Join the conversation:
The third and final episode recorded at the AI and Big Data Expo, that we produced in partnership with Exfluency, the AI driven translation and localisation system. The event took place at the RAI in Amsterdam on 26th-27th September. We recorded a series of interviews from the Exfluency booth with a number of the speakers and attendees at the event. Our guests for this episode were:
Aoibhinn talked about the key points that were made on a panel that she attended called ‘Building an Augmented Workforce’. She said that building an augmented workforce is exactly what she does with clients. She works in the public sector, and can see that many of these organisations are struggling to do what they would like to do for citizens with the amount of people that they have. These clients are coming into this question like, how do we use AI from an efficiency standpoint? Pretty much like most organizations, they are wondering and people in the audience are wondering, there’s so much how do we understand what will actually work for us and how can we do it in a secure way, and how do we find out what will be the winning use cases for us that we should invest in? So, it was very interesting. She summarised her thoughts as a lot of the questions that came from the audience were like, Oh, the speed is so high, and should we be scared? What happens when we automate everything? And the panel offered some nuance there where we do recognize the complexity that these technologies bring alongside them. So, it’s a bit different if you’re talking about like out-of-the-box solutions like Microsoft is integrating co-pilot. So that might be a bit more straightforward to implement, although you would need to bring your organization along to some certain level of AI literacy. She then said if you’re making these custom machine learning solutions, there’s just a world of complexity that comes with that. So, it’s not something that you can do in a day. So, yeah, that’s what we talked about.
Aoibhinn explained her outlook on AI. She discussed the great question are we going to be automated away? And she thought that kind of makes the assumption that we’re going to use AI to do everything we’re doing now, and then humans become irrelevant. Whilst what her understanding is her clients, they’re not necessarily meeting the service delivery that they would like to. So, for example, citizens might have to wait several weeks to get certain services. So, there’s pretty well defined way that we would like to spend humans time if we didn’t have to do kind of machine work. So, she doesn’t see AI as necessarily as a big threat for today or tomorrow. But then of course you can go into the sci fi future where we can automate everything. She thought that will be very complex to rule out but then she considered what if that did happen do we all need to be working 40 hour weeks would be scale down could that be an option?
Aoibhinn talked about how she thinks AI is going to change our working lives, she exclaimed that she doesn’t like it when people make these big declarations and know exactly what’s going to happen in the next few years because everyone is seeing these big speeds and innovations and taking different turns every few months. So she decided to talked about what she sees based on the present. She said AI being used as a productivity strategy and how can we use AI to support these people to deliver that service and what she was interested in is foundational models like GPT 4, for example and how we are democratizing the way that AI can be used because until recently when you would have needed massive amounts of data, large resources to train these models. And now they can be used as foundations provided by big tech companies and then fine-tune them for our own efforts. Therefore she said that will accelerate AI adoption into like not necessarily tech companies and she said the most exciting things are in the public sector related to personalisation.
Aoibhinn said there is work to be done in terms of progress in increasing female representation in AI. Basing this off of World Economic Forum’s report, they make annual reports based on LinkedIn data. So, in 2022, we saw that 30% of AI talent were female and women’s representation in high-level leadership roles, like we’re talking VP Csuite that drops a lot lower. So, you’re seeing about 13%.
She then explained a way to improve this low number. She said AI and machine learning models are nonbiased as data goes into them but the industry that is creating these AI models is not diverse. She also stated that teams with diverse perspectives perform better, they challenge assumptions. There is a lack of talent pool in AI therefore we should be investing in a diverse group. She said there is a rise in women graduating from STEM university courses and entering the workforce but then there is a retention issue.
