How combining linguistics and mathematics turned into a successful company

In this episode, we sit down with Marco Antonio Stranisci, a postdoctoral researcher in Natural Language Processing at the University of Turin and founder of aequa-tech, an AI startup focused on social impact. Marco shares his journey from humanities to computer science, his activism against hate speech, and the creation of Debunker-Assistant, a tool designed to combat misinformation.

We explore the ethical dimensions of AI, the challenges of launching a startup in a crowded tech landscape, and the importance of participatory design in building inclusive technologies. Marco also offers advice to young researchers navigating academia and industry, and invites listeners to contribute to his open-source initiative, the Citizen Dataset Lab.

We express our gratitude to University of Turin and personally Lucia Salto for the guest of the podcast Marco Antonio Stranisci.

🔑 Key Topics Covered

• Marco’s academic journey: from humanities to computational linguistics
• The intersection of activism and AI: detecting hate speech
• Founding aequa-tech and building Debunker-Assistant
• Challenges of entrepreneurship in the AI space
• Ethical concerns in AI development and data collection
• Participatory design and citizen involvement in tech
• Open-source vs. closed-source models in AI
• Advice for PhD students and early-career researchers
• The future of NLP and interdisciplinary AI
• The Citizen Dataset Lab initiative

⏱️ Question Timestamps

• 01:33 – Marco’s academic journey: switching from humanities to IT
• 02:57 – How activism led to a PhD in computer science
• 03:53 – Translating research into startup innovation
• 05:12 – Why Marco chose entrepreneurship
• 06:00 – Emotional highs and lows of startup life
• 07:26 – Benefits and drawbacks of leaving academia
• 08:45 – The crowded AI landscape post-ChatGPT
• 11:56 – Marco’s elevator pitch for aequa-tech
• 13:45 – Debunker-Assistant and participatory design
• 15:02 – Challenges of analyzing social media data
• 16:38 – Open-source values and transparency
• 17:58 – Future plans: interdisciplinary and efficient AI
• 19:00 – Advice for PhD students using AI
• 21:45 – Ethical concerns and the importance of learning
• 23:01 – Humanities in STEM: is something missing?
• 26:18 – How listeners can help: Citizen Dataset Lab
• 27:36 – Multilingual participation and open collaboration

Transcript

Copy Transcript Copied

SPEAKER_03 0:02

Hello and welcome to Eurex at Smart Talks.

SPEAKER_01 0:05

The Smart Talks Project Series is an EU funded Eurexis project which provides information and assistance to researchers at any career stage. We support researchers in mobility, careers within and outside academia, innovation and entrepreneurship.

SPEAKER_03 0:29

We strive to increase scientific collaboration between Europe and the world. Hi, I'm Daria, a senior expert at Euro Access National Contact Point in Latvia.

SPEAKER_01 0:44

And I'm Joan, a teaching and research assistant from the Faculty of Mechanical Engineering in Nicholserbia.

SPEAKER_03 0:49

And you are listening to the Euro Access Smart Talks podcast.

SPEAKER_02 1:09

The PhD is an experience that in which you can uh uh uh uh learn to learn.

SPEAKER_00 1:16

Today are guests is Marco Antonio Sunici from the Turing University, whose desire to create something with a social impact as well to the success of a company that combines linguistics and mathematics. Marco, well everyone, thank you.

SPEAKER_01 1:35

I would like to start off uh with a question regarding your academic journey. Could you please tell us uh how you made the switch from uh the humanities to IT and a little more about uh the process itself, uh where you went to uh university and stuff like that?

SPEAKER_02 1:56

Okay, sure. Yes, uh it's a long story and non-linear because uh uh my dream when I were when I went to the high school was to become a teacher of humanities, and then I discovered uh at the middle of the of this path that it wasn't really my my dream, and I fell in love with uh linguistic and computational linguistic, which is the which is the study of uh how we teach uh machines to learn to to speak and to talk in natural language, and uh in parallel, uh I had an experience as an activist uh for human rights, for which I created some uh technologies uh for the automatic detection of hate speech, and this experience led me at a certain point of my life to try to do a PhD in computer science. So I had this uh this witch from humanities that was led by linguistics, my my interest in linguistics and my activism uh against hate speech uh in social media.

