Drive your life is a truly special opportunity. Eric is on the road for 8 weeks, exploring beautiful places and working from them. However, it is not all he is doing, he is also able to meet with exciting advisors, researchers, and entrepreneurs. One of our recent events put him face-to-face with Sachin Gaur, a senior advisor for NORA, the Norwegian Artificial Intelligence Research Consortium.
Here, you can find their exchange in an interview format:
Eric We’re here today with Sachin Gaur from NORA, and also Thimm Zwiener, who is joining on this drive your life trip, and I’m Eric McDermott.
We’ve just heard a bit about NORA, and all of the breakthrough kind of Norwegian technologies and companies emerging right now and in the pipeline. You’ve given us a little brief overview, but maybe you can introduce yourself and talk a bit about what NORA does.
Sachin Yeah, great, I am working at NORA in the role of Senior Advisor. It is extremely rewarding because I get to talk to people on the cutting edge, applying AI across topics from healthcare to climate change. It’s very rewarding in the sense that in a role like this, learning never stops.
From NORA’s perspective, it was created to coordinate AI efforts in Norway. It has behind it 8 public universities, 3 university colleges, and 4 research institutes. Nora constitutes like a national environment, and of course, as the name suggests, it surrounds research, but the research must lead to innovation. NORA also engages in a variety of topics: in practice we try to help startups which are doing research in AI. We also have an EU network with which we help our stakeholders get more EU funding. We are also involved in increasingly more and more initiatives, which we think can help with the Norwegian efforts. For example, something is being built right now, it’s called the Norway AI cloud. There is a need because, AI models are not small anymore, and become increasingly bigger. New AI models have 500 billion parameters. We were successful last year installing some of the large models on this infrastructure and providing it to researchers as a service… they don’t have to worry about the infrastructure but just do what they love.
Thimm That’s cool. Do you have to apply for these? What is the process behind it?
Sachin When we are talking about research, we are mostly targeting PhD students, that’s our primary customer here. But of course, any researcher is also relevant. When you’re doing PhD research today, you may face a lot of bottlenecks, infrastructure is one of them, dataset is another one, and if you are applying it to a specific domain, then you may not be a computer science expert, but you may be a genetics expert, and so and so forth.
We also need a cross-disciplinary approach. That’s where we are trying to create an intervention between colleges, universities, and research institutes; where if you are a part of them, the membership is free. We also created NORA startup. Any Norwegian startup can apply, as long as they are having a research co-interest in their model.
Eric Wow, it seems like you guys are involved from the start to the finish, as you said, going from research into industry and really supporting people along this entire timeline.
Sachin Well, yes, but I don’t know if I could say until the finish. Finish means that the company is successful. I think we have more on the enablement side, we are quite nimble in that sense. Norway is a small country population wise but it’s quite large geography wise. People are in different cities; therefore, you need an umbrella framework for them to understand who’s doing what.
But also, when we are talking internationally. We have recently done a collaboration the Helmholtz Information & Data Science Academy in Germany. They have 18 research institutes on different topics, with a 5-billion-euro budget. These are the kinds of relationships that NORA creates. We currently collaborate with three German researchers as we are talking in Norway. And we have just kind of finalized a similar exchange program with Alan Turing Institute in UK.
That is something that NORA really likes to kind of facilitate: how can we exchange researchers with the top AI environments in the world? Hopefully, in future we will do some more collaboration. Research within the AI context cannot do alone, and you need to put the best minds together. For example, some of the problems that we’re facing today are planet scale problems, climate change – like your friend Thimm here and what he’s trying to do – I think people in the end need to join hands. I think that’s the kind of spirit at NORA: how can we join hands with relevant stakeholders.
Eric It’s kind of a coincidence that these AI models are becoming larger and larger and needing more and more parts and so are the things we’re trying to tackle with them. As you mentioned, with this rapid climate change, the source problems causing it to contain more and more parts, and so we need better tools and more and more people to come together to work on solving it.
Sachin NORA is also working in some of these areas, trust areas, if I may say, for example, you have a couple of billion euros in this project; called Long Horn project. We’re trying to take carbon dioxide out from a cement factory and store it in oil wells.
These kinds of projects require political will and money, right? I think Norway has some good possibilities. The strength areas in Norway, I would say that Norway has digital data for the last 50 years when it comes to healthcare, for example, so if you want to do good research in healthcare, you need longitudinal data, you need to understand what happened to the patient or the person involved. You cannot overnight produce this data. This society is lucky in that sense in that it has this data over the last 50 years.
So I think about healthcare, climate change, think about other topics like maritime, fisheries, all this required sustainability for the future; I think there is a mission driven approach here that is needed. And I think that’s where we need bright minds to come and solve those challenges and problems.
