. It is processed on the vehicle, and then it is sent to the city of Schenectady. So, they feed a dashboard, and they can make decisions with the data in this dashboard. After this is an ongoing project. We have of course this tool has been improved over the years. Given the project that we had with Schenectady, CDTA, the Capital District Transportation Authority, also wanted to use that tool. I cannot remember how you say in English, you know the places where people wait for the bus…
HANY ELGALA:
The bus shelter.
MILA GASCÓ-HERNÁNDEZ:
The bus shelter. Thank you. So, they want to see if those bus shelters are clean. You know there are people waiting, etc. So, we are also using this device, which we call the CAT, Community Asset Tracking. They're also using the CAT to be able to identify this data and this information. And again, the CDTA will be able to make decisions with the information that comes from the CAT. So, we are using AI-based tool that has an impact on the community. The work that we have done with the city of Schenectady, for example, has been embedded in a strategy to address urban blight, which is the deterioration of cities that that has happened in Detroit, but has also happened in Troy, in Schenectady, even in Albany, and so clearly addressing urban blight means addressing crime, addressing economic situations, addressing marginalization of communities. So, it does really have an impact on the community. And
HANY ELGALA:
I believe this project, they had students involved in this, like collecting the data, processing the data, getting the experience of the actual workflow on the ground, and being able to communicate what they manage. So, these projects are an example of how to get different stakeholders together, and AI and machine learning is the common technology.
MARY HUNT:
I wonder if there's a way, or it's even practical, to look at a class as they enter a cohort and look at them again in a year or two years or three years to see are they working in… after graduation are they working in fields where AI has become a significant part of their work across the disciplines. Is anything like that a possibility or even of interest? Would that tell you anything?
MILA GASCÓ-HERNÁNDEZ:
I think that will tell us that the work that we are doing with students is in the right direction. I was going to say that one of the one of the things that we want to do is with these AI microcredentials is to improve the employability of our students. So, if we really see that our students are getting good jobs and they're getting them in a good amount of time, and these employers are valuing their knowledge about AI, I think that we can be successful. We haven't done that so, but at some point, particularly when the students that are taking the microcredentials graduate, we will have to go and talk with employers. We have some reports. Some people have already talked to employers, and they say that they value AI knowledge, that they value microcredentials about AI. But we haven't done this research ourselves. We will have to do that and see what impact what we are doing has on employability.
MARY HUNT:
In July, there was an incident involving AI agents, and I'm going to ask Hany to explain what an agent is in a moment, that went beyond their intended sort of testing environment and hacked into I think some businesses that they weren't supposed to. They weren't explicitly given those instructions, so they acted on their own. So as AI moves beyond, say, answering our questions and more simple activities, to actually taking actions on our behalf, does that change the risks we need to be thinking about? And as a university that's experimenting with some of these technologies ourselves, what implications does it have for us in terms of what we need to think about before we give AI more autonomy.
MILA GASCÓ-HERNÁNDEZ:
Maybe you can start with the explanation of AI agent.
[Laughter]
HANY ELGALA:
Yeah, that’s a very…
MARY HUNT:
AI agent first, Hany. Tell us what
HANY ELGALA:
Yeah, before I tell you what the agent is, there are different ways to interact with AI and AI models, yeah. One of them is the most popular interface like ChatGPT or Cloud or Gemini from Google where you prompt and you get your answer. Yeah, whether you ask to have an answer in the form of a text or a generated image or whatsoever. Yeah, there is another level of interaction where you work with an agent in the background or an assistant. Actually, like whether an agent or van assistant, conceptually they are the same. And this is where it's not a step-by-step sort of question and answer or prompt and response. It's a goal-based, like you define a certain goal, and this model you give the model access to different tools to enable the model to achieve this goal, or the agent or the assistant to achieve this goal. That's the main difference. It's almost close to automation. Like you request something, the goal is specific, and this agent or this assistant has several tools. It will be able to somehow identify the context of your goal, identify what resources are needed and what tools are needed. It will orchestrate between all these items, and it depends actually on how you configure it. You can somehow approve the different steps, or you can let go, yeah, which is the risky part of it. You can let go, and it goes from one step to another, because it has the capability to create files on your device, it has the capability to search whatever folders and documents on your files. It has the capability to go and search the internet and come back with whatever answers it can. It can also execute programs on your behalf on the machine, so it's as if you have a person, that's why it's called assistant or an agent, taking care of whatever request that you are asking. Yeah, so that's the main difference in terms of the classical AI models, where you call them large language models, and the agents or the assistants where they are capable of reasoning and making sense of the context and making sense of the recipe that is needed to achieve this goal.