SPEAKER_01 3:10

And I think it's uh pretty clear that uh you are able to translate a lot of this directly into your startup because most of the products that Aquatec uh does do are connected with moderation, uh enabling the citizens to experience the uh relative realistic content which isn't uh full of uh misinformation. So thank you.

SPEAKER_03 3:40

This uh sounds even actually tricky, you know, when you read the description, uh sometimes uh you get an impression that uh you will be evaluated what you read. I mean, the news, for example, that you can upload and screen and something that's uh like uh stop reading that. Uh the police is uh already after you, but uh um hopefully it's not, at least right now. Uh so but but but how did you decide to become an entrepreneur with this startup?

SPEAKER_02 4:15

Uh yes, there are different uh motivations about that. The first one is that uh I would like to uh continue the work that I started with uh with the no-profit with this project that was called Control Audio Against Hate. And uh, so I decided to uh found uh a company uh to build technologies with a high social impact. We just don't want to be uh tech tech guys that do stuff uh with uh artificial intelligence, but uh people who want to engage with citizens in building this type of technologies. And the other ingredient of this decision was that I met uh five other wonderful colleagues during the PhD, so I decided okay, they are aligned with me, with my core values, and so with them I did I decided that uh it was worth trying to begin this adventure without them. I would never start this type of uh this type of uh idea. Um that's that's it. It's one year that we we were founded in July 2023, so uh it's uh relatively new a new experience to me.

SPEAKER_03 5:36

How is it?

SPEAKER_02 5:39

The experience, yes. Uh I would I would say that there is some oscillation. Oscillation. Okay, there are there is some swinging between uh very excitement and a little depression because it's difficult to because it's it's new and also it's it's not easy to say, okay, we are in the AI field, and so there are a lot of people now uh that are expert, even if they are not expert of the AI. So finding the right voice in this crowd is uh is very challenging, but uh you need to be creative, and uh this is the big part, uh the the big interesting part uh and exciting part because you have a lot of new things that you that comes to your mind and you can and can transform your idea and uh make you make you feel that you are doing something new and interesting, not just for uh your experience but for society.

SPEAKER_01 6:43

That's very interesting, and uh uh I would also like to note that uh your access is primarily based on uh researchers and academics switching from the academia to industry and uh giving options for mobility. Uh taking this into consideration, your path uh is very interesting. How you managed to uh find fellow colleagues, like mind-minded people who uh together with you started uh your company. I'm interested uh seeing that uh your approach was somewhat different, that you are primarily based in business now. Uh, how do you see the benefits and the drawbacks uh of uh your uh present status? Yes.

SPEAKER_02 7:35

First of all, the crowded environment is a really uh big challenge because uh we uh started thinking about it before ChatGPT. Then ChatGPT arrives and it was a mess because everybody that were that were expert in cryptos now are expert in uh large language models, so it's uh it's an issue. But uh this is also something that made us reflect a lot. And the other uh the other aspect is also related to um to the fact that uh there is this enthusiasm, this idea that uh technology is something really um really destructive that will change the world tomorrow, a sort of terminator for social goods, but it's not like this. It's it's artificial, but it's not intelligent, the artificial intelligence. And if you study a bit, you know that uh it's just uh matrix multiplication with uh a lot of uh GPUs in your computer. Um, so it's difficult to find your voice in this in this crowd, but there are also these benefits. This is also this also led leads to a benefit that is that uh you you you you have to be creative, you have to find your own way, and we are doing it, and we are doing it uh uh putting at the core our interdisciplinarity because we are uh from linguistics, from network science, from math from mathematics. So we are creating a new creative way of thinking about our uh experience as uh NLP researchers. And the other thing is that we are a group and you fight a lot, you discuss a lot, and you uh become friends a lot. So we I am very lucky to be uh in this journey with uh the other uh five people from Equatech, and we are building um uh a team that is really worth uh regardless the the output uh the outcomes of our uh business uh business inquiry. And finally, the the things that uh that made me uh change a lot my perspective on the world was uh discussing with clients. Uh trying to find the the needs of clients and be transformed by their needs and correct our uh our uh direction is uh really rewarding.

