Featured Technology VISUA News

VIDEOS

Podcast: Visual-AI DIY? Are You Crazy?

See Transcript

[00:00:04.730]
Hi, I’m Franco De Bonis, marketing director here at VISUA. And in this first episode in the series, we ask, ‘are you crazy to even think about developing a Visual-AI application?’

[00:00:18.190]
We can do it ourselves. It’s every company’s battle cry, which can lead to amazing success, freedom and financial benefits, but can equally lead to massive failure, wasted resources and budget,

[00:00:29.690]
and lost opportunities as you’re slow to market. So when it comes to computer vision or Visual-AI based projects, how do you decide whether to roll up your sleeves and start developing or calling the experts with ready to go solutions? Let’s find out, as I’m joined by Luca Boschin, CEO and co-founder of VISUA, Declan McGonigle, VP of Sales and Marketing at VISUA. And we’re really delighted to be joined by Gian-Paolo Valero, data and AI specialist at Microsoft. Welcome, each of you.

[00:01:00.280]
Thank you for inviting us.

[00:01:01.580]
Thank you.

[00:01:02.400]
Wonderful. If you can just give us sort of 30 seconds overview. Firstly of what each of you does. Just to give some context, we’ll start with Luca.

[00:01:10.420]
Sure.

[00:01:10.720]
So I co founded VISUA with my co founder Alessandro. And what we do, we focus on building and delivering great image recognition technology.

[00:01:21.160]
Fantastic, Declan

[00:01:23.080]
Hi, I’m the VP of Sales and Marketing at VISUA. I have been working here for three years since we really started to progress commercially and work along the lines of all the different clients we have and all the different vertical markets with Visual-AI.

[00:01:39.130]
Wonderful. And lastly, Gian-Paolo what are you up to?

[00:01:42.970]
Thank you for the inviting, and lovely to be joined with all of you from VISUA. So my role at Microsoft is I work with customers, a set of customers to help them develop their projects from data from SQL stuff to connectivity services, Visual-AI, spelling checks, all of those AI services that Microsoft offer.

[00:02:03.630]
Excellent. So each of you have really direct experience dealing with customers and projects. So I’m really looking forward to this discussion. So let’s get started. I guess the first question really, and I’ll start with you, Luca. The first question is why are we even having this conversation? What is it about Visual-AI that makes it different in each of your opinions to any other typical project that a company might just begin and start working on?

[00:02:33.330]
Sure.

[00:02:33.640]
Well, AI is complex by itself. It’s a new doctrine in general, but even more complex is Visual-AI. So in a way, it is a challenge of a higher level of magnitude. And the reason for the complexity, specifically with Visual-AI, lays in the fact that it’s an extremely unstructured challenge to solve. So if you compare Visual-AI with other forms of AI, say, for example, NLP, natural language processing, what happens with Visual-AI is that a concept like a dog, for example, is going to look extremely different in each single image where a dog is present. So it’s extremely unstructured. What if you take somebody pronouncing or writing down the word dog, that’s it. It’s very specific. So the challenge with Visual-AI is the level of unstructuredness that it brings along. And it’s interesting because you can also prove its complexity through neuroscience. So not many of us know that the larger part of our brain is actually dedicated to processing visual signals. So it kind of proves the point that even evolutionarily, detecting concepts and things visually is a very complicated task.

[00:03:55.160]
That’s interesting. Gian-Paolo, anything to add there for you?

[00:03:58.720]
Yeah, I think what Luca was saying, every AI, every model, every machine learning model has noise to tackle. But I do think that Visual-AI is specifically taxing and complex to really nail down the model. Just think about I think we have all the computer power that we can think of, and we’re still trying to create AI that could describe an image. And the reason for that is, for example, if you look at my background, you might see that I have a sofa, I might have a frame that I’m wearing a hoodie. So it depends on what you’re looking for. And also it adds a lot to the noise, whereas with text it’s easier. Sound? You look at the wave with Visual-AI. I think it’s much more complicated on that sense.

[00:04:47.810]
That’s really interesting. So I guess at this point then, so we’ve identified it’s a real challenge, but I guess we need to also understand what do we mean by when we talk about we’ve phrased this as DIY, do it yourself. What do we mean by that in this context? And what is the alternative? So I mean, Gian-Paolo, obviously, from a Microsoft point of view, what does that mean to you?