MARY HUNT:
How do we guard against or limit what it can do? It did some… I don't know if it did harm in this instance, but it did things it wasn't intended to do, which raised some antennae.
HANY ELGALA:
No, let's give you some context when it comes to this attack on a company, this company is called Hugging Face. Yeah, so it was basically like cybersecurity research. They had this agent within a sandbox. So, a sandbox is a term that is used to highlight a secure environment, so a controlled environment. However, and the model or the agent was asked to in AI machine learning in order to evaluate these agents or assistants or models, you have benchmarks. Yeah. So, benchmarks conceptually, you have a certain task or certain questions, or depending on the what exactly you would like to evaluate. So, the model was asked to perform on this benchmark. However, in order to really score high on this benchmark, the model decided to figure out a way to escape this controlled environment sandbox by finding a security hole, and actually just to make it more interesting, it was not just one agent, to be honest. Yeah. It was, if I remember the number correctly, it was like 1200 agents. 1200 agents. And what's interesting is within this environment, they created like a chatting environment, so as they can communicate with together to brainstorm, to reason, to figure out ways. So as if you have a group of agents in a like a chat room, they are communicating messages, and 700 of them contributed to attack, so not all of them. So not all, like not all of them are somehow contributing. Like some of them are innocent.
[Laughter]
HANY ELGALA:
So, like 700 managed to figure out this hole. They went to Hugging Face because Hugging Face has data, has models, all resources that agents need so as to score successfully on this benchmark. So that's the story. IT sounds scary, but the good news is that these agents, for your audience, did not create consciousness. They didn't want to escape. They are not evil. They just would like to fulfill what they were asked to do is to score high in in a certain benchmark. So that's the reality.
MARY HUNT:
Wow! So, they didn't intentionally… There was no harmful intention, but it could have created some harm. I'm wondering, what do we learn from that, or what does a university learn in terms of the experiments that are taking place there, or what can you take from that?
HANY ELGALA:
Several things, yeah. I will say the technical parts, and maybe Mila say the other parts. Yeah, I mean this automation has to be granted gradually. If it has to be granted, it has to be granted gradually. The privilege and access to credentials has also to be limited. Accessing the internet and keeping logs and monitoring, setting whatever is required to monitor the behavior of the model has to be set as well, as well as also maybe that's in general thing, there must be like a continuous investigations, or somehow we call it, and I'm not a cybersecurity researcher, but they call it red teaming. So red teaming means figuring out any issues with the data, the model, the tools, whatever, the whole ecosystem, it has to be continuously red teaming to figure out any weaknesses within the deployment. Yeah, and this is valid whenever the university has to deploy certain models on campus or valid for any entity that would run models and use models in their workflow.
MILA GASCÓ-HERNÁNDEZ:
Yeah, I think that in the end it comes down to having policies, having guidelines regulating how we use AI and what AI can do. I think that we cannot and I know that there are some people who assume or believe that we should let AI do whatever they want, but I think that we are realizing more and more that even if it's not intentional, and hopefully it will never be intentional, or we'll have a problem, harm may be made, and we need to avoid that. Several months ago, one of our public engagement fellows invited the director of a nonprofit, which is called TechTonic Justice, Kevin De Liban, that was his name, and he is a lawyer. He had a case several years ago where a group of people wanted to sue the government because of the misuse of AI in eligibility of benefits, and we did a podcast at the Rockefeller Institute of Government, and the interviewer asked us what do governments need to do to avoid these risks of AI, and I of course gave my perspective. And then he very, very plainly said governments should not use AI, and I was like how Interesting still in the age of AI, there are people who think AI should not be used at all, and I think that we need to think a little more critically, if you will, about how we use AI, with what purpose, and are we willing to have certain negative bad implications to achieve other goals.
