Published on February 08, 2022 --- 0 min read
By Clearbox AI

How to harness the power of AI in companies

Share this article

Have you ever run into Google’s ‘People also ask’ section and found the questions you were about to ask? Periodically, we are going to answer the most researched questions about AI and Machine Learning with our guests on Clearbox AI’s new interview series ‘People also Ask...and we answer’. Enjoy!

In recent years we are witnessing the rise of AI in many different situations of our daily life and this also relates to companies. Nevertheless, the latter is still an area of high concern for many reasons. So, what does it mean for companies to adopt AI in their businesses and how can they take advantage of its power? We will discuss this with our guests Shalini Kurapati, CEO and co-founder at Clearbox AI, and Marina Geymonat, Head of Innovation Lab at Sisal and AI Expert for Italian Ministry of Economic Development (MISE).

Introducing our guest

Marina is an information scientist and she has a lot of experience in the use of Artificial Intelligence to solve real problems in companies. Recently she joined Sisal as Head of Innovation Lab, a brand new department created to foster innovation in the company. Before that, she worked in the innovation department at Telecom Italia for almost 30 years with a focus on data, data analytics and especially on AI.

Why is artificial intelligence important to business?

Marina: Based on my experience I think that there are at least three directions in which AI can help businesses. The first context is probably the simplest, and it’s where you can use very simple AI algorithms that can help taking simple decisions which are not so critical and that humans take with no neurons almost but they have to do it, so it goes strictly coupled with automation. There is another way that AI can help, and it's a little bit more difficult, and it's doing things that humans are not able to do. That is especially when we talk about companies that have a huge amount of data and no human could ever make any good decision, even less in real time but also every quarter of an hour, or hour, by looking at millions of different very small data that make no sense to a human person. What we do now is help people with very smart visualizations so that they can immediately catch what's happening and make the right decision. This is where AI really comes into play: if you can automate the understanding of what's going on in this very huge and complex amount of data and explain it to humans, so that they can take the right decision, or embed the decision to take in the AI algorithm, that would help a lot to achieve a much better operation, a much better decision. Another example is very based on visual search, so we talk about a kind of anomaly detection that typically is done by visual search, visual analysis, and it can also be done by checking the amount of data to verify if there are anomalies. Here AI can help a lot as well, because you can check and verify many products, animals, people or whatever, and verify that everything is going well or if there is something wrong. Lastly, we can find speech recognition, which is useful for support. Customer care is one of the most used contexts where this can be applied, and you can understand what's the customer problem or question automating it and solving the problem more easily, or getting to answer the question with real data taken from the real world.

Shalini, do you think there are business sectors that need AI integration more than others?

Shalini: Now companies have massive amounts of data and what AI does is, it makes this analysis or extraction of information from data much more efficient, fast and often in real time if done properly - real time is still a tricky word in this business -. If you look at the energy networks, they have massive amounts of information and also you have new players coming in with smaller grids, so energy consumption is definitely a use case, I would say. And also the banking and financial sector, and healthcare is definitely another fertile ground for using AI. And drug discovery is another field because what we know is that we have immense amounts of information, there's so many permutations and combinations about which medicine would work for what and they're also trying to use AI for drug discovery. So in terms of fields, these are the ones that I've already experienced in one way or the other that are making big gains with AI.

How does AI adoption work in companies and what are the requirements?

Marina: The first requirement is to have data. And then there are technical requirements such as having the data in a form that can be put together, can be queried, can be explored. I think that if you have your business plan and you want to go in that direction, that's not the big problem. The big problem, first, is to know what you want to do with AI. You have to know what you want to achieve: you want to make predictions, you want to make classifications, you want to automate, you want to…whatever. That must be very clear and it seems easy but it's not, because whenever you want to do something using AI, you need to express your requirements in terms of what can you extract from your data, with algorithms, any stuff, but that must be in the data: it cannot be injected by people's mind, or semantics, or common sense, that's what's really lacking there. So you need to have everything that you want to achieve, that must be hidden in the data that you are using and that I think is the most important requirement. Another big one which is on another level, it's an organizational level, is that whenever you want to apply and have benefit by an AI algorithm, you cannot limit your design of an AI system to a single department. Even if that's the department that will benefit from the AI solution, normally a good AI solution must be designed taking into account a lot of different departments that will have to put data, or have to put knowledge, or have to put their point of view in designing the system. In my experience this is one of the most important requirements on the organizational level: to be able to work together and have a very heterogeneous table of people designing this AI system, so that you don't miss any of the key points that you want to put into artificial intelligence.