[00:05:17.210]
Well, I think I’m blessed to be in the position that I am because I do believe we’re going to make amazing things in the next couple of years in terms of AI and specifically Visual-AI from a Microsoft point of view, our job as a company and that’s actually in our mission statement. We try to make everybody achieve more with every tool that we make. So we try to our main goal, especially in this one, is Visual-AI, is to create democratization of Visual-AI tools. And we do it at three levels. We do it for non-technical people. So we try to create tools that even people that would see like a PowerPoint type of software where you can drag and drop and do stuff, and that would create AI models in the background. We also then have Lego block style, and we have complete DIY, where we just give you the computing power and you create all of your models. So we do try to create all of these tools that everybody can do their own AI. But what I found in practice is that’s never the case. You do facilitate the use of technology, but I’ve had the opportunity to work with more than 200 accounts or different companies and there’s a specific pattern where each individual company that has been joined with experts in the area, their products have been successful. Otherwise, it’s quite difficult.

[00:06:46.490]
That’s really interesting. Often people will think that some of these modular systems are plug and go, and it’s like ‘we have a solution!’ So Declan, obviously, VISUA doesn’t provide Visual-AI in the same way that Microsoft does this Lego block and very nontechnical interfaces and that kind of thing. So do you see this concept of DIY in the same way, or how do you see it?

[00:07:15.880]
I think one of the things that we’ve seen right across the board is visual data is growing exponentially everywhere, and so is AI. And AI is a very kind of large, complex field. So there’s a specific challenge and a unique challenge around identifying objects and getting information and data from all these visual data around us. So, in some instances, companies have been successful in being able to take down kind of open source libraries, been able to address small projects using AI, whether it be voice, whether it be text, whether it be visual, et cetera. In some instances, these have been successful, and successful maybe a POC stage. But we’ve also found a lot of our clients, even if they have technology, even if they have engineers, et cetera, who have some experience in it, once it goes past a certain size or past a certain level of complexity, it can often fail. And it can often be a lot more costly than if you identify specific experts in the field who can often deliver something for you much quicker. So it’s something that people have to be aware of is that every element of AI isn’t the same. The challenges for every area aren’t the same. And as we develop within it, there are experts beginning to appear for all the different various areas of it. And it’s important to keep an eye out on who those are and why those people are being successful.

[00:08:50.510]
Okay.

[00:08:51.010]
And I would only add that there if you allow me, because I think sometimes, for example, we have two tools and Microsoft that are very nontechnical. So one is Machine Learning Studio, which basically anybody can login. It’s a Web browser experience where you literally drag and drop models, and you create your own machine learning models. And that’s not a specific of VIsual-AI, but machine learning overall. And then we have another tool which is based on computer vision, where you can create your own models to identify images. So, for example, the demo that I do on this is that you create bikes, and then you have mountain bikes or city bikes. Right. And we can tell the differences between bike and a car. But unless you train the model, it’s hard to differentiate a mountain bike and a city bike. Right. So basically, you put all the pictures and then you’ve trained the model without any tech or any programming behind to identify now a subdivision of bikes. So I think Microsoft, in our case, we’re happily guilty to have made things easy so people can actually think they’re quite achievable because they are.

[00:10:01.610]
But the problem is we’re giving away the Ferrari as free as we can and as easy as we can. But it’s still a Ferrari. Okay. And if you give me, if I drive a Ferrari, luckily I won’t crash it, but it won’t go as fast as if Michael Schumacher would be driving it. And I think that’s where the question arises is, who’s driving the Ferrari?

[00:10:26.640]
That’s really interesting. So, the takeaway that I’ve got from this is, DIY is really about getting to a point where perhaps you’ve proven the concept and you’ve then understood your project a lot better. And at that point, there’s a different decision to be made. It might not be, for example, delivering the whole thing, start to finish at scale and whatever. So that’s a really nice segue into the next question that I have.

[00:10:59.450]
Basically, I’m just going to open this to the floor. So whoever wants to jump in. So when should companies consider? I mean, I would imagine there are scenarios where Gian-Paolo, for example, there are scenarios where someone could use the lowest level entry level solution from Microsoft and just create something which they can that is usable for them. And then there are going to be times when even your top level, as you said, without expert assistance, they are going to get nowhere. I guess the question is when to DIY and when not to DIY. So who wants to jump in?