HANY ELGALA:
I just want to mention that all these capabilities are for not the general intelligence or super intelligent. They are still like somehow classical sort of agents. So, if we reach like general intelligency or super intelligency, where these models are like I don't know, more than like hundreds of IQ, for example, just more than us, whether they did it intentionally or not, but they are intelligent somehow. How an entity that is less intelligent control an entity that is super intelligent, it’s a very weird situation,
MILA GASCÓ-HERNÁNDEZ:
Right. So, do we want in X years from now to wake up and be in an environment in a society where we cannot control what we produce several years back? In order for that not to happen, it is now the time to put limitations to AI development.
HANY ELGALA:
Agree, agree, and that's why the tone is being changed here. Like because it's showing capabilities that is developing very fast. Actually, you can see from the U.S. government now, they have some regulations where you need to share this model before you release it with the government to double check and some of these models that were capable in cybersecurity, like the government asked to stop making them accessible to the public, and they had to… the company had to stop the access of these models.
MARY HUNT:
That’s interesting. I'm going to ask you in a moment about Bill Gates. He had a recent essay on the future of AI, but he even said in that essay, he said, I know guys in the cybersecurity field that are brilliant, and they're scared of what they're hearing others are capable of doing — their counterparts are capable of doing. So even the smartest among us, you know, are concerned about these issues.
HANY ELGALA:
And yeah, I think because the thing is that they are different from us, even if we are intelligent creatures, and they are intelligent. Although they score IQ higher now, you can go and check. There is a website that showing the IQ.
MARY HUNT:
The IQ of the AI agents?
HANY ELGALA:
Yeah, yeah, yeah.
MARY HUNT:
Oh, I was talking about real people. He was talking about people that are even afraid of you know what other people who manipulate the AI data can do. They're just getting so smart, but I didn't realize they are actually characterizing the IQ of the AI agents.
HANY ELGALA:
I mentioned previously the benchmark. So, this is the more practical way, but just to make it easy to digest how intelligent they are, they let the models do IQ tests, so as to make a comparison between humans and the average human is scoring like 100. And last time I checked it a couple of weeks ago, it was about 140 plus. So, there is a gap. Yeah.
[Laughter]
MARY HUNT:
Oh boy.
[Laughter]
HANY ELGALA:
Yeah, yeah. So, the security, as you were saying, like cybersecurity is one of the critical or very dangerous, like sort of a bad actor could do it. Not just the models themselves. They decide to… they created the consciousness themselves, and they decided to do bad things. The bad actor can use the models that are capable and like do cybersecurity attacks. It’s a double-edge sword. Like because you can use this model to do the attack as well as you can use this model to defend yourself from the attack. So, it still there is war and race everywhere, like whether on the company level, whether on the specific industry level, like in terms of the cybersecurity, the attacker and the defenders.
MILA GASCÓ-HERNÁNDEZ:
But the thing is that we have learned with the use of technology that bad use by bad actors happens faster and more efficiently that good use by good actors. So, they're always ahead of the good actors. We have seen that with police, for example, terrorism, cyber security.
HANY ELGALA:
That’s an excellent point because we were talking about regulation and promoting regulation. Regulations are for good actors.
MILA GASCÓ-HERNÁNDEZ:
Yes, that’s true.
HANY ELGALA:
They are not for bad actors, yeah. So, you don't do regulation for bad actors, yeah. So, it's good to seriously like consider what possibly bad actors can do.
MARY HUNT:
Gotta out smart ‘em, I guess, if that's possible. You both come to AI from very different backgrounds. I'm just curious if there's something in your own training or your lived experience, your own your work experience that causes you to look at AI differently than most people, or what's your special sauce, your secret sauce? What do you bring to the to the college?
HANY ELGALA:
So I'm the, I mean, like one thing to know, like the leadership of the college by design is from a different background to have an interdisciplinary approach in terms of leadership. So, I'm the engineer in the college. I was exposed to AI definitely before ChatGPT. So, I used it in my research around 2019, yeah around this time, 2020, yeah and I applied in my research and I thought like okay like there was definitely before the AI Plus initiative and the college like why don't we offer a course at the Electrical and Computer Engineering Department? So, I reach out to the department chair and designed the course and offer it now for four plus years, I believe. So, yeah, I started research first from the engineering, like I do wireless communications. So, it's a pure STEM discipline. I applied in research a couple of years. I took the initiative to offer it in as a teaching, but somehow something changed because I decided to get more involved, more engaged. So, I submitted my application to be part of the provost’s initiative, like being a Provost Fellow on AI and Academic Integrity. I served there for one year. Also was supported by like being part of the like leadership academy, so as to get also the skill sets. So, I got the skill sets from a leadership perspective and the AI perspective. Yeah, and I’m spending most of my time on AI on different levels, as well as a heavy user of AI as well. Yeah, and the final stage so far is the college, yeah. The college started like 2025, so it's a young college, and this is where I'm happy to work within the college, yeah.
