Antonio Grasso started his career as a software engineer in the 80s. Now he is the Founder and CEO of Digital Business Innovation Srl, a recognized member of Intel's Internet of Things Software Innovation Program, and an ambassador, fellow, and advisor to organizations too numerous to list.
Antonio Grasso joins us to explain how he empowers some of the biggest companies in the world to use AI in a meaningful way and explains the two ways his company goes about this. You’ll hear about what Antonio believes is coming down the pipeline in terms of the Internet of Things, especially when it comes to edge computing, and why network traffic has become a huge concern. We discuss where edge computing begins and ends with regards to the difference between the device and its computational resources. In light of the fact that one can infer at the edge but not train at the edge, Antonio shares his views on why he disagrees that the ultimate goal should be to train at the edge. He also provides a helpful resource for AI practitioners to calculate an AI readiness index.
Key Points From This Episode:
Tweetables:
“‘Wow, this is really unbelievable! We can also create not [only] code software with direct explicit instruction, we can also [create] code software that learns from experience!’ That really [caught] me and I fell in love with this kind of technology.” — @antgrasso [0:03:13]
“I started on Social Media to share my knowledge, my experience, because I think you must share what you see because everyone can benefit of it too.” — @antgrasso [0:03:39]
“We need to shift to better understand what is the meaning of edge computing but we must divide the device itself from the computational resources that we put [there] to harness the power of computational power in proximity.” — @antgrasso [0:16:15]
“I can not imagine training at the edge. — We can do it, yes, but my question is why?” — @antgrasso [0:20:50]
Links Mentioned in Today’s Episode:
Digital Business Innovation Srl
AI Singapore (AIRI Assessment)
EPISODE 34
[INTRO]
[00:00:04] RS: Welcome to How AI Happens, a podcast where experts explain their work at the cutting edge of artificial intelligence. You'll hear from AI researchers, data scientists, and machine learning engineers, as they get technical about the most exciting developments in their field and the challenges they're facing along the way. I'm your host, Rob Stevenson, and we're about to learn how AI happens.
[INTERVIEW]
[00:00:32] RS: Today on How AI Happens, joining me is the founder and CEO of Digital Business Innovation. He's a recognized member of Intel's Internet of Things Software Innovation Program and an ambassador, fellow, and advisor to organizations too numerous to list. Antonio Grasso, welcome to the podcast. How are you this morning?
[00:00:49] AG: Hi, Rob. Thank you for having me. It's nice. Today, in Naples, yeah, is a sunny days, like in Italy every time. But it is also interesting to speak about AI with you. Thank you for having me.
[00:01:01] RS: Yeah, of course. But benvenuto to you. I should point out, it's more morning for me, but you're winding down your day on the other side of the Atlantic there. So I do appreciate you joining me today. Antonio, there are so many things I want to speak with you about you have your finger in a lot of pies. I guess I'll put it that way. You kind of got your start as a software engineer. You're a mathematician originally by trade. I guess I would just love to hear from you a little bit about your journey. How did you wind up in the field of AI and where you are now, influencing all these various different technological topics?
[00:01:32] AG: I started as a software developer in 1983, what was really the starting early day of IT for business. I start coding the software ERP for companies. Then it's only in the last years that AI has emerged. I started, as I said, in 1983 with a computer that was really slow compared to the one. Then I beat my career over the business, AI CT, information technology for business.
Then I have experience in the last 10 years with a USA company about the financial prediction. I worked with the some big firm, big investment firm, from 2008 to 2012, elaborating an algorithm that can predict the investment feature technical analysis about a financial asset. That was my starting to shift from deterministic software. That is the usual software we used to deterministic software that are programmed step by step. If the user pressed this, do that. If result equal 10, do this one. We call them deterministic software.
The shift from deterministic software to probabilistic software using neural network training, there is a big shift, in my opinion. I embraced that shift because I was catched up by the shift. Wow, this is really unbelievable. We can also create not code software with direct explicit instruction. We can also code software that learn from experience. That really captured me, and I felt in love with this kind of technology.
Now, I become an influencer. Yes, as you say influencer. I talked because I started on social media to share my knowledge, my experience because I think you must share what you say because everyone can benefit of it. So suddenly, people start following me on Twitter and on LinkedIn. I say, “Okay, it's good.” Then I was even more motivated, even more motivated. Now, I cannot go there anymore. Sorry. I like to code develop software but I cannot to date. Now, I'm in the brain phase. I'm in the thinker phase, people that think how technology impact our business, our society and help the future.