[00:11:37.610]
I can give an overview of what I’ve seen sometimes in the market. So there’s been times where we’ve had people in conversations or maybe inbound chats where people have come along with us with a specific issue, and after we’ve kind of gone through with them and identified what their issue was, sometimes the project is so small that you kind of nearly even say, so why would you use AI to do this? It’s something that’s so kind of small and minimal. In the bigger scheme of things, it doesn’t make sense to do it because in some instances, people are kind of very fixated on saying AI must solve everything. So it has its place. The other area where it makes sense, maybe for people to do a DIY kind of prototype might be in a very small POC of a use case. So they’re trying to figure out would AI or would some sort of AI models help them with a specific use case, and it might make sense for them to do it. If they have some engineers then to say, take some stuff down, run it through and see, can they figure out that, yes, this at scale would make sense.

[00:12:44.240]
What they then have to do, though, is make the decision to say, okay, in being able to do that. And at that small scale, is this something that we can scale out? What are the other levels of complexity that are going to be involved? What are the other decisions that are involved? What are the integrations are we going to have to have? And then they have to be very clear about do they have all of that expertise to be able to produce a working model of something or a working product with it? So they have to be very careful that at the outset somebody may be able to create a small POC for something, but then they have to make sure that they are aware of all of the other elements that are required to bring it to production. So it’s just something that needs to be considered every time somebody looks at a project like this.

[00:13:29.560]
Yes, scaling up is definitely a complete different story. And maybe just to add on that, also it’s difficult to see cases where you can be successful, but definitely if you have very low complexity. So, say that you have to understand only one concept in extremely repeatable data, then perhaps you can manage to build something yourself and in a way and make it scale. But it’s very rare the case where it’s just one very specific model you need to identify in extremely repeatable data, and especially in the context of computer vision / Visual-AI. As we were saying previously, a concept can look so different in each single image that there you go you rise with the complexity of scaling it up.

[00:14:20.580]
I think that’s the jump out point about the bicycles. If you’re just looking for bicycles, then that’s one thing. If you’re looking for mountain bikes versus city bikes versus kids bikes versus whatever, that’s a whole different level of complexity. Are we talking about that kind of difference?

[00:14:35.710]
Yeah, as well, absolutely. So if maybe you’re just looking for the mega concept of bicycles and you’re happy to maybe only identify 60-70% of the bikes that actually appear, then probably a DIY will even work. If you need to scale it up to more concepts or identifying all the bikes and maybe start looking into the variance of bikes, it becomes difficult and you need a bit of more expertise.

[00:15:03.990]
And how does scale fit into this? So obviously, let’s say you just got one concept or one type, one classifier. Let’s say whatever it is, that’s cool. But how does scale impact on that? So the complexity could be simple, but the scale effects that also.

[00:15:21.620]
Yeah, probably it’s around cost and false positives. So as soon as you need to scale things, you’re going to see incredible server builds, because it’s not an obvious task to scale up. Ai running on a lot of models simultaneously, tens of thousands of models simultaneously, if not more. And again, the problem of the false positive. So if you’re looking just for one model and once in a while, you get a mistake, that’s fine. If not, there’s multiple models, multiple variations. The number of wrong answers starts increasing at an incredible magnitude and the data is basically garbage and you can’t use it.

[00:16:04.480]
And obviously, then it depends on the sensitivity to those false positives, negatives, whatever it may be. Anything to add, Gian-Paolo at this point.

[00:16:14.670]
Yes, I think what I’m hearing from you guys, I think you’ve put in, I think very rightly. So kind of a quadrant of technically, when you’re performing Visual-AI, when to do it and when not to do it. I would add another angle to that kind of quadrant because I would also argue that it’s not only about the technical point of view, whether you’re able or capable to doing the DIY Visual-AI, it’s also about the business. So I try to challenge companies, like if your core strategy revolves around that Visual-AI and your business model is built on top of what you’re going to build on the Visual-AI, then obviously you’re going to want to make the invest to create a team and to build a team to be able to tackle the Visual-AI problem that you’re trying to solve. Okay. But if you’re a retail company or if you are an engineering company, even an engineering company, or if you are an energy company, whatever company you are and you’re just creating or adding on top of your business model, I would argue whether you really need to DIY on Visual-AI. Okay.

[00:17:38.850]
And why not the same way you do it with other areas, finance and other stuff. And sometimes you bring other people that are expert, why not talk to experts on building or adding the Visual-AI business model and then get that help? So I think I would also add not only the technical point of view, but also the business conversations on top of that to make a decision.