MARY HUNT:
Mila, what about you? Yeah. So, in my case, I am a social scientist. I am in the Department of Public Administration. I'm the research director of the Center for Technology in Government, and therefore my teaching and my research has always been in teaching in technology in government, and I've done anything and everything that has to do with technology used and implemented in government organizations. So, it was natural for me to start studying AI. We actually at CTG we studied… we started doing projects on AI before the AI Plus initiative at the university. At some point, the AI Plus initiative came up, so we decided to align our research as much as possible with the university's priorities. And more and more, our projects were about AI, and then I also, of course, was embedded in those projects. So, it was kind of a natural transition from the work that I have been doing when it comes to technology in government. I think it has been very interesting because I've realized that we are not starting from scratch when it comes to AI. Yes, AI has different capabilities at previous technologies, but still there are certain lessons that we can apply also when it comes to AI, CTG is 32 years old, so we've seen any technology, right? And we have a lot of lessons that we have learned from past projects, from past moments in time, and many of them we can still see when it comes to AI. So, I think that I am bringing also that expertise when it comes to technology.
MARY HUNT:
It's a big responsibility running the college, and you spend a lot of time talking about and thinking about important issues and heady issues. So, I have to ask you, how do you unplug?
[Laughter]
MILA GASCÓ-HERNÁNDEZ:
I'm going to tell our anecdote. So, I unplug singing, dancing in Spanish Hany unplugs listening to podcasts in AI. So, we were teaching this session of the AI Academy in Cobleskill, I believe, right? And I do not know how that… I cannot remember how that came up. But we always say in these sessions because of our different backgrounds, we are going to focus on different things, and our perspectives are going to be different. And sometimes we don't agree at all. So, and that's completely okay. We are open to discussion, and then I don't know how it came up, but I told them, see, in my way to from Albany to Cobleskill, I was singing my Spanish songs in Spanish while Hany was listening to a podcast about AI.
HANY ELGALA:
For me, it's difficult to unplug. Yeah, so yeah, I somehow, I enjoy it. For me, it's like a hobby. Yeah. So, you know when you read fiction, nonfiction, whatever. Like for me, that's my hobby. Like a recent hobby. Like reading about it. Like using the different tools and options and making comparisons. I try to do some extra activities, but I mean like AI is a big part of my life. Yeah.
MARY HUNT:
It’s where your heart is. That’s nice. I would be remiss if I didn't ask you a question about the data centers. Just curious because AI for a lot of us is sort of an invisible technology. We're only recently realizing that there's a lot of physical infrastructure behind AI.
MILA GASCÓ-HERNÁNDEZ:
Mm-hmm.
MARY HUNT:
Communities are being asked to absorb much of the physical costs of AI, whether it's energy, water consumption, land use, new infrastructure, what should they get in return, and who should have a voice in those decisions? Is this a government decision? Is it communities in partnership with government? I mean, what are your thoughts on thIS? We're hearing a lot about this on the news, and communities are pushing back, and a lot of our leaders are changing their viewpoints on it when they're hearing from communities. Any thoughts on that?
MILA GASCÓ-HERNÁNDEZ:
From my perspective, ideally we should say this is a co-decision between government and the community. The truth is that governments are making those decisions without taking into account communities' perspectives, and that's why we're seeing so much community pushback to these data centers. The problem with communities that we are seeing is that these communities are usually marginalized communities. Socioeconomically speaking, they are in the low range, and therefore we do know that those communities do not participate in government decision making, and therefore their voices are hard to be heard. That also has a lot of implications in the location of these data centers because they are precisely impacting those communities who cannot say a lot about those data centers. Not only because they do not have the time and the channels and the mechanisms to do so, and sometimes the knowledge to do so, but also because they do not understand the implications of having that data center in their communities. So ideally, it should be a co-decision. I am not seeing that yet, and I am happy that more and more communities are trying to push back, usually with the help of the help and support of advocacy groups.
