The transcript is also available at Temi.
Michael (00:35):
This is Episode 1: Inception.
Cliff (00:39):
We can start with a question from the community. This is an username Bard: "can you give us a basic primer about your utility tokens and how they work? And, how does Hadron compare to other similar tokens out there, like Filecoin?" I think the easiest way to put it is that Filecoin is for storage; Ethereum is for smart contracts; Uniswap is for DeF; Theta is for video; and Hadron is for AI. That's really simplified, but that's probably the easiest way to think about it. AI is kind of frequently not characterized very clearly in the media. We'll go into that more in the later podcast. For now, we're going to be just using the broadest and biggest bucket, which is AI, as opposed to specifically ML, or machine learning, or deep learning.
Cliff (01:32):
We'll just put everything into the AI bucket. There's some pretty famous examples of AI, like AlphaGo. AlphaGo beat Lee Sedol in this ancient game of Go. And a lot of people didn't think this was possible with current technology because the number of different possible board layouts for Go exceeds the number of atoms in the known universe. It's not really a problem you can brute force and solve. You really needed an advanced AI approach. That's what Google figured out in pretty spectacular fashion. Another example of AI, that's in common use is, for advertising. It's used by all the biggest websites and apps out there to optimize what kind of ads you see. Companies that use this, of course, Google, Facebook, Tencent, Baidu, they're used to great effect. And of course there's also AI used to help control drones and self-driving cars like Teslas.
Cliff (02:25):
I say "help" because a lot of it's used for inference to detect what kind of objects that they're seeing, but the actual control mechanism is often used using conventional programming. That's also changing a bit. One common theme among all those examples though, is that they're developed and controlled by a very small group of elite companies and very knowledgeable engineers and scientists. A lot of the innovation in the world happens that way, but we think that there is value in a decentralized approach to help democratize the tools that you need to build these AI systems. Besides these famous examples, there are a lot of use cases that can help individuals and communities that may have been overlooked. For example, AI chat bots can be trained in local dialects or local languages to help guide people on where to obtain safe vaccinations; AI pest detectors and weed removing robots can help protect farmers' crops without needing to buy toxic pesticides and herbicides from big chemical companies. Or, you can build an AI to automatically help diagnose common injuries or infections in a local underserved population.
Cliff (03:31):
And, all of these things are actually already being built. I think the problem is the tools to build these and the tools needed to train these systems to very high accuracy are not readily available. A lot of them do require maybe living in a certain place or having access to a certain kind of payment system like credit cards. That's one of the reasons we think it makes sense to decentralize this and to make it more available to a broader population.
John (04:02):
That's a great point. Cliff. One thing I'd like to add is that Uniswap covers a subset of DeFi, specifically just DEX.
Michael (04:09):
Cliff just talked about AI. Ev, can you go into some more detail about what we mean when we say AI?
Ev (04:16):
Sure, Michael. AI, machine learning, deep learning -- these terms kind of are used interchangeably. They actually mean subtly different things. AI is probably the broadest term. It really encompasses anything that kind of simulates human intelligence and it's been around for decades. It could be something as simple as when you call your credit card company, and they say "press one to hear your account balance; press two to make a payment". Machine learning is the more recent revolution that has happened probably in the last decade or two. And it's enabled a great leap forward with new capabilities like self-driving cars. So the way to think about machine learning is that the contrast to it is logic based coding. That is the traditional way of coding that you may already be familiar with, where a software engineer will write all the lines of code and they specify all the parameters and a logic that goes into it.
Ev (05:15):
Most of the apps you already use from the spreadsheets on your computers, to the games in your X-Box, they're all probably logic based coding. The thing that makes machine learning so exciting is that now this logic and these parameters are actually set by the computer based on training examples that we humans provide. This enabled an entirely new revolution in software programming because it's a qualitatively different way of solving a problem. And the result is that problems that traditional logic tended to struggle with can now be solved through machine learning. These kind of fuzzier problems, like what is spam, for instance. Deep learning is just a model with a lot of different layers, hence a lot of depth. So it's really a subsegment within machine learning. We'll go into what all this means in terms of layers and depths in future podcasts. And finally, Hadron is really about AI writ large. We include all the different facets of AI, including some of these older technologies before machine learning. Because they're still useful for different applications, just like logic based coding is a little bit older, but it's still incredibly useful, and it's going to be around quite a long time.
Michael (06:31):
On the next episode of the Hadron Multiverse, we chat with Darick about the challenges of artificial intelligence development in ways to build better algorithms. Followed by an episode, discussing tokenomics in depth. Find us on iTunes or your favorite podcast platform by searching for the Hadron Multiverse!