[00:03:59] RS: So you have the opportunity to consult with some of the biggest brands, household names. Just to scroll the website and see all the companies you work with. So I'm curious, when you go into these companies, the IBMs, for example, and you are going to consult them on their AI practices, what does it typically involve? Is it about an implementation? Is it about the future use cases for them? How are you helping these companies position themselves to use AI in a more meaningful way?
[00:04:24] AG: I have a double face. My company has double face. This is why we created also a new company. Digital Business Innovation is a technology company that helps customers to build technological projects. So it’s in the execution operation. Then I have also a strategy that is very important, but let me see. With the big multinational company, I help them to explain value proposition. It became more complicated to understand how to introduce the technology because technology is infusing AI with the IoT, blockchain with smart contract. Let me say financial. So it is very unbelievable how you can explain you can pursue your communication goal.
With this large company, I collaborate the most in the marketing, where I collaborate with the communication department, where together we talk about how can introduce this concept. We found this kind of new breakthrough. How can we introduce? How can we talk about it? So researched value proposition usually, and I like to create also infographics articles, etc.
On the other end, I work on a technological project with many companies and also startups. I like to rise a startup from idea to success. Many companies in Italy and also in Europe, I'm helping startups to create their project, to go to the market. It’s an amazing scenario. I really like too much because I feel comfortable. I feel myself writing the article because I infused my personal experience in one article, because after almost 40 years of experience, I have a lot of experience to talk about it.
For example, in the case of IBM, I was the first Italian developer to write the software using the PowerPC. The PowerPC, this is announcement about the reduced instruction set computer against the complex instruction set computer. That was a breakthrough because in one cycle, the CPU can do more against the other complex instruction set computer. Now, the development of the risk as reduction instruction set computer risk has created really amazing thing, something that turnover is important. For example, real time analytics.
So I helped this company to understand how to introduce value proposition, how it's better to talk, and they want an expert. They want an expert. They don't want the marketer. They want a technical expert because they know that it's hard to understand and to explain. Like Einstein say, “You understood well when you are able to explain to your grandmother that thing.” So I'm the guy that can explain to grandmothers.
[00:07:26] RS: I'm not sure how many 'nonnas” are out there listening to this. But if there are any, then you are the guy. That's for sure. So is this attached to just getting past the hype cycle and helping people understand what is possible with this technology and how to make it more mainstream? What is the goal of bringing you in as a technical expert to disseminate this information?
[00:07:48] AG: In a certain way, so to give a trust to what you're saying because this company of a market or communication, they’re a marketer. They’re a journalist. They are not technical. Then they have operational technical. There are silos. I know it’s sad to say. But many companies are silos where technical don't like to talk with the marketers. Marketers don't like to talk with communication, etc.
In this case, here is my consultancy, my collaboration with this company. They try to in an independent way. Sorry, I forgot to say the time in depth. Then in my work, I will not publish something that I do not generate as opinion about the available positions. It's very important because also to the pacing. Acceleration of the market is very high. We have unbelievable fast things that are changing day by day, so the process of a multinational.
But this is also a problem of universities. Universities are not able to maintain the pace of innovation that research and development by the companies create and put on the market. This is the technology transfer. We were used in the past to think about we need to transfer technology from research to the industry. But we can pass through university that also bring resurrection to the market. But now, the problem is that you can miss the link between research and the market because the market is going fast. Then the research is able.
We have many companies that are creating industrial research, prototype development. This is very interesting, but we lose the opportunity to have a good flow of transfer. In many cases, they cannot maintain the pace of innovation. Let me say also, they like my view about what is happening in the new world, my experience. I work with the tens of companies. For example, for Intel, I'm an Intel Software Innovator. This means that I can also propose but participate in the IoT technical field. I work with Intel, but Intel will ask me, “Antonio, can you explain this value proposition?” Then I write an article.
A technical guy that they become a tough leader for opportunity. Yes, this is an opportunity, of course. They pay me, so it's an opportunity for sure. But it's also a passion. It’s a passion that I do with my determination and, let me say, with the love. I love technology. It’s one of my best love.
[00:10:30] RS: Yeah, of course. So I'm glad you mentioned Internet of Things because I wanted to ask you particularly about some of the opportunity there. What do you think is coming down the pipeline? With regards to AI and the Internet of Things, particularly with edge computing, how does this all play together? What do you think is the opportunity there?