SPEAKER_01 10:10

I think that's an excellent point. Uh sorry, Daria, for interrupting.

SPEAKER_03 10:14

No, everything fine.

SPEAKER_01 10:15

Uh clients and business are important part and should be considered an important part uh also in research because mostly we are based on theoretical knowledge and uh what we need to accomplish in university in order to get our degrees. But on the other hand, the more important uh stuff is what what is happening currently in the industry, and if we are listening to the industry and preparing the new students for uh their future work. So I hope also that uh your professors were able to help you and your colleagues in this regard.

SPEAKER_02 10:54

Uh yes, one of them was uh had an experience as a startupper as well uh 15 years ago. Uh his name is Giancarlo Ruffo, and uh it was really inspiring. He helped us a lot uh in some critical moments, also to make some uh decisions because another thing that you you need to learn if you want to do this stuff is to decide, because in the university you often do not decide, it's it's more collaborative in the good cases. Um in this case, you have to take decisions, you have to be fast sometimes, and this is something that you learn. It's you you do not uh you do not uh start with this uh this skill.

SPEAKER_03 11:41

Yeah, exactly. And uh uh about the skills, by the way. As a startupper, you definitely know what the difference is between TED talks, elevator talks, and pitch sessions. So, can you uh can we ask you uh for a small um elevator talk about your idea behind this startup? So, what is what are the core ingredients of it?

SPEAKER_02 12:06

I'm not sure I'm able to, but uh I try. Okay, the uh idea, the general idea behind this is that uh we want to create technologies with a high social impact. So the idea is that uh Equatech uh was born to fill a gap, and this gap is that there are a lot of people that are excluded from this so-called revolution uh because technologies are biased towards uh certain categories of people. So we want to uh create uh technologies that fit this need. And uh in doing so, we want to uh revolutionize the way in which we collect data because we are living an ex uh extractive capitalism uh stage. Big companies uh extract data from people, even if people do not want. We want to uh flip these uh this type of process and start from citizens uh and co-shape with them our technologies, our annotation schemes, and our uh uh and our direction. So our idea is also to uh revolutionize the way in which we uh collect data for uh for technologies and for society. So this is what EcoTech is for.

SPEAKER_03 13:27

Uh you mean you would like uh more like citizens to contribute uh to this technological okay.

SPEAKER_02 13:37

Yes. Uh let me let me explain with an example that maybe it's uh easier. We our first product is the banker assistant, which is an AI tool uh for that support the fight against uh misinformation. Okay, we have some ideas, we have developed our first prototype that is in uh almost in beta. Uh but uh our next step is to uh create together with non-profit uh journalists, also common citizens, the database for this type of uh this type of technology. We do not want to uh propose something from uh from uh from the top of our experience, but we want to collect to to collect from uh uh from society the real needs and from them to change uh to change also the technology uh that we built uh until now and enhanced it. So this is what what we what we what I mean what when I talk about uh uh participatory design of uh our technologies.

SPEAKER_01 14:45

Uh that's very interesting. And uh you mentioned uh you mentioned the news and uh uh an assistant that would help prevent the spread of misinformation. Uh, how about uh social media, other platforms? Are there uh uh products that your firm is developing which could help uh decrease the spread of uh hate speech, which is apparently uh more and more on the rise, as we can see in political debates uh and uh on these platforms?

