Andrea Amenta - The Hidden Cost of AI: And How Blockchain Might Save It

Good morning everyone and welcome back to stage like this better. Okay, welcome back to day two of Napolit. Hope you had a fantastic time yesterday. Hope you had enjoyed the uh initial event side event as well and you didn't make it too late uh yesterday evening and we were super excited this morning in this beautiful church to just get started talking about blockchain and uh and AI. My name is Kat and I will be your host for uh for this morning and we will get started straight away with our guest u guest speaker.

Uh our first talks uh really really explores one of the biggest hidden cost in tech today which is the environmental and operational impact of AI and data dependency. So to lead this talk is going to be Andrea Mena operation manager at dig now. Welcome.

Thank you. Hello everybody. So let's start with a question. Who of you has used it any AI gener generative generative tool and in the last days? How many of you know the data that PI is using is coming from?

Do you know if this data is reliable or not? Nobody. Yeah. That's why we have a trust problem in EI. And today I'm going to explain to you why we have this trust problem.

Everyone is using EI today. A lot of tools, JPT, Gemini, Lyama, a lot of tools are already in the market, but nobody knows the hidden cost be behind them. And today I'm going to explain you why. I am Hello everybody. Yeah, thank you for being here and thanks Ethereum Naples to host me.

I am Andrea Amenta. I am originally from Italy as well, but I've been living in Barcelona in the last 10 years. And I'm here to introduce you this topic. Okay, that is something that uh really uh touched me. The last 10 years I've been working Hello.

I've been working in um tech in uh from data management to software development and I've been managing projects all over the world and in the last years I've been into tech okay but more innovative tech like EI and blockchain and I'm always been attracted by the intersection between EI and blockchain that's why I joined now that is a small startup that is actually building a data infrastructure for AI, a verified one using blockchain and the II. But let's see the trust problem. As I mentioned before, we have a trust problem in EI. Okay, why we have this trust problem? First, some numbers.

Okay, some research I've been doing in the last years. First of all, we have an analysis from Gartner that is saying that 85% of EI projects is not delivering ROI. What does it mean? ROI is return of investment. So 85% of projects doesn't deliver anything.

So they are going to fail. Another shocking analysis is from McKenzie that is giving us 22% of projects don't give any value to anybody. It means that one every five projects is going to fail. And another analysis that is shockling shocking as well is that data scientists someone some people that are the smartest in the world are actually wasting 80% of time in data analytics data cleaning and data verification. while they can do something else like innovate like create something new but instead they are just wasting their time.

The trust problem bring us to three hidden costs that are related to EI. Okay, the first one is efficiency. The efficiency cost means we are using a lot of computation to run EI servers to run computation and so on. And this efficiency cost is really impacting in our world as well. The economic waste because we are not talking about uh monopoly money.

Okay. We are always talking about dollars, crypto, whatever you want to uh mean economic value. But as well we will see that trillions are wasted every year for AI that is not efficient. And as well the environmental impact that is something that nobody almost nobody is talking about because people are always pushing for money for performances and so on. But the environmental impact of this technology is high impacting.

But let's dive into these three hidden costs. Sorry I am not uh very good with this. But the first one the efficiency drain as I mentioned before we have a problem with computation we are using computation for running AI and so on but let's deep dive on it an analysis from Google cloud so one of the most the one of the biggest data provider in the world is saying that from 15 to 20% of data all around the world is waste So we are using this data that is not useful at all and the AI processing is based on this data. So we are processing data and data and data but is useless. So we are wasting all this AI computation and all this data as well.

Google Cloud says that software engineers are actually using 2.3 hours of the time to clean the data and create verification systems. That means that the 30% of their time is wasted using verification and cleaning data as well. Something impactful we know GPT that is for end users. Okay.

But as well there are AI technologies that are for businesses that could be like uh fraud detection or supply chain management and so on. So in these systems the AI responses do to the verification layers that are redundant are from 12 to 18 seconds. It means if you detect a throat, you need to wait 18 seconds to get a response from who is the guy that stole you the money. So, as you can imagine, it's not that wonderful, you know. Next one, let's deep dive on economic cost.

We have a big analysis from IBM that is one of the biggest tech company in the world. Okay, that is saying that the AI cost for uh the businesses on on average is 12.9 million per year. That means that on average so we are not talking only about Fortune 500 companies big corporation we are we are talking about average. This average means that in the mid businesses these 129 million annual cost are actually their entry budget.

