AI isn't the solution to everything
A year ago no one could shut up about block chains and everyone was scrambling to find a way to integrate it into everything. Now, we're all about machine learning and AI.
I'm going to say something I feel like I say far too often; no solution is universal.
Personally, I'm not even sure what field block chain is valid in. All I can say is, it that isn't cryptocurrency.
AI and machine learning definitely are applicable to a great many more things. But, AI and ML are like short cuts to reasonable answers. And, today they require a LOT of smarts to get right.
In my opinion, there are simply a lot of fields where AI offers no advantage over classical computing methods, or where the advantage isn't worth the cost.
What sort of tasks is ML good at today? If you can put data into columnar grid (like an Excel spreadsheet), and you know which data points are relevant and exactly how they are relevant and can tell a computer that. BUT, there is too much for a human to process... then you have a problem modern machine learning is good at.
I'm simplifying the living shit out of this. Firstly, the arrays something like TensorFlow can manage are N-dimensional, not just two. And, in theory, you could probably model anything you want in such a structure. The problem is really about adding the other two pieces into the equation. You need to be able to build a set of rules which understands your data and the effort to do so must be less than the human investment required to derive the answer on your own.
I would say it requires a problem of a reasonable level of complexity, but which can be broken down into