She explained at women in AI they try to improve perception and recognition and representation. So, they partner with this organization called Equals, and they do this role model campaign in Amsterdam. So, Women in AI have these massive posters and they put women with really cool jobs in tech and AI so that they’re spread out through the whole city and people can see them, so they exist. And they also do our Women in AI Awards, so these women get recognition and, they get a light shone on them and they have our mentorship program. Regarding the issue with retention she mentioned earlier Women in Ai try to combat this by having mentor, mentee relationships in all stages of a career, so they have students who are being mentored by women who are like in their early phases of their career. But then there’s also more senior women mentoring women that are struggling to find like, what’s the next step in my career? And they also have an Activate program, which is they go to areas in Amsterdam at the moment where you don’t see that many students, like high school students, flowing into these STEM bachelors. So, they take the kids and they conduct workshop with them for the day and they are partnering with ING and Schiphol on that. They also do monthly meetups with groups of women and have mini lecture.
Michiel explained model lifecycle as basically the beginning of a model to the end of the model and ING are part of that process. A model has been developed by ING on a local level or a global level and for ING it’s part of their job to determine where the key risks are in that model. He explained that the model has been developed in 2 to 3 years so how much further will the mode be in another 2 to 3 years. ING’s goal is to see if there are any risks in this model and identify the ones they need to to make the bank safer.
Michiel talked about why it is so important to have model validation in a bank. He said it depends on where you are in a bank. ING has a lot of models and the scope of these models also varies quite a lot. They have advanced analytics models which are very much related to machine learning and they also have credit risk models related to the amount of capital that they need to hold. So it really depends on the domain. But for example, for us, the most important thing is to make the bank safer. But you can also imagine an external regulator like the European Central Bank is also going to look at what we have performed right, how safe we are as an organization. So it’s also important to take a stakeholder like that into account when we’re doing these validations.
Michiel said they have nine validation dimensions at ING and therefore they look at nine different aspects of a model and that includes looking at the data. They check the quality of the data and look at the model design and the design choices such as what type of features you have created, how you come to the outcome of a model. And they look at the output and look at how well the model performing, does it align with their expectations. They also look at the limitations of a model and s sometimes they build an AI model just to check does this model perform well and would an AI solution be better than this business experience?
Michiel talked about how using AI can make a bank better for example customer interactions and targeting the right people, and then using AI to target specific customers advertisement. These are parts that fall into Non-regulatory space. He also said they have a little bit of freedom to also use AI to make our processes better, to make the bank better. From their perspective as model validation, they also use AI to challenge the more standard models. An they use AI to challenge these models to see how well they are performing and how well the design has been made. Michiel expressed KYC is a big space and they also consider anti money laundering, fraud and anomaly detection models which look at how suspicious each transaction can be compared to others. AI systems try to reduce the amount of false positives there are from these rule based systems enabling them to only need to look into suspicious cases.
Guy talked about atNorth and described it as a Pan-nordic data centre provider that is active in all Nordic countries in Europe. atNorth build data centres for the highest performance computing and artificial intelligence as well as they have built a whole stack of infrastructure as a service on top to run AI and HPC applications in the most sustainable way you can imagine. Guy explained the key advantages of the Nordic region for these data centres. He said the massive amount of renewable energy that is available there, but also the climate is ideal for data centres that consume and run and consume and produce a lot of energy. So, they are actually way more efficient than actually in other places in the world, in Central Europe and the US. And that’s why clients from all over the world are coming to Nordics to run their heavy workloads, their back-end calculations like processing of large language models or AI applications or big simulations that come and run them in the Nordics.
Guy talked about a US client called Tomorrow.IO. They are an extremely fast weather forecasting company. They have also launched now recently lots of satellites. So they collect data from the satellites from all over the world. They let them calculate the data in Nordics, in Iceland and in Sweden. And so they then expose them, the results of it, expose it back over APIs to applications all over the world like Uber or airlines. And so that’s one example. Another example is BNP Paribas, one of Europe’s biggest banks. So they bring their massive risk calculations and massive calculations again Sweden, and Iceland primarily to achieve their sustainability objectives. So they were running it previously in France, UK, somewhere else in Europe, but they have abandoned that and brought these heavy workloads to atNorth to help them really achieve primarily their sustainability objectives, but also to get more compute for the same money and to get more done with the same expenditure.