SPEAKER_02 15:20

Yes, unfortunately, there is something that you can do and something that you cannot. And uh now uh if you want to scrape gather data from Twitter, you have to pay uh 30,000 euros per month. You can't uh scrape data from Facebook and Instagram almost legally, you can't you can technically, but you can't legally, so we can't say okay, you shouldn't uh you should okay. Uh so uh we it's impossible to propose a service that analyzes social media posts. Uh so this is uh a limitation, but uh we can uh uh we are we we can try to find a way to um to identify some sources that could be relevant uh from a social networking perspective. For instance, telegram channels uh uh allows you to download some data so you can so we can do something uh there, but uh we can't interact with uh social media. And this is one of the biggest problems that we want to challenge uh to challenge because our uh one of our ideas is to be fully open source. We are living in uh in a in an era in a period where big companies close their technologies, close their data, and we want to open them and be fully open uh uh throughout our uh entire experience. That's very noble.

SPEAKER_03 17:00

Yeah. Um so but but what are your plans for the future? Uh, how do you think your idea could be developed or and what you need for that? Okay, okay, apart from funding, maybe money, obviously. Money, or yeah, definitely. How scale or something like what are the plans?

SPEAKER_02 17:26

Uh our idea is to develop technologies that are uh uh interdisciplinary because now everything is large language model, but uh this is only a small piece, a small part of uh what is artificial intelligence and what is natural language processing. So we are integrating different subsets of computer science in solutions that could be uh possibly disruptive without being so big, so economically costly, and so on. So we would like to propose new approaches to artificial intelligence that are smaller but more uh more uh efficient, efficient. And the other stuff I told I I told you uh before, we would like to um to go uh all in into the uh participatory design. Uh so we want to engage with create a community an open source uh a community, uh an open source community with people from uh with coders that can help us support and create and improve the uh the debunker assistant and the other our technologies, and also involve people. Our idea, uh our core value is to uh create uh trust links, trust uh relationship with uh with people that could be uh to that could benefit from our technologies.

SPEAKER_01 19:00

Taking into account that uh you're in a very niche and advanced field, uh, I was wondering uh if you had any advice for uh young PhD students and young researchers on how they could use AI to uh benefit their research, or where the true uh advantages in AI lie. And on the other hand, uh should it be uh restricted in some sense uh in the university process, or is this a question of uh the calculator? Let's say new technology put in. Uh some professors think that it's wrong uh to use it to write your papers, maybe it is. Uh, but does it have other uses that could also help uh students, researchers, um, and other academics in their daily work?

SPEAKER_02 19:57

Uh I am not against the usage of uh these technologies uh for helping you uh, I don't know, writing coding also, copilot uh algorithm copilot technologies that help you support you create writing programming. Programs are very useful. Uh there is an ethical issue behind this because uh every time I use copilot or GPT to do this stuff, I'm feeding uh these technologies with uh information about myself and so they can use them, and I don't know how they will use them because they are closed source. So there is an ethical issue, but I don't see the issue related to um related to the the usage itself. I think that uh in order to uh about the support call the support of AI for coding using this type of technology, to use this type of technologies, you need to know how uh programming works. So you have to study it. Uh you you you are not a parrot, so you you need to study to understand the logic, you need to know how to ask, and this is something that you can't uh delegate to artificial intelligence, so it's not something that will replace the the the the learning path. But uh another thing that I would like to suggest to uh to people who are in the academia is as as a um former PhD student that graduated six months ago, I think that this type of congratulations. I think that uh the PhD is an experience that in which you can uh uh learn to learn, right? Because uh we we do a lot of stuff, uh we feel a lot of uh anxiety, but uh I think that if We are able to be meta uh to to create our meta condition cognition of what we are doing. I think it's a really important experience to improve a lot of soft skills that you don't have before starting this type of uh experience.

SPEAKER_01 22:17

Maybe you would also consider developing a PhD assistant, uh AI chatbot that would help PhD students with anxiety. That would mean a lot to the whole community.

SPEAKER_03 22:30

It could help with supervision, you know. I know some guys in academia who are lacking uh very much uh the uh supervision from their uh professors, but uh you know, this is another topic, another big, big issue. Uh my one question, my one takeaway that uh what you said uh made me think of. Do you think that these guys in in STEM uh lack a little bit of soul from humanities and social sciences? Human touch.