[00:18:03.570]
Okay. So let’s get a takeaway from this and then we’ll move on to the next section that I wanted to cover. So we’ve identified you could all argue a Quadrant between top left is low complexity, high scale. We’ve got high complexity, high scale. And then the opposite is low complexity, low scale, high complexity, low scale. And depending on the complexity and the scale, like if we went to the extreme with the highest complexity and highest scale, what I’m hearing from you guys, don’t even try. Go get the expert opinion, go get the expert help, whether that be expert solutions, expert brain power, whatever it is to implement that perhaps to Declan’s point, if it’s low complexity, low scale, you almost need to think about, do I need to make this Visual-AI or is it so such low scale and so simple that I can put humans to task on this at a reasonable cost? And then to your point, Gian-Paola, we have think about the business needs and how reliant we are going to be on this technology and wherever you are in the Quadrant, if we’re super reliant on it, we need to go and get help. Have I summarized that accurately?

[00:19:20.250]
Yeah, but I think there’s another point and I don’t know if we have time, but then you have the security and Privacy piece with the Visual-AI, which is a whole set of topic about it. But yeah, I think you’ve summarized it accurately.

[00:19:34.600]
Okay. Let’s spend just a few minutes, very short amount of time just to understand then to some degree, we’ve identified that there is a place for doing this yourself to some degree. So before we move on to all the issues and what happens when how you should outsource this, let’s say what are the benefits of at least doing the proof of concept and getting to a stage where you’ve proven the need for it and the ability that you can do it. So what are the benefits of doing that, Luca, Declan? I mean, maybe one of you wants to..

[00:20:17.400]
The benefits can often be like anything at the beginning of a project. If somebody wants to look at developing something new, put some new products and put something new into their business workflow, they have to be able to go and present it to senior management. They have to be able to articulate the concept, they have to be able to show some sort of proof of concept to see where it would fit and how they could put together a business case for it. So it can often be useful to be able to do that in house where you understand what the kind of big level goals are, but be able to articulate it and get some buy in from people that you might need within the organization to help you to bring it forward. So you’re always going to have to have multiple stakeholders if you’re going to provide some new product or some new products set. So in that instance, it can be basically timely and it can also be something that’s very focused and clear for your own business, for your speaking your own businesses language to explain this is something that we see benefit in and where we want to go and take it.

[00:21:18.990]
That being said, then what you have to make sure that you do is it something that you can do yourself all the way through and make sure that you’ve thought about every single element that needs to be done to put it into production. But from that point of view, the DIY element can be done and can be useful at the beginning where you don’t have to rely on outside people and try to have to explain it all to them when it can be done much more quickly internally.

[00:21:43.930]
Okay, Luca, Gian-Paolo of anything?

[00:21:46.890]
I mean, just on Declan’s point. The important part, as you were touching at the end, Declan, is then not to be naïve and think that you can go from a demo to a production grade technology internally, or if you think you can do it, you should seriously analyze it rather than just thinking, sure, we can do it.

[00:22:10.810]
How quickly can things fall apart at that stage? You’ve got a working demo, it looks fantastic. The board see it and go, wow, this is amazing. Does it kind of creak at the seams for a while or does it just sink to the bottom of the sea straight away when you start putting scale of pressure on it again?

[00:22:29.190]
Because there are so many different moving parts. Where we’ve seen, obviously we don’t say but where we’ve seen and we’ve been told by clients where issues arose, was that because the demo was set up, because the proof of concept was done, even if it wasn’t expressly said the feeling was within the company, oh, that’s great. So we can do it ourselves. You just need to kind of multiply that out now by 10X and we’ll have a working product, then you have the difficulty. If that isn’t the way it works, that people have to go back then and say, okay, look, we now need outside help to do this. And it can be something that can be very difficult to explain maybe to senior management or something that even people might be a bit embarrassed to go back and say, look, we can’t actually deliver this full scale. So there’s an element of making sure that the expectation is set that if this is a proof of concept and it works, that before they would come back and say how we would go to production, that all of the elements have been investigated and not just given the impression to kind of senior management, the company that once you can do something that’s a particular size that you can just 10X it or 20X it.

[00:23:38.000]
So it’s not that all of them fail at specific points, but if the expectation hasn’t been set to say, okay, if we are going to go into production with this, we must go and look at all of the elements that need to be put together to do it and then come back again to the business and explain how it will be done rather than just letting people assume it can be just multiplied out.