Guy then talked about the data, the time and the expenditure needed to fuel all of this innovation. He said there is a lot of initiatives in the world of training and retraining large language models and foundation models to really build generative AI. And it’s an extremely promising new domain that make companies, enterprises or users, way more efficient. So they bring first these large language models to calculate. So they are extremely complex mathematical models. If you look at GPT 3, where OpenAI has based ChatGPT on, that alone, that takes like 10,000 GPUs for weeks, months, to just calculate it once. And that is then calculated based on information that comes from the Internet. But then when enterprises start to use this, they don’t want to use Internet data. They don’t want to have the data from their own enterprise exposed to others. So they want to run that then in a private cloud. So they want to set that up as a private environment. So, they train the models on their own data and then make their own language models out of it for their own use with their clients and for their customer support. Health care companies are doing that. Banks are doing that. So, it’s a massive amount of data, a massive amount of compute that needs to run, get the most out of it for the investment. But it’s like a rat race for getting most intelligent models with all the newest data and the most privacy data that comes to us.
Guy expressed how managing all this data can be done sustainably. He said people go to just the public cloud and run it. That’s good when you have a small test and when you have no idea where it’s going to end try it out. But when it becomes very massive calculations. A, the, the public cloud is too expensive, but also where does it run in the public cloud? You actually don’t know. It’s like Amazon Azure. Whoever decides where it runs, it runs in the data centres where they are. It’s somewhere else in the world. So, it’s not done in the most sustainable way, but it’s also not done in the most economical way. So that’s what atNorth try to achieve with their clients is they want to get the most out of it, but when also to do it in a sustainable way so that they don’t jeopardize the climate and don’t jeopardize the future for kids and ourselves.
Vidya explained the three horizons of data, he said they are essentially using your data to maximum today, whatever data exists today. If you look a bit a little bit out, one, two, three years down the line, then you want to probably start thinking about, hey, what are the new data sources that I could make available? And this could be existing data sources, but just in bad quality or bad shape. So, the next horizon is actually working on those data sets and making them useful over that horizon. And even beyond that, it’s all about creating data by systematically digitizing your processes. So there’s plenty of processes today in businesses which I mean are digitized. Maybe they don’t happen on pen and paper, but they actually are not systematically collecting their data. So, if you’re looking at a horizon, which maybe is around 2 to 5 years out, you want to actually start thinking about, what are the data sources that I want to create for my next foundation for digital transformation.
Vidya then talked about the importance to why he works on these three horizons. He said if you think about the second horizon, which is all about improving data quality, it does not happen overnight. You need to get activate the organisation; you need to activate the leadership. You have to sort of provide transparency into where the data quality is bad. Maybe even prioritize where you want to improve data quality. So, you got to invest a lot of time of the organization in doing that and you have to activate all this machinery. So, if you want to benefit 1 or 3 years out, you actually have to start today with it. So that’s with respect to the second horizon and the third horizon is equally well, if you want to create a good set of data, good quality data, you have to start collecting it for some time. So once again, you have to start now.