Eric Certainly. This cyclical process that you mentioned with the harvesting of carbon from cement factories and relocating it back into oil wells reminded me of your project, Thimm, with this regenerative agriculture approach trying to regenerate soils.
Thimm Yeah, one big part of regenerative agriculture is exactly that, we want to sequester carbon in the soils. We can do that with plants, and various plants do it in different ways. It is a less technical approach. But nevertheless, like it’s something that needs to be measured and needs to be measured at scale. I think methods like machine learning and remote sensing will play an essential role.
Eric Definitely, I think leveraging these technologies will help with this – and on that note, do you currently see any kind of emerging trends, since you have this kind of overview over many different departments and initiatives?
Sachin AI models are becoming larger. It is becoming very difficult for any individual to make an impact unless we make concerted efforts to make it easy for an individual to contribute. Leading research today is happening in companies, not in open society. As Universities or as democratic societies, we don’t want our future to be controlled by select few. We want it to be open.
That’s where I think these EU-wide projects are important because not every individual European country may have sufficient weight in this domain. HuggingFace, for example does a lot of these Open-Source models. They received EU support, to go and make open models, and they have now released those open models!
Now, I think we need that political will, and we need that leadership at the highest level. Emmanuel Macron together with Yann LeCun for example had a big discussion panel at VivaTech. We need that kind of reciprocation from every European leader. Once it becomes a political thing in Europe, I’m sure that it will provide motivation to citizens, to think about the big problems and how technology can help?
Eric Yeah, and I guess this whole movement of AutoML is based upon this enablement of everybody, every kind of scientist, or so to say a normal citizen can use these powerful machine learning and AI models to uncover some truths about their data, or even the world.
Sachin It talks about the values we have in Europe, not just do we develop these things, but we have to develop them responsibly, particularly trustworthy, is the key word here. I think that requires a larger participation. So only a select few cannot decide, it requires a bit more consultative, and participative approach, and that’s why it needs to happen. The cutting edge research should not happen behind closed doors, big companies, maybe we don’t know what they have built, but it should happen in open research.
Thimm And what are some cool projects that NORA is financing or facilitating?
Sachin We are mostly facilitating, for example, last year, we installed this algorithm called AlphaFold. It was a 50-year breakthrough in the field, I believe somebody must get the Nobel Prize for this. Since 1972, it was a standing problem to predict protein structures from the sequence of ATGC, and say how this protein would fold. This was a Norwegian American, Nobel Prize winner, Christian Anfinsen, who articulated this challenge. And everyone or rather two years since then, there has been this challenge called CASP (Critical Assessment of Protein Structure Prediction), where scientists try to predict the molecule structure with mathematical or computational approaches. And in 2020, AlphaFold won that competition.
Then there was a lot of pressure on AlphaFold, because the world wanted to know how they did it. So, in 2021, in August, they released the code as a paper. But not just them; there was a woman from University of Washington, from Institute of Protein Design, Minkyung Baek, she also made her own model inspired from AlphaFold, called RosettaFold. What a big company did, was replicated by an individual researcher. Once the code of AlphaFold was released, we thought it’s an excellent opportunity to provide it to everyone. it was around 3 terabytes of data. It took us five days to get it.
Thimm The model itself was 3 terabytes of data?!
Sachin Yeah, it has a lot of dependencies. Protein databases is huge. We downloaded it, we installed it, we reached out to some of the people behind CASP. And we said, we want to organize this workshop. This was the first open workshop on this in the world. And overnight – you know, we were thinking we are doing like a smaller Norwegian thing – with five or ten people here trying to discuss what I proposed, and that will give them a service, they will type some sequences, and they will be allowed to see how the protein structure looks like. But we had 700! Some of the top minds in the world participated in this because we were the first ones to have this kind of workshop.
Thimm Wow, so is this algorithm still running? What are the requirements – say as a PhD researcher – to work with it, what do I have to do?
Sachin You don’t have to do anything, you just have to write an email to us, and, you know, as I said that most of our services are free for Norwegian students. It is just that they have to know that this exists and that’s our job, to make people in the life sciences, for example, familiar with our services.
But tomorrow, if something else comes up, even with our capabilities, to set this up, it may take us weeks. For an individual PhD student, they will spend a lot of time not doing what is not actually adding value.
Eric Right – the actual research. So much time would be used to set up their tools and methods, and here you’re providing that service. That’s awesome. And is it also available to non-students?
Sachin Right now, we cannot really subsidize other people’s work. From, you know, a competition perspective. We can’t start going out, pick up like Amazon or others can do from a commercial standpoint. We are restricting ourselves to people who need it most, which are PhD students.