[00:10:48] AG: I think it's huge because you’ll say Internet of Things, and then the Internet of Things, if you interpret the means of the words, it’s connected object. Yeah, many things. Many say connected object. But let me say we must split as a force to better understand. We must split between IoT and IoT that is industrial Internet of Things, where you can say that edge computing is present in both aspects. But the industrial Internet of Things is most.
One thing that is really interesting is the convergence of many technologies. For example, you have Internet of Things now that are connected object, but they are stupid. They are deterministic objects that connect the two because these are software. As I said before, software can be deterministic. But now today, AI can be probabilistic. So shifting intelligence to the age is one of the top challenges.
Now, we did it. We did it because many devices have computational capabilities, able to inference. Inference is the process where AI can run the fish, recreate the results. At the moment, we have not big opportunity in the training because training need more resources. But we don't need because we can transfer the model and then we can run. Let me say Internet of Things, connected object is something that is going lateral because we interact with these object.
For example, Alexa, that is a natural language processing devices. You can talk with Alexa, but it’s a limited function about voice command. So we can talk about voice assistance. We need to classify. We need to segment what are virtual assistants, what are the chat bots, what are the prediction algos that can interpret data. Let me say now the trend is that IoT collects data. These data are elaborated apart in proximity, so in the edge. You're using edge devices, part in the cloud because they feed training.
One thing of the artificial intelligence is that they use data to train, and this is one aspect that is really important. You have data that are becoming – They have more instruction. So it's important to get there and to manage the correct flow of data from connected objects. Connected objects are no more sensors, temperature measuring. Connected object can be a car, for example, a truck for a company. It can be a machinery. It can be a device.
We are pushing. With edge computing, we are bringing computational power to the source where the data are generated. Then we say, “Okay, I don't need to transfer all the data to the cloud or to my server. Some data can be elaborated on the edge.” I do and then I give the result immediately. What are the advantage? The advantage of edge computing mainly is latency time. Latency time, it’s something that you can harness to avoid. Then the other is network traffic. Network traffic, that has become a huge concern in industry. For example, we have hundreds of devices that send the data to the server. It’s a congestion. It’s something that can create congestion. In English, it’s congestion.
Then this is important also to have something related to that. So edge computing and its architecture because we talk that edge computing that is something that is a paradigm that bring computational power to the edge. But in terms of architecture, it is a bit more complex because when you talk about, for example, a manufacturer that has planes that is big, you can need various layer of architecture. So we talk about edge computing. But before reaching the cloud and the elaboration, you have a fog computing. Fog, F-O-G, fog computing that is capable of managing. It's like a gateway, but it's more like a gateway because it has computational power.
So edge computing, it's very important to bring computational power to the edge with avoiding that data come up and down to the network and reducing the network traffic, enhancing the latency. Let me say that many people, one thing that I’m noting that many people, they talk about edge computing. They see as a black box. They say this device is capable elaborating. No, this is not the device. This is a normal device with embedded computational resources. Let me see, talk about the fridge, for example.
Or talk about a machinery or a car. The car itself is not a device, an edge device. The car is a car. Then we put computational resources, computational power to give them. We call it edge computing, but we cannot see as a black box. We need to shift to better understand what is the meaning of edge computing. We must divide the device itself from the computational resources that we put to harness the power of computational power in proximity.
[00:16:15] RS: It’s important to call out, as you did, that one can infer at the edge but not train at the edge. I'm curious, given the trend of computational power to be smaller, more accessible, more commoditized, if that's an inevitability that there will be more training at the edge, can we expect that, and what are some of the challenges? You mentioned that there's more possibility for a black box sort of algorithm at the edge that’s less easy to understand. Should that be the goal to be doing training at the edge?
[00:16:45] AG: I don't think so. I don't think so. Because before the training, you need to be worried about data quality. You train AI with data. If data is not clean and is not – Many say healthy data, cleaning data. So I think, in my opinion, you cannot. It can be in some specific case. But generally, you also need to act the data governance procedure, where the data flow is managed, is assisted, and then need to become the training data.
While inference is – Let me say it's a read-only execution. I read and then execute. Training is a read and write. Let me simplify this way. So flow of data, you create the linkage inside the model. You created the model. I don't know if we need the training at the edge. For sure, we have computational power miniaturization. But I think you need to relay on data quality to also have the algos that are training well. So in my opinion, you can do it but you don't do it because the quality can be not what’s expected.