So all this money that could be invested in something else are actually wasted not wasted use on EI. The wasted value is the $3.1 trillion that is actually wasted because of redundant data because of data verification and so on. This 3.1 million 3.

1 trillion we are talking about trillions not billions are just for make some example enough to hand the hunger in the world two times or give us a gift an iPhone to every person in the planet or funding the NASA budget for 130 years so it's actually a big impact on money And as well another analysis from EBM say that 40 to 50% of the budget is spent in data analytics. So all this budget and um all this money are spent in that analytics that is needed to clean the data that is connected to AI. But now let's go to the environmental cost of this. Okay, that is something that really touched me. Okay, first of all, let's understand how much energy you are wasting for EI.

It is estimated that from two to 5% of the global electricity all around the world is used to power AI. This is a data from 2024. But just to make you like to give you a bit of context, this amount of energy is around 500,000 to 1,250,000 gigawatt hour per year. This amount could be connected like two countries. Okay, we always talk about uh Bitcoin uh electricity waste and so on.

But as well is consuming a lot. For example, to make a comparison, we see that EI waste the same amount of energy that is using Russia to power the entire country all around the year. And uh in 2030 if we continue in this way without the efficiency checking without um control of the data is it is projected that the global electricity that the EI will consume will be the 10% of the global electricity all around the world. Another impact on data is related to the emissions. Okay.

So we have the emissions first of all for the training. Okay, I put an example of G GPT4 that is a very known uh AI generated tool. GPT4 for every training consumes 15,000 tons of CO2. And I don't know if you know but yeah uh GPT4 is trained several times per year. But as well we need to consider the operations of GPT4.

And every day of using GPT4, it is compared to running uh 100 cars every day no stop or 50 flights from New York to London or powering and yeah the consum the consumption of 45 US homes per day. So that's something impactful. But we can see as well some graph that inefficient data of VI protocols and as well the running of EI is estimated to contribute to from two to 3% of global emissions all around the world and that's comparable to the aviation industry. So the same amount that the global elevation industry is consuming right now is more or less the same as the inefficient data processing OBI and from this we can see as we say before that Google cloud say that 20% of this uh energy and this um the computation is completely wasted and that's what I control the vicious cycle. Okay, we have problem with the eye.

We have a trust problem with the eye. Okay, we are in we have inefficiency. We have redundancy and so on. So what is happening? What I call the vicious vicious cycle.

Vious cycle starts from bad data. Okay, the bad data is entering in the system. AI is hallucinating. We have problems because the data is not good. The data brings to distrust.

It is uh there is an analysis that uh 24% of uh direct directives are not datadriven. So they don't trust on the data that is provided by the anal data analytics teams of or from AI that lead to the leads to assign task to their people to their colleagues about verification. So 62% of um of development time goes to API testing and verification. So we create layers and layers of verification because people don't trust their data and at the end this brings to a lot of resources that are wasted of people of energy of potential and as well of time and the solution is here. Okay, we are in um web three blockchain Ethereum conference.

So the solution is web three. Why I'm not going to talk about Bitcoin speculation, NFTs and so on? Because we know that the media are being uh very bad with blockchain and web three in general because they are always associating blockchain with these bad things. Okay. But we know that uh the reality is that blockchain will be used as a cryptographic consensus connected to ma to mathematical trust.

So we can have trust without institution trust without people trust with the centralized people. So this could be a solution that the data that we input in our systems the on the system that are all around the world could be verified once with um cryptographical ash ash graph and could be stored and could be used by the EI systems all around the world. there is a study that uh we have a lot of economic opportunity. Okay, first of all I wanted to focus on an industry supply chain industry. There is a research from McKenzie that if we verify uh blockchain data in supply chain we can have uh a a reduction of cost from five to 10 billions only in the supply chain business.

But as well if we connect again to EI we see the EI market will hit 1.8 trillions by 2030. And if we verify the data if we confirm this data if we have data that is valuable we can reduce up to 50% of this cost. This mean 900 billion that could be used to I don't know solve something else in the world innovate cure the cancer so something different and valuable that could be use it instead of being wasted and I see I foresee two futures for us and for everyone. Okay the future one is the status quo.