HANY ELGALA:
For me, just briefly, like for me, I acknowledge the concerns of the community when it comes to data centers and the environmental effect and the resources. I mean, just mentioning because hopefully this could be coordinated better, and the feedback or the input from the community is going to inform the decisions and the regulations, and somehow the incentives that should be given to the community and controlling the technology behind the different data centers, but there is also something that is being mentioned regularly which is with the prosperity that we are going to get with AI, there could be like a universal basic income or universal high income, or so yeah, hopefully, we can as a government, as a community work together to minimize the or control the negative impacts of the technology, and hopefully, we reach a point where this is going to have a positive impact on everyone, yeah. But it's problematic. It's a complex situation, yeah. But hopefully, there is a bright future soon.
MILA GASCÓ-HERNÁNDEZ:
I think that one of the problems with these benefits that everyone will get eventually is that they are not concrete to the people, right? I have done a lot of research on participation and the use of technology in participation processes. And when the topic is so broad, nobody wants to participate. Now, if you say we want to do something about that streetlamp that doesn't really work, then everybody wants to participate. I think that here we have kind of the same problem or the same challenge, right? Yes, maybe AI will heal people that are sick with unknown sicknesses, right? But that is not affecting me directly. So, do I care about that? Hmm…
HANY ELGALA:
No, no, I totally understand. It's not like a simple problem. You have many stakeholders and different interests and different players. It’s good to have transparency. It's good to set priorities. It's good to yeah to work together like on different levels.
MARY HUNT:
I think that interdisciplinary approach you're taking in the college and at the university is key too, because it's not just from one perspective and one profession, one discipline is not going to solve this problem alone. It's going to take a lot of expertise across the different disciplines. I mentioned the Bill Gates’ essay on the future of AI, and you know it was pretty provocative. It was scary in one sense and very optimistic in the other. And I think what his point was: Hey, we've got to get planning. You know, we've really got to do some serious planning. I'm not seeing that happening, and he's trying to sort of get everybody on board about that. And he did say, “AI will either be the greatest equalizer ever invented or the worst source of injustice.” Just kind of curious for your thoughts if you saw the essay, what you thought. He also made a point... You talked, Hany, about workforce. He was fearful about what might happen to the workforce, and perhaps if we planned appropriately, we could head off some of those issues. But thoughts on, you know, AI's potential impact on the workforce, and you know where are we headed?
HANY ELGALA:
Yeah, I can start with the problem is that like the criteria for designing these models or agents is not to augment humans so they became more productive, better? It’s about like a potential to replace humans, yeah, so this is a wrong criteria from the get-go, yeah. I think that's what makes the students skeptics, scared, and many people are also… they are not sure what the future will bring. So that's the problem with the workforce. In terms of the potentials and the negative impacts, I mean currently you can see that there is a concentration of power, like the knowledge and the progress and advancement-it just like you can count on your hands how many companies are having this power. So, this is showing that there is something that could lead to like losing like agency, like in terms of democracy, also could be endangered, so the current situation somehow has to be adjusted somehow, so people can see the more the positives, can feel more secure, can see the engagement and the contribution to the decision making and their future. Yeah, but currently the situation that's why maybe Bill Gates was, yeah, maybe not sure how things will develop. There is definitely potential to solve many human problems, whether it's the climate change or curing disease or discovering new materials. I can count forever, yeah, but in order to reach these goals, the current situation is very scary in terms of the setup.
MILA GASCÓ-HERNÁNDEZ:
Yeah, I think that the concentration of power that Hany is talking about is key. One of my students is working on procurement of AI by government agencies, of course, and we are seeing that only four or five big companies are procuring all governments in the United States. What does it tell you about concentration of power and therefore, you know, impact of those developments? I think that is a very, very important issue. I also want to say something. I believe that, and I think that history has shown that when every time there is a change, a revolution of something, technological revolution for that matter, the fear of technology or of that change taking over jobs is there. If you think of the Industrial Revolution, for example, and the automation of factories that has happened for 200 years, that has always been a fear. And before we had this many people working on the lines, and now they have to press a button, and robots do most of the work, and that has happened inadvertently. People have not really realized that it was happening, and there was a lot of concern in the very beginning, and now people have kind of forgotten because new jobs have arisen, and new ways of doing things have arisen as well. This might also be the case with AI. We do not know yet, but maybe we need to be creative and plan, as Bill Gates was saying, what new professions, what new jobs we can think of that give meaning to humanity.