SPEAKER_02 23:07

I I don't know if uh the if it's marketing, this type of uh narrative of a superhuman uh like Elon Musk and uh Mark Zuckerberg. Uh but uh it's definitely wrong what they are doing. The it's definitely wrong how they are transforming their uh environments because also you know that Facebook a few days ago told that uh they will uh dismiss all their effort against misinformation, and that's something that uh clearly uh makes us uh conceive that the fact that these are not these are not public, this is not public sphere, and we need to be more aware about it. Uh it's private company that offer services to speculate on our way of talking and interacting uh with other people.

SPEAKER_03 24:07

Uh it this raises one more question about you know, uh, right now science is open source, open access. We are really struggling with this. Uh want to share our knowledge and research with people. And now these guys who are in on social media are super close. You you exactly, you know, this is the the number that you named, how much it uh costs to download uh the at least to have an overview of a definite topic. It's still enormous uh amount of money for someone from academia. You know, if you do not have a any private investor in in or uh interest in this, uh then you definitely shouldn't do that even. Uh, then uh maybe this is a competition between the platforms. Someone are more open, some are someone are like more closed for researchers. Uh who wins, I don't know. But uh, but still, I I mean this is a trickier thing because uh in case you can browse on the internet and find some information, and exactly the news are super open, you can download anything and um and check, but social media uh are on in a on a different planet right now, at least as I see it.

SPEAKER_02 25:29

Yes, and and even if they uh uh in the competition for the large language model, so uh uh in January uh it comes they come up with uh GPT and I don't know what's the number, then uh Gemini from Google and so on. If we look uh none of them release data uh uh on which these models are trained. So even if uh Facebook with Yama is more open sourced, uh because it released the the code of the language model, we don't know what uh in on which data is trained. So it's uh they they all they are all closed in a certain in a certain extent to a certain extent. So I agree with you that it's uh it's a yeah.

SPEAKER_03 26:15

Uh can you think of a way how our podcast could could help you? Maybe you have some like uh message uh for the listeners. We usually have this early stage career researchers who are listening. Maybe you need some help or anything.

SPEAKER_02 26:33

Oh yes, yes, I have. Uh actually, yeah, sure. Grab your chance. We have uh we're about participatory design of our products. We are launching uh something that is called Citizen Dataset Lab. So in February, we will start with uh we start collecting uh uh fact, collecting and evaluating uh uh articles that are misinformative in order to create a public open database to develop to train our models so everybody can contribute uh to these uh to this uh uh initiative. So if uh some of you want to want to contribute, you could write me at my email marco.stranici uh equamenotech uh dot com.

SPEAKER_03 27:24

Uh thank you Mark. I I hope that uh you will succeed in in your efforts. Oh, yes, sorry. Uh the language the language of these publications.

SPEAKER_02 27:35

We started with uh Italian and English, but uh this type of uh initiative is of is for every language because we want to we want to go multilingual uh with any language uh so if you are a citizen from a certain country and you want to participate with your specific language uh you are welcome. So don't worry about it.

SPEAKER_03 27:59

Yeah, that was the thing that I want. Thank you very much. So no limitations, actually. Thank you. Uh no limitations, yeah. Okay, good. Thank you very much for being a part of our podcast. We wish you really good luck, scale up, uh, be a unicorn soon. So everything the best to you and your startup.

SPEAKER_02 28:20

Thank you very much.

SPEAKER_01 28:24

Thank you for listening to this episode of The Access Smart Talks Podcast.

SPEAKER_03 28:28

We hope that you enjoyed this episode and can't wait to share more with you next time.

SPEAKER_01 28:33

You can follow us and find more episodes on the Reaccess Portal, our social media profiles, or wherever you listen to podcasts.

SPEAKER_03 28:41

This episode was prepared by hosts Daddy Aksandov and Yongelovich editor Easter Pearce and is dedicated in the memoriam of Professor Miroslav Trajanovich.