[00:24:01.570]
You also have to take into account for everybody who’s listening. Right. Let’s call it out the way it is, when companies try to say that they are going to build their product on their own and which is understandable it’s because they want control and they think it’s going to be cheaper. Right. Because when somebody expert comes, they’ll charge you expert premium prices. And then you say, oh, I hire maybe an intern that is coming right out of college, a freshman and will do the project, which I’m not saying they won’t be able to do it, but sometimes I think that could be. What’s the word that I’m looking for? It’s not real. It’s a fantasy. You’re trying to get control Privacy because sometimes you don’t want to share what you’re doing with other companies, thinking that you’re going to be stolen and then it’s going to be cheaper. And I would argue back that the three of them prove me wrong. I do believe that those three things that you’re coming up to do, the Visual-AI yourself, are not really true. And you’re going to end up paying more, have less control and probably less Privacy even because again, when you’re doing Visual-AI or any AI for that matter, you also have to take security into account, which is not even part of the conversation today.

[00:25:26.350]
Okay, so we’re going to move on to the pitfalls of it. But before I do, I want to leave this bit on a positive note because it was about the benefits. What I am hearing from you, though, is that it can help you to prove the concept, understand the challenge, understand the scope of what you need to do, because as you work through it, maybe before you start engaging, I think Declan mentioned this. Before you start engaging with the experts, you really need to understand what you’re trying to achieve. And maybe that first step in building out the first part will help you do that better. So when you do engage with the experts, you can now show them what you’ve done so far, show them the scale of the issue and you have a better understanding. Is that a good summary of that section?

[00:26:18.590]
So let’s move on to the next part, which is so we’ve identified that there is a great starting point there. But what are the pitfalls? What are the issues? Let’s say someone’s watching this, saying whatever you guys have said there, I don’t agree with you. I can do this.

[00:26:38.390]
So what do you say to those in terms of the pitfalls that they’re going to face if they do decide to press ahead and just do this themselves?

[00:26:50.560]
I think from experience, the pitfalls that I’ve seen is just people underestimating the complexity of various elements of AI. As we said earlier, AI is a very broad term. And just because somebody has been successful and maybe developing some AI projects either at another company or internally in your own company, doesn’t necessarily mean that every AI project that they approach is going to be successful. And I think the key pitfall that you have to make sure that you don’t fall into is if you see that, if we take computer vision, so there are computer vision, there are libraries of models, open source models, there are services you can purchase. And people think, oh, I can just take a bit of this and a bit of that, put it all together and it will create what I want. And like everything, the complexity and the devil is always in the detail. So it’s very important that people understand the clear complexity of any specific project that they want to address and that they can figure out are all of the resources that they’re using going to match and address all of those concerns? Because if they go down a route and they get too far down and then realize that they’re missing certain elements, it’s very difficult to come back up again and start again.

[00:28:17.070]
So it’s that initial element of just making sure AI is all quite new. And it’s very difficult to just decide to get something off the shelf or something from an open source library where because you may have had some experience, a bit of experience and some elements of it that you’re necessarily going to be successful in it, it has to be something very seriously considered.

[00:28:45.930]
And Luca, have you got any sort of specific considerations if someone they need to take into account or plan for?

[00:28:53.040]
I mean, the main thing is that if you get it wrong, it’s going to cost you a lot more. Perhaps that’s a big challenge and it’s going to cost you in terms of dollars, human resources, time wasted. So to me, it’s mostly a thing of being efficient. It’s like if you get sick, if you have mild things, you can take care of them yourself. As soon as you have something more serious to look at, you need to get a doctor in the room to make sure that he gets the right diagnosis straight away so you can move on with your thing. So losing the time. To me, it’s perhaps the most troubling part for a company where they may be losing market share opportunities as a result of trying to scramble something together

[00:29:44.570]
Or even an opportunity internally, right. And I’ve seen this with customers. For me, it’s also a little bit even more painful to watch because I talk to customers directly and I am measured internally on how much our companies use our different tools. Okay. So the more they use the tool, the more I get a pain in the back. Right. And sometimes when I talk to customers, it’s challenging to see that they want to do it themselves. And you already see that that one is going to fail. And then what ends up happening is that the product completely falls apart because that person lose credibility. So internally, you already lost the opportunity to even show to your post as part of directors something that you maybe were really passionate about and thought it was very easy. What I’m looking for. How do you call when you’re in the desert and you see water but there’s no water there?

[00:30:44.260]
A Mirage?

[00:30:44.690]
Yeah, exactly. That’s the one. Right. Again, I think we’re happily guilty for Microsoft sometimes of doing this. If you go right now, if you Google or Bing, whatever you want to do. I am from Microsoft. I have to say Bing, Bing Computer, Vision Azure, it’s going to get you to a website that you can actually paste a URL.