Vidya then pointed out the challenges of bringing the three horizons together. He said one thing he typically observes is very often you need three different kinds of skills to work on the three horizons. So if you think about the first horizon, you’re looking at the classical DNA kind of resources. So you’re looking at cloud engineers, data engineers, data modelers, data scientists, these kind of people. And that is what people think about when they think about data analytics today. This is by far the composition of DNA teams as well. If you want to work on the next horizon, you have to think about digitizing processes, which actually is a more traditional IT skill, but it’s not necessarily available in the same organization. And the third bit, which is all about creating data for the future which we tend to use things like Power platform, which is kind of a very low code, easy way to digitize your smaller processes. And once again, that’s a completely different skill, which is not really available within the DNA community. So I think the biggest problem you have probably is co-location of all these skills, and we were fortunate that we could do it in our organization because we started from scratch with our digital unit. But in many other organizations that’s not the case, very often you just have the DNA people involved. Even if you look into the IT organizations, they are actually also siloed in a very similar way. So, in IT, you’ve got three different departments and you’ve got to actually work with three different departments if you want to address three different issues. Frieslandcampina were fortunate enough to actually start from scratch, Vidya explained. This actually shaped their digital unit, thinking about these three horizons and how Frieslandcampina want to work in the future
Vidya talked about what more he hopes to achieve at Frieslandcampina. He said when they started a digital journey he started with the ambition that they want to improve speed and efficiency of their R&D organization. Over the past three years that is what they have been working on. He said they are essentially working towards the set goal that we set about for ourselves, which is to basically increase the speed and efficiency of their R&D.
Rowan said sustainability is a very big driver because sustainability is a hot topic. He mentioned the Paris Agreement, and that companies want to save their energy, save money and that kind of stuff. Rowan talked about how his company help clients achieve this and it starts at the bottom line of the question because you want to achieve something. But in order to achieve something, you have some questions. You have some KPIs, some key performance indicators that are important for these customers because you really need to know what kind of ideas a company has in order to support them in their way of sustainable operations.
Rowan expressed that analytics for Industry is a company he works for, and they are basically a company that supports or facilitates manufacturing companies in order to become more sustainable. So, in order to become more sustainable, they deliver dashboarding and try to have a better look on the data because the data sets need to be of good quality. You have a good quality data and, based on that, you can move forward to these KPIs of sustainable operations.
Rowan said to be able to achieve this for clients the company has created a standardized energy monitoring system. So this is basically a full-blown package with four sets of dashboards that has some insights in how much energy you consumed per batch per product, per line, per asset, but also per work order. It’s very important for different organizations because you have the process data, the sensor data, but you also have the information about relational data. And those two needs to be contextualized to each other. And that contextualization is happening before dashboards. And that’s something they try to deliver to the customers. He gave an example of a current client which is a food manufacturer in the Netherlands and they have two questions. One question is about overall equipment effectiveness. So how can we improve our operation by availability, performance of my assets, but also the quality of my products. But also the other side of the story is energy monitoring. They really want an energy monitoring solution to be able to meet their energy goals. For example, they want to save 2% of gigajoules per milk because they try to save more money on the input side of the factory, for example.
Rowan said AI is often being picked at the conversation because AI is a buzzword. People really want to have AI, they want to predict their maintenance, they want to predict their energy consumption over the next decade. He said honestly and realistically you need to know what you’re analysing and where it’s coming from. And that’s something the energy monitoring solution for manufacturing, offer is a starting point for this journey, for this data journey of this customer.
Giuseppe talked about some of the key points that he made in a panel earlier on in the day. One is about how we make data available for our business consumers and how to make most of the data touching one point, which is data democratization and how data should be made available for people that are not tech aware. And what about IT? What roles do we have in the IT organization to enable that? Another one quite important is how to break data silos that are currently existing in your siloed company, which means how currently you connect to your source systems, how you process that data into a consumable format for the end user. And what about the silos that you are creating right now in your company? And the third element, which is also important there is how the data silos and data democratization have a saying on how every company should be pursuing, energy transition and actually collaborating towards that global purpose that we have. So, he touched on those three elements, and he had a quite interesting contender, which is Foot Locker and Van Oord. So, Van Oord is about ships, and we have 108 of them. If you don’t know about Van Oord, just Google the Palms of Dubai and you will find that’s us. We go with the ship, take sand and build interesting stuff. But Van Oord also build wind farms at the North Sea. So, in a way, they’re collaborating towards this transition in a way that they are making sure that every new element of energy is actually based on helping the climate to not be where it is right now and where it’s heading, which is quite bad.