[00:18:03] RS: Do we expect it'll happen eventually though, as we are able to more ensure that clean data, annotated data happens, perhaps even automatically?
[00:18:10] AG: You need to manage the process. You need to manage the process, if you are able to create data to ensure that are not duplication. If you think that is, you can do it, absolutely. But I'm learning that it’s better you go carefully. I will collaborate with many companies that work on data elaboration, data cleaning. One is Sama, for example. That is a leader in this. I had the opportunity to collaborate with Sama about their value proposition. I think it is very interesting.
So if you watch that process, the process is not easy. Yes, you can do. You can do it at the edge. But let me say why. This is why it can be interesting, yes. But why? You don't need. It can be you don't need. I don't know. Facial recognition, for example, in a shop, you can know new customers. But facial recognition is a part, cognitive image, computer revision. It’s like something different from, for example, I don't know, checking the correct shape of a product, quality control. Quality control is something that is important, but I don't know. I cannot imagine how you can do quality control on the edge training the machine to control the quality, if behind it is not an engineer or a human that can say, “This is correct. This is not correct.”
Or as many cases, you need human contribution because we must rely on AI as a tool. AI is a tool for us that, yeah, it’s a good tool, absolutely. It’s probabilistic. It’s capable of cognitive operation, small cognitive operation but [inaudible 00:19:55]. So I don't know in which case you can need to put training at the edge.
[00:20:04] RS: Yeah, it makes sense. With sufficient connection to server computing power, then I guess the question is why do we even need to bother training at the edge?
[00:20:12] AG: Yeah, because they really are an operation, and operation you need to manage the execution usually. It's like sending your rules to the edge, to the machinery, but don't let the machine learn himself, itself on the edge because I don't know how you can manage. It can be it's my fault. Yeah. I know it can be it's my fault. But I really don't – I cannot imagine how the training at the edge. Yeah, you can see the process is standardized. We have a lot of things. We can do it. Yes. But my question is why.
[00:20:48] RS: Right. Yeah, fair enough. Antonio, I wish I had more time with you today, but I have to let you go because you have companies to consult and processes to influence. But before I let you go, I just wanted to ask you one more question at the end here. What advice would you give to the AI practitioner out there, just so that they can be more effective and more impactful in their organization as they develop and implement this technology?
[00:21:10] AG: I have an experience that I want to share with you and with the rest. I collaborate with the European Commission about the AI for Europe. AI for Europe is a project to raise awareness about AI in business in Europe. We are lagging behind the outside like USA and China about AI. We are reducing the gap but we are still behind. I am an expert and analyst part of the evaluation committee.
Given this work, collaborate with them, I discovered that is some good tools that call AI readiness, where you can calculate AI Readiness Index. It's made from Singapore. Singapore elaborated this framework that is a self-assessment framework. You can find on AIRI. Sorry for my pronunciation. I'm Italian, but it’s A-I-R-I, AIRI, and it’s Singapore AIRI. It’s a self-assessment model where you can just responding to simple question. You can evaluate your level of awareness about AI, and it’s for business leaders. I advise, I suggest, I recommend on the business leader listening to search on Google for AI. Or you can put a link in the text. I don't know.
This is a self-assessment where you can evaluate aware, and then you can determine if your company – Many now will say that the difference between low-tech company and the high-tech company is by the AI awareness. It can be. I don't know. Many are saying that. But basically, you can also assess how your company is engaged with the data. You can become more data literacy, etc. So one suggestion is to evaluate, to do a self-assessment with this tool. It can be interesting to understand better how the company engage with new technology like AI. Ist's very simple.
[00:21:54] RS: Antonio, that's great advice. I really love learning for you today. Many, many grazie for being here, and I really appreciate your time.
[00:23:16] AG: Thanks to you, Rob, for having me. Mille grazie. I hope everyone enjoyed this podcast, and thank you for your time.
[END OF INTERVIEW]
[00:23:28] RS: How AI Happens is brought to you by Sama. Sama provides accurate data for ambitious AI, specializing in image, video, and sensor data annotation and validation for machine learning algorithms in industries such as transportation, retail, e-commerce, media, medtech, robotics, and agriculture. For more information, head to sama.com.
[END]