HANY ELGALA:
If I may add some positivity, maybe this will change the dynamics a bit. Like you are aware that there are closed-source models, like proprietary models, and there are open-source models. Closed-source models are models that like you need you don't have access to like the weights of the model. Weights of the model… this is like as if you download the model, you don't have… you cannot download the model. You have no idea what the data and the configuration that was used to train the model. So, this is closed source. We are talking about some of the models that belong to OpenAI, like different models or Gemini or Anthropic, yeah. Although like these companies also have open-source versions, yeah, but they are not as good as these. Currently there are opensource models coming from China, and these models are really good. Like the gap is narrowing down between closed-source and open-source, so maybe this is going to disrupt this concentration of power because the society, the communities, they are going to have access to capable models. It's not only about their performance, but they can fine-tune it to their specific needs, because with closed-source you can still fine-tune it, but it's like they are too big, so you cannot just run them locally on your machine, and you don't have the funding even to buy whatever resources are needed, but in open source models, especially the small size ones, you can fine-tune it so as to serve specifically and augment you in whatever activity you would like to apply AI on. So open-source movement could also disrupt this concentration of power somehow, and like while listening to a podcast on my way (laughter), like it was mentioned that NVIDIA just acquired Hugging Face, and Hugging Face is the platform for open source. Like if you go to Hugging Face, like this is where like they offer different services, but they offer like-open-source models, fine-tuned models. So, this is a move from NVIDIA to somehow support the open-source move. Definitely, they have a like financial interest in this, but also this is also going to be very useful so as to have this capability of the technology available almost for free to be used and to be fine-tuned for specific activities.
MARY HUNT:
You know, back to your point about the IQ of the AI agents and assistants. I think what Gates was saying too is that, you know, they have the potential to surpass human expertise in all fields, and questioning, you know, are there some things that we just don't let AI near. We say that's for humans only. There are some activities that just are off limits, if it's even possible to do. I don't know, but are there some things that we just say that's too valuable or that's too precious?
HANY ELGALA:
I mean, definitely. Like whenever like the decision is going to have an impact on humans. we cannot let it be made by AI. At least humans have to be in the loop and making the final decision, considering the individual situations. That’s a most important thing. Also, we shouldn't be like as I was saying, like this human interaction. We shouldn't have applications that can lead to isolation. I mean, it's a case-by-case, not all one-size-fits-all because like teenagers and like different generation they interact with AI differently, but also there must be some applications that if they have a negative impact on humans, like being lonely, having like mental health related issues, all these things, we shouldn't let the AI intervene. There are some areas where, when it comes to defense, like if the decision is made to deploy AI with a high percentage in defense, you are forcing others to do the same. So, somehow, we will be in this vicious cycle of improve and deploy and then improve and deploy. I think this will bring us to a different situation that we cannot imagine how this will end up being.
MARY HUNT:
Mila, decisions you think should never be delegated to AI, regardless of how sophisticated it is, and what should we teach our students never to outsource to AI?
MILA GASCÓ-HERNÁNDEZ:
I have several areas where I would never use AI, even if we can do some sort of human-in-the-loop thing. I think that's in the end again an ideal thing. It cannot be done really well, and I will give you an example. But clearly, anything that has to do with counseling, for example, to provide specific examples, I would not allow AI to do. Although, we see a lot of counseling doing by or done by AI, and I was reading the other day a very interesting article that said that people feel more comfortable talking to an AI and explaining their problems and their situations than talking to a human, and yet I think that's an area where we should not be using AI. Defense —I am very reluctant to use as well because of the many concerns that we have already discussed, and in general decision making that requires creativity I think should be done by the human. I am very skeptical about the human-in-the-loop thing. I think that using AI is unavoidable. Although we say this area should not be using AI, we see that everyone is using it, and then we kind of let ourselves be relaxed because we say, "Oh, but we'll do human-in-the-loop,” right? So again, going back to the work that I do, what we've seen in governments is that before a public manager or a public servant had, for example, 10 cases for a housing, child welfare, whatever area this person was working on, he or she had to deal with these 10 cases, and now they are using AI to make decisions about cases. And because AI makes things easier, now instead of 10, they are giving this person 150 cases, and they are telling him, "Oh, we are going to be good because you are going to review all those decisions.” So before it was 10, now it's 150, and of course, many of those cases are not reviewed any longer by the human. So, I think that we are making ourselves feel good by saying human-in-the-loop, but practically it doesn't really work. So, I think that we need to get creative on how we can address this.