[00:31:06.960]
It’s going to give you a description. So you see it and you’re like, oh, wow, anybody can do this, right? But then when you’re in the little pieces, it’s like, if I took a course on finance and all of a sudden I believe I can be an accountant. It doesn’t work that way.

[00:31:26.410]
The other thing that we’ve kind of seen through some experience of people we’ve worked with in the past is sometimes people are very enthusiastic about kind of taking on a project, doing it themselves. And when they see that project isn’t going to be successful, like Gian-Paolo was saying, sometimes people would say they can kind of see the writing on the wall and a team gets disbanded and people start to leave and then the company is left with a big problem whereby they haven’t got what they were looking for. The teams have kind of left or split and then they have to try and start the whole thing all over again so it can be have a lot of further consequences and not figuring out exactly how to deliver that project from the beginning because expectations can be set and are very difficult to change.

[00:32:17.140]
And I think the best way to describe it is the opposite. So I actually had a company and they were paying the most premium price, like Microsoft also has consultancy services. Okay. And they were paying for Microsoft consultancy services, which is that only big companies would do that, right. Because it’s such a premium price. I was curious and I asked this company, what drove you to hire Microsoft services to that? And what he said blew my mind because what he said is like, look, we’re going to take and it’s a project on Visual-AI, they have to agriculture and grow fish that they grow organically and trying to, instead of fishing, they grow it in a controlled environment. Right. So they wanted to use Visual-AI to make sure how they were going to go. Anyway, the point is, when I asked him, why did you decided to go with Microsoft construction services? He said, ‘because if it fails, nobody is going to blame me’, right? So if it fails, obviously he’s very passionate about it and we believe we’re going to do it because we’re going to do it. But he said, I also need that expertise companion to make sure that everybody who sees the project know that this will be successful.

[00:33:40.650]
And if it’s not successful for any reason, it’s not my fault or my team’s fault because we were accompanied by the most expert people and I found out that was an amazing way of viewing things because it’s like, yeah, he’s actually even creating a safe passage for himself. It’s going to be successful. But if it isn’t, it wasn’t his fault.

[00:34:06.670]
It goes back to the old days of no one ever got fired for buying IBM back in the day of when PCs were the first thing. So I understand that. I think the Declan said something earlier, which I think is probably the crux of this whole thing. The biggest pitfall is setting the right expectations. If you set the expectations right of what you’re doing and understanding where you’re going to go next, I think then what I’m hearing from you guys is you’re not going to go far wrong. If you set the expectation that this little proof of concept that you’ve pulled a few modules together and you’ve shown that it can achieve something, is going to run the whole business part of this, then you’re setting the wrong expectation and you’ll look like a failure. So, okay, that’s a really interesting approach. We talked about a brain drain. If people leave, Luca mentioned about potholes and dead ends, if you hit a dead end because you can’t progress any further, then that’s an issue as well. So that’s interesting. So let’s talk about, because I’m really interested and I guess you’ve segued into this really well Gian-Paolo, let’s talk about the benefits of using experts, because despite the fact that you have these different levels, layers, ultimately, I guess what I heard from you is you drive people to the point to reach out to experts, whether it’s your own internal consultants or whether it’s external consultants who bring in the expertise.

[00:35:48.260]
And there’s obviously a reason for that, otherwise you wouldn’t do it. And I’m mindful of the fact that even in human life, as we learn more, we’ve become more niche. A PhD in one subject doesn’t mean they can do everything else. And it seems to me that people think about AI in a very broad term. So I can do AI, so I could do AI. I can do any AI, but that’s what I’m hearing from you all is that’s not the case. It’s much more specific than that. So let’s talk about the benefits of using experts. What does that bring to the party? Luca, I’ll start with you.

[00:36:32.270]
Yeah. To me, it’s mostly the guarantees that it brings. So if you choose the right partner, you see with who they work before and where they’re able to solve the challenges that previous clients asked. It already gives you the comfort that you’re going down the right direction. And then you compound on top of that, a lot of other warranties, safety aspects like you’re going to know how much is going to cost. You can get a good idea from that partner, how long it’s going to take to deploy so you can properly synchronize it with delivering what you’re looking to deliver to the market as fast as possible, and then also making sure that you have somebody that keeps running the engine successfully. AI needs constant maintenance, and you may need in your solution to add new models or change models or adjust them and having somebody that on the fly can perform those tasks for you because they do it constantly for multiple other clients. To me, it’s something that gives me a lot of comfort in the idea of running a business, because I know where I’m going to end up. There’s a lot of more safety and security there.