Eric And also those who give back, because the work they’re doing is often published and then open to the public.
Sachin There are also startups which reach out to us, and we are more than happy to help them. Sometimes they will not have capabilities, normally we then help them with the computation side of things, but more importantly there is a PhD student which can help them compute those structures which they need for their work through important research.
Eric OK. Just going back to this trend question… on the one side we had this kind of bringing AI back, say, out of being exclusive for larger corporations and rather toward the enablement of researchers and startups and so on – but kind of on the other side of that question, we have more-so the trends or innovations with respect to topics. Do you see anything very exciting?
Sachin On the creative side, we are all seeing Dall-E. Definitely affecting your kind of domain.
Eric Yeah this whole kind of synthetic data generation. Dall-E is a model where you can just type in something and it creates an artificially generated image.
Sachin Yes, and in the natural language processing space so far, I can say when a big company builds a model, they focus where their market is, so you will not see a language like Norwegian as part of these large models. Which means that it’s dependent on a particular country, if you take a country like Finland, Sweden or Norway, first thing from the AI perspective you see of strategic influence is if the AI models are having their own languages which give an equal opportunity to the population which lives in this region. There isn’t so far.
If you depend on a large company, which maybe five years down the line when you’re relevant from a market perspective, right. But recently, Facebook has released a model that has a lot more languages, they call it No Language Left Behind, and it supports Norwegian for example.
As an immigrant, for example, it impacts my everyday life. You know, could I translate, spoken Norwegian, in real time into the language that I prefer? I think the world we are getting into, which is hyper connected, where you could chat with anybody in your own language, I think it’s a big enablement. And that’s where the question we have to ask, is that who should decide which language should be there in the model or not? What are the biases that we encode in those models?
A researcher from the Alan Turing Institute for example was showing it to us that if you translate language from Turkish to English, any positive word will be translated with a male bias, and if you translate anything which is negative, it is translated as a female bias. Nobody is knowingly building these biases, but they exist.
Eric Yeah, but they’re very important to look out for –
Sachin Yeah, imagine one of you is talking to someone in Turkish, and unknowingly, your translations are getting goofed up. You can imagine how unpleasant that could be.
Eric Yeah, this whole concept of Safe AI, Ethical AI, is also an extremely important and relevant topic.
Sachin Yes, so we need to do investigations there. And I think, the Cosmopolitan magazine, did a good AI-based magazine cover using Dall-E. And it was a female astronaut. So, recently a professor here from Berlin, she wanted to get an image generated for Sally, that is a female professor with a skirt, with a, uh, view from the bottom, okay, so you can imagine going to the gray zone. So that’s the thing that you must know, if this kind of thing gets into the wrong hands, what kinds of things people will do with it.
Thimm With respect to Dall-E, I’m pretty sure they already have some filters and so on already running on there, as I have not seen any Dall-E picture which contains nudity.
Sachin I’m 100% sure what you’re saying is correct, but at the same time what I mean is that we need investigations which are open where we can learn more about these things. There are now guidebooks coming out where people are trying to summarize it.
Thimm And you don’t want to have these filters owned by companies, you want to ensure that the filters that are used have the consensus from society.
Sachin I think these things are easier said than done, and I think everybody has a different starting point. Norway may have a different starting point, Germany may have a different starting point, Kenya may have a different starting point, with different requirements, cultural values, how you look at things.
And I think where we are now is that the AIs that we develop are very homogenous. It also becomes that you can enforce a lot of your value systems with code, because what you end up exporting is a lot of technology from one part of the world to another part. AI will play a role in how we perceive things.
Eric Definitely. Even with this example you gave previous the interview with the company that is creating AR overlays on sports games or applying advertisements in the language of the television viewer in real-time, I mean, these are also subtly changing and influencing the way that we perceive our world and our environment in real-time. AI will certainly have a bigger and bigger impact in the years to come, and the spheres of influence will expand.
Sachin Initially, when messaging tools were becoming popular. The emojis that you sent were mostly in white skin colors and now these days, if you see emojis they are diverse and you can choose the skin tone.
It’s a gradual process by which, just based on the starting point, so if you take an AI model from America, maybe it starts with that value system. But if it’s able to appreciate the nuances of a culture, then it can adapt better. That’s where we need this kind of a healthy dialogue between society and the developers of AI, so that it is closer towards the society needs, not what the developer builds.
Eric Yeah, just becoming gradually and gradually more inclusive and aware of all ethical concerns that arise with these powerful techniques.
So, Sachin, I think we are almost out of time here. So let us just say thank you, for tons of insight and good conversation.
Sachin Thank you and I wish you the best on your trip, and especially the time you spend in Norway.
Thimm Thank you, and we are already enjoying it!