MARY HUNT:
I think I'd like to recommend you both for an advisory council on AI regulation or policy.
[Laughter]
MARY HUNT:
You seem to be proponents for AI, but you seem to say there's a lot of issues here that we need to consider and we need to be thoughtful about this process. Do you think this makes the role of the university more important?
MILA GASCÓ-HERNÁNDEZ:
I think so. We are we are educating the future leaders or the leaders of tomorrow. I don't know how to put it, right. But we are educating. They are here. They are working with us, and so clearly families play a very important role, but these four years or more that they are going to spend with us, we have the opportunity to support the learning about AI, but the learning about using AI responsible and thinking critically about AI. So, I do think we have a very important role. These people are going to be running industries, governments, nonprofits. They're going to be citizens, whatever it is that we can help them with today, it's going to impact the future of our society. So, I do think that higher education institutions have a very important role to play. In the specific case of the University at Albany, I think that we are already playing that role with our AI Plus initiative. We have three branches, if you will, or legs, if you will, of this initiative: the AI academics one, which is led by the AI and Society College, the AI Research, which is led by the AI Plus Institute, and embedded in it is the AI and Society Research Center and the AI infrastructure for all the more equipment-related issues. I think that having this broad perspective, recognizing that we can influence our students, but also our faculty and staff in how they use and how they think about AI, is already making the university play a very important role in the Albany area and probably beyond, because we are often contacted from other universities, even abroad, that want to know about our experience and about the work that we are doing.
HANY ELGALA:
Just to add on what Mila said, because we are preparing leaders for the future, because they are not just going to be users of AI; they are going to be designers of AI. They are going to make the decision to decide to deploy AI in certain place. So, like they will play different roles, and they should be prepared accordingly. And it's good to get this, like, full picture in terms of like the technical, the ethical, the responsible, as well as the hands-on experiential learning that we also are good in offering, whether it's the larger-scale access to the resources at on-prem here, or even the AI maker space that we mentioned as well.
MARY HUNT:
If you could put one warning label or user label on every AI system, what would it say?
MILA GASCÓ-HERNÁNDEZ:
Handle with care.
[Laughter]
HANY ELGALA:
Designed by humans.
MILA GASCÓ-HERNÁNDEZ:
Oh, that's a good one.
MARY HUNT:
That is a good one. How about for humans? Designed by humans, for humans.
[Music fades in]
HANY ELGALA:
I'm always very positive, so there might be like some from today in 10 years like some ups and downs and challenges, but hopefully it's going to be a happy ending.
MARY HUNT:
Well, I think with you folks educating the next generation, the people, as you say, will be making the tools and making the decisions and using them, I think that's very optimistic and very hopeful. So best of luck with that important work.
HANY ELGALA:
Thank you.
MILA GASCÓ-HERNÁNDEZ:
Thank you.
MARY HUNT:
To be continued, I guess. Right?
HANY ELGALA:
Yes. Sure.
MILA GASCÓ-HERNÁNDEZ:
Absolutely. Thank you.
HANY ELGALA:
Thank you so much for having us.
MILA GASCÓ-HERNÁNDEZ:
Yes.
ANNOUNCER/MARY HUNT:
My thanks to Mila and Hany for such a thoughtful and thought-provoking conversation. AI is changing quickly, and none of us knows exactly where it will take us. But conversations like this remind us that the future of AI isn't only about technology; it's also about people, the choices we make, the questions we ask, and the kind of future we want to build. Thanks for listening.
For more information on Dr. Mila Gascó-Hernández, Dr. Hany Elgala and the AI and Society College at the University at Albany, visit the resource page for this podcast at the engagement dash ring dot Simplecast dot com.
The Engagement Ring is produced by the University at Albany's Office for Public Engagement. Special thanks to our recording engineer Scott Freedman of the University’s Digital Media Services. If you have questions or comments or want to share an idea for an upcoming podcast, email us at UAlbany O P E at Albany dot E D U.
[Music fades out]