[00:37:47.570]
Okay, Gian-Paolo. I’ll come to you next. In terms of Microsoft.

[00:37:51.890]
I don’t even know whoever’s asking that question. I think when I proposed to my wife, did this crazy thing that I did a skydive jump, a tandem Skydive jump, because I’m not a professional skydiver, and I proposed on that. And that was very nice to do because she had to say yes. Right? Otherwise the other guy wouldn’t open the parachute. I’m just kidding. That wasn’t going to happen. But basically, that was my first time doing skydiving. And when the guy approached me, I said, Look, I’m going to be your tandem guy. And I’ve done 150,000 jumps before I could even get jump alone. And that was like, okay, look, I’m safe. The guy knows everything he gave me you’re going to do this. Imagine my feeling if I said, okay, I’m going to do this on my own. Or even the guy who came with me said, oh, this is my first time doing a tandem thing. I would have gone crazy. I would maybe not even do it. So even the question doesn’t make sense. Like, who doesn’t want to have an why wouldn’t you want to have an expert doing it next to you and making sure that project is going to be successful?

[00:39:03.960]
I don’t understand. I’m sorry. Maybe I’m a little bit irreverent here, but it’s just that you’re going to pay less money. For what? I don’t want my tandem instructor to pay less. But with the risk of falling apart, I don’t know. Unless you’re very sure that you’re able to do it and you’re 100% confident, I will argue, have you done a skydiving before? Because if you haven’t, you’re always going to need somebody to teach you first.

[00:39:37.930]
And if you keep in mind the medical bills after crashing after the site, they gonna cost you less anyhow.

[00:39:49.130]
If you even have to have the medical bills. Yeah, exactly. I mean, Declan, you’ve spoken to numerous companies who are maybe at the beginning of projects. What are those discussions like in terms of why they reached out to VISUA and why they might reach out to experts?

[00:40:10.550]
I suppose the examples I’ve had instances where we’ve spoken with people who have gone through the track of trying to do something themselves, creating a nice POC, then trying to expand it out and coming across all the pitfalls that we spoke. And I think the biggest thing they got from speaking with us was they could immediately see because we were experts in our field, we could identify very quickly where the gaps were. We could give them real clarity about how we would solve them. But we were also able to because we’re expert and focused on what we’re doing, we’re also able to say, this is what we do. This is the time we do it in. This is when we deliver it, and this is what it will cost. And you could see the relief, just the kind of comfort in being able to say, okay, we have this problem. We now know how we’re going to achieve it, and we can go back and explain to the leadership how we’re going to address it. And because then they could feel that they could come over and back to us all the time if they came up to any other problem in this area.

[00:41:20.070]
Because this is our business, this is what we’re focused on. You could see that they then became very comfortable with working with us. And because we were experts in that field, we were able to solve that problem, which is, in effect, what business is it’s about solving a set of problems every day? And they could say, okay, we can, in effect, outsource that problem now to somebody who knows what they’re doing and we know how much it’s going to cost us, and we can keep moving and start developing our tools. So it’s like somebody who’s going to come to a scenario where not through any fault of their own specifically, but they came to a roadblock where they couldn’t see their way past the problem, came along with somebody who knew how to address it and were then, in effect, able to outsource it to them.

[00:42:02.530]
Very good. Are there any sort of final comments from you guys as we draw to a close on this generally?

[00:42:11.530]
I think the only thing I would say to people is that there are lots of things to consider and lots and lots of moving parts around AI. I think one thing that you should always consider when you’re looking at using any provider, any service, is to make sure that it’s not just like a faceless service where you present data and you get something back and you have no interaction and no way of configuring and discussing what your requirements are. It’s always important to make sure that if you’re engaging with some experts, if you’re engaging with any teams to make sure that you have access to experts within their teams where they can say and they can show that they’re really interested in doing is solving your problem, not interested in just giving you some technology or some tools, etc. That they’re really interested in solving your problem. And if that’s their goal, and if that’s the way they focus on their business, that’s where you’ll get the most value for trying to provide your own solutions.

[00:43:15.800]
Okay.

[00:43:16.340]
Given the complexity of AI, something I always like to keep in mind is also focus on the values of those providers of AI have. Because AI is so complex perhaps you’d rather have somebody that is very transparent in the conversation. So if there is any challenge, it’s going to come up straight away. You’re going to be able to discuss it together and understand how to solve it. But also somebody that is extremely passionate about what they’re doing because AI is something new and those that are truly passionate about it will have an edge from those that are just doing it to do a business right. Those values is something I would keep in mind as well when looking for partners to work with.