Giuseppe talked about data-driven Marine ingenuity which they have at Van Oord. Marine ingenuity is about the value that we create for our customers. And data driven means a lot of cultural behaviours that consumers have. Data-driven has been just a hot topic across time and right now it’s shifting in a big way in the sense of the following. BI systems, business intelligence systems are right now shifting into another state, which is the augmented consumer way. They think about things like ChatGPT integrating into Microsoft PowerBI products and ChatGPT actually being the developer for you. So, you have now the ability to prompt, if you prompt well, and actually get insights on that data. And that means a change in the behaviours in the culture that we should expect as you as a manager in order to query and get data. So, data driven means not only knowing how to query your data, prompt it, but also, in this new terms, means that we need to make sure that people are trained in a way that they know what is fact what is a dimension, and perhaps tell them how to create proper context, proper questions towards querying those BI systems that we are providing. So, data driven touches a lot of points and 80% is the human part and 20% the technology that we should be making available for this new environment. One of the things that he touched in the conversation was, okay, we have a lot of data pipelines, 100 of those connecting to an ERP system or CRM and so on. But right now we have technologies that are bringing all the data into one lake, so it’s not anymore the chief data Officer is actually chief integration officer. That’s currently what his company have been doing in the last ten years, connecting systems and making lakes to create a relations on that. But it’s more about, okay, that not a problem anymore integrating data, what is the problem now? How all that data it’s used and now it’s the time to think about how your company is going to evolve in the cultural aspects so that people that are not technically aware are able to get into the data in a proper fashion. He expressed we have a gigantic leap to make, which is, How do you set up a proper behaviour like agenda to train the managers how to be precise if they spot data quality mistakes, raise the red flag. Number one element on data driven behaviour. Number two, if it says 53, why are you going right? You should go straight. There’s a factor of people just not relying on their opinions. And in that sense, we need to evolve in a way that data now is accessible. Now what about what you would do in the behavioral part?
Giuseppe went on to talk about LLMs large language models. He said the technology is enterprise made available, not there yet. So, if there is a company already doing that, then he thinks they went into the preferred channels of the main providers call it ChatGPT. But if you go there, one of the things that he had a conversation with Microsoft, which is really inspiring, is that you have the infrastructure on how ChatGPT works is so massive that you cannot host it in your company, only the data centre of Microsoft. But what if you are allowed to actually have your own LLM on the company and train them with the knowledge of your company in the context of Van Oord we take the knowledge that they have of 150 years for 108 ships and based on this knowledge, let’s train this LLM so that it makes proper conclusions about atmosphere, about current, about waves, about how to create a dredge the sand or the soil from X to Y. Actually, LLMs are going to be vastly at hand and hopefully they can go into the preferred channel because then people are enabled companies like Van Oord to train their own LLMs sourcing, all the infra from huge parties like Microsoft and start to implement those as the real context for every user.
Gary Talked about how he at attended a lot of these sort of events over the years and it is amongst one of the best he has attended. He expressed how much traffic the Exfluency booth has attained.
He explained that Exfluency’s main aim from attending the event was to introduce Exfluency to the market and make more people aware of what Exfluency can do for them and they also attended to learn. He expressed the intelligence that people hold at this event and interactions he has had over the past couple of days have proved to be very interesting. He also expressed his happiness of how well his colleague Jaro’s speech went down with this particular audience. He called his colleague a bit of a rock star after hearing how well he did at the networking party on the first evening of the event due to his experience in the sector and lots of like-minded people wanted to have a conversation with him.
Gary explained the challenge for a company like Exfluency is to get your name out there. He said once people understand what it is Exfluency can provide for their businesses, they’re genuinely interested. It’s a very expensive thing to be able to get your name out there though, and you can meet so many potential buyers in a very concentrated area and time if you come to these sorts of events.
In partnership with
PRCA members receive 10 CPD points for listening to this podcast if they log it on the CPD programme.