Shalini: Yeah absolutely, you really hit the nail on the head about the communication and the organizational impact, because it's not just a technological solution. There's a lot of change of mindset required, and first of all what's very important is to set the right expectations: what can AI do for me at the moment? Because sometimes it can turn out to be some kind of vanity project where we say “AI is going to solve everything”, and that's because everybody's talking about AI so it must solve all my problems. That's a bad attitude to start with. So, to have the right expectations, the right infrastructure, people, skills, and most importantly data: do I have enough data to make sense for this particular problem?

What can be the most significant barriers to AI adoption and efficiency?

Marina: I really agree with what Shalini was saying about expectations. This could be a big barrier because you start a project and then after a short time people are already expecting whatever they have in their minds. It’s an advantage of AI that when you set the system, the system works at some acceptable level, but then the more you use the system, the more you give feedback to the system, the more it should improve its results. So the only way for a system to really flourish and reach the best results, is to use it, it’s to have it in the wild because that's the only way that real data can come, and then feedback can be given, and the system can enhance its results. Another very key point is to understand the results. It seems simple for those working on AI to understand concepts through which we deliver results, such as precision, recall, accuracy but also lift, there are many specific words that can be explained very easily - well not now here - but of course a few hours are enough to understand, or a few days of full immersion on AI are enough. That would bring an enormous benefit to the AI world, because in this way people could really understand what I mean when I say "this works at 99%": what does it mean? And then, like I already said before, the requirements can turn into barriers, such as the problems of compliance, of privacy, so the need to access a lot of data that must be checked against all the other structures in the company that take care of all these legal issues which are really important.

Shalini: I like the way Marina said “I would like the models to be in the wild” and not in an incubator somewhere. So being in the wild is being in production, being in an operational environment, and to get there, there are a lot of issues, and one of the biggest issues is the data quality and accessibility. Everybody has this attitude that "this data is mine, I won't give it away" even if it doesn't have privacy sensitive data, it might not have any sensitive data sometimes, but they're like "oh why should i give this data to somebody else". One aspect is definitely privacy, so you have GDPR in Europe you have the California Consumer Privacy Act in the US, you have the HIPAA for healthcare in the US, so you have so many, it's not just a European thing. There are a lot of privacy compliance rules which say that if you collect your data for one particular purpose, you cannot use it for something else, so sometimes that data might be brilliant to start an AI project but you cannot use it. And the other aspect is the data quality, so what Marina was saying is that it's very hard to get good quality data, data cleaning or getting it in a format or a shape where you can use it, it’s not a sexy work, you know. As always, barriers and solutions go hand in hand, so, when I talk about some of the solutions might sound as barriers if you don't implement them.

How will artificial intelligence change the future of business?

Marina: AI can change a lot. If the companies will start -and they will, I presume- to use AI in all processes, in optimizing a lot of their operations in every field, AI will take care of the day by day ,and people will take care of what can be done differently in the future, things that with AI cannot do. There's always the human design behind this. And what about the small and medium enterprises? On the one hand I believe that for those companies as well, that can make the difference, and I think that there are many cases where implementing AI systems could have those companies to jump ahead. Of course, in order for this to happen, it's important that there is a very pervasive knowledge of AI. Also, if you're a small medium enterprise you don't need to use such enormous and very expensive algorithms that need hundreds of cores in order to run, but you probably can benefit from starting with some small, simple, very well focused algorithm, that can relieve humans of a lot of day by day work, or have it done better, or achieve better quality, whatever is the context in which we applied. The risk here is that if the small/medium enterprises are left behind in this evolution, what could happen in the worst case is that the big companies take the business, the space, the knowledge, the know-how, the very important solution that the small company has discovered. The big company using data can come to know this and make it their normal operating way, so that the small company can lose their business value and the reason why they are on the market.

Shalini: We all hear about reports that say that by 2030 AI is going to create x trillion dollars in global economy, but where is this going to is something that's unknown, and there could be economies of scale here, because sometimes if you have massive amounts of data, the amount of hardware you need or the kind of algorithms that you need to run might give you a great competitive advantage, and how can we catch up with this divide? Small medium enterprises don't have access to these resources, but they're nimble enough, they're able to change their course very quickly, so they could be the ones building critical pieces of this AI infrastructure, they could be the ones developing innovative solutions that can plug into this massive technological stack, if you may. We should also be aware of the importance of data from the beginning, and get your infrastructure right, have iterative developments about your AI processes. Even if it may sound boring, I think they should get together some standards and processes in building AI, data models and pipelines. I think it will grow organically from there.


Picture of Clearbox AI
'People also ask...and we answer!' is the new interview series by Clearbox AI. We try to answer Google's most researched questions on a specific topic about Artificial Intelligence or Machine Learning with our guests. Enjoy!