[00:44:04.980]
I guess to your point, there’s also a lot of ethical questions around AI generally, so you’ve got to make sure the ethics of the provider matches your ethics. You don’t want a provider that’s doing funky things somewhere that you don’t want to be associated with. So I get that. Any last comments from you, Gian-Paolo?

[00:44:22.210]
I think maybe the conversation seems like we’re telling people, oh, don’t do it right. Only do it if your parents is with you or something like that, right. And you’re going to get burned. But actually no, I think it’s an easy two step because it’s actually easy. And I think Declan put it very nicely. The tools, Microsoft Tools and other providers tools. They allow you to build POC very quickly, so you know you’re not jumping in a ghost town or in a forest that you don’t even know where to start. So at least that’s given to you. And then once you’ve realized this is actually achievable, then make sure that you bring somebody along with you to make sure that you land the project correctly. And that’s the beauty of it. Before, sometimes you would tackle a project where you have no idea what to expect, or even the thought of having a pilot would be so dear that you’re like, oh, this is such a high mountain to climb right now. You can pilot it, you can trial, okay, but make sure that’s not a production environment.

[00:45:25.210]
Very good, guys. Thank you so much. I think we can leave it there. I’ll summarize as follows. What I’ve learned, hopefully everyone watching this and listening has learnt too, is that there is a time and place for both DIY Visual-AI and for outsourcing. I guess it’s all about understanding the position you’re in, the size, scope and complexity of the project, and when you need it, and if you have the budget to see it through. Having your own solution has many long term benefits, but there are also many rabbit holes and pitfalls along the way. So if you’re not able or prepared to see it through, you could summarize it and don’t even start. So thank you, Declan, Luca for your expertise and obviously very special thank you to you, Gian-Paolo for lending us your time and valuable insights from different provider’s perspective.

[00:46:14.670]
Look out for our next episode where we’ll be looking at fishing that’s the kind smile with a PH and how Visual-AI provides a really intriguing new approach to detecting the latest kind of graphical based attack vectors used by bad actors. So see you next time. Thank you very much.

[00:46:31.200]
Bye.

[00:46:31.720]
And thank you guys again.

[00:46:32.910]
Thanks for the invite. Thanks.

A Poke in The AI – Episode 1 – Visual-AI DIY? Are You Crazy?

Introducing the Visual-AI podcast you’ve been waiting for! A Poke in The AI will see VISUA marketing director and host Franco De Bonis joined by guests to discuss a specific topic related to Visual-AI.

In this episode he is joined by special guest Gianpaolo Valero from Microsoft alongside VISUA’s CEO Luca Boschin and VP of Sales and Marketing, Declan McGonigle.

In this episode, our guests dissect the idea of doing Visual-AI in-house; the situations in which it will work and the situations in which it won’t. We talk about the benefits of taking on the task yourself and those of outsourcing the work to a third party. You might even get a great idea for a marriage proposal. Yes – it’s the podcast that has everything!

You can watch the podcast below, or you can listen on Spotify, Apple Podcasts or Soundcloud.

If you’d like to take part in the Podcast email [email protected]

RELATED

BLOG BLOG
Press Release: Conviva Partners with VISUA

Reading Time: 3 minutes Conviva Partners With VISUA To Integrate Visual-AI Technology In Its Leading Insights Platform VISUA’s Visual-AI suite will enhance the insights delivered to […]

VISUA News
BLOG BLOG
The European Digital Services Act – How it Affects you

Reading Time: 6 minutes TLDR: The European Digital Services Act has been ratified into law by the European Union and will have wide-ranging implications for companies […]

Content Moderation Featured
BLOG BLOG
Press Release: A New Era In Sports Sponsorship Valuation

Reading Time: 3 minutes Visual-AI Company VISUA Ushers In New Era In Sports Sponsorship Valuation with Disruptive Commercial Models VISUA aims to change the face of […]

Featured Sponsorship Monitoring Technology VISUA News

Trusted by the world's leading platforms, marketplaces and agencies

Integrate Visual-AI Into Your Platform

Seamlessly integrating our API is quick and easy, and if you have questions, there are real people here to help. So start today; complete the contact form and our team will get straight back to you.

  • This field is for validation purposes and should be left unchanged.