Okay the status quo is where we are right now. So we continue as we are. We continue wasting. We continue with inefficiency and we continue to waste trillions of money. We continue to pollute the planet.

But there is another future that is possible that is where we use blockchain to verify the data and where we have efficient all around the EI verification and the sources. So which future do you want? of course is the second future because everyone wants a world that is better what you can do in your daytoday life. Okay, I don't know how many of you are working in software development. Okay.

But if you're working in any AI uh company or blockchain company and so on also datadriven companies always audit your AI systems also question every redundancy that is in um in uh in your processes uh explore verify data solutions and as well make efficiency requirements. to track KPIs, understand better the data that they're using and so on and as well for different stakeholder this solution can be used first of all yeah for uh developers because next time yeah that you build something that is uh distrusted is nice to ask this question is the data is coming that I using is trustworthy or not and as well for investors is uh something good to understand because the next unicorn will probably come from data verification systems and not from that uh that is part for executives as well because you need to look at your return of investment and as I mentioned before most of the EI companies are actually failing and for everyone because it's something that is affecting everyone nowadays. And now I will introduce you DIG now. Okay, that is the startup I'm working for. DIG now is actually a small crypto NDI startup.

Okay, we are creating a data infrastructure that combines both AI efficiency, a decentralized community of humans that is used for accuracy and blockchain for permanence. So as I mentioned before, we use the blockchain to verify the data at the end. Just to make you understand how it works, okay, we collect data from many sources. It means it could be from uh databases, from uh uh web scrapping, from uh all over all over the world. So wherever there is data, we can collect it.

We first have a EI assessment. So this uh data is uh evaluated by our AI engine. This AI engine give a crypto score to this data to understand if it is valuable or not. If the data is not valuable, it will pass through a decentralized community of humans. Okay.

We have a platform that is called task where the people enter and will get rewards to validate and verify the data and at the end when the data is verified 100% we storage it in the blockchain with a cryptographic hash and this data once is uh created and validated is uh available through API to the people to the companies to the protocols that need it. In here there is more or less the technical architecture. Okay. So as I mentioned before we collect the data from uh wherever it could be. So from the internet from the data providers from the humans and so on.

There is the verification layer that the first step is the I supervisor that assign a trust score. If the score is high it's already saved in the smart storage. If it is low, there is the the centralized community of people that is verifying the data. And once the community is verifying the data and the score is stored again and then this data through smart contract is uh stored in the blockchain and then delivered through API or an MS MCP servers. We actually launched a pilot with a customer in the defi space.

Okay. And we in just two months of operation, we processed and verified over 15,000 data points. Thanks to the AI verification layer and to the community of people, we have already 5,200 people that are verifying the data as well. We had a 99% of accuracy of the data that we provided to our uh customer and as well we saw that uh we saved 67% of time by reducing verification of the data from the customer that we are working with. How we are working right now we are focusing on crypto projects.

Okay. because we know that uh web 3 needs trust. So we are verifi verifying project data for crypto project mainly at the moment because as I mentioned before we are quite small. So we are verifying data related to teams to toomics to all the data that you can imagine about crypto projects but the idea is in the future to go through other markets. Okay, that could be like financial data.

Imagine like uh financial data that is verified once and could be accessed by all the brokers all around the world. Healthcare as well could be a great use case because you can imagine some a patient that is going to a hospital and has already the data verified from that hospital and then can go all over all around the world to have the same data uh shared it with everyone and as well in the supply chain because having uh verified data for example order tracking that is verified could be verified and connected to everyone in the supply chain business and now yeah if you want to know more you can explore about dign now of course so as I mentioned before we are already 5,200 verifiers that are part of our decentralized community of humans that are verifying data we grow more or less by 100 people every every week and we are verifying 1,000 data points per day already. So we are not building just a product okay but we are trying to build a movement okay that where we have a decentralized community of people as well and I don't know if I mentioned before but these people will get rewarded with tokens and oh yeah as I mentioned before we need to stop this revolution okay because to compensate all the three hidden costs that we have we need to make actions Okay. So we need to stop accepting inefficiency. We need to start building AI that works and as well we need to join the movement of building the trusting layers for AI.

And that's it I think from my side. Uh if you want to connect you can uh connect with me in telegram link edin or if you want to explore more about ding now this is the QR code for link tree and that's it from my side and now I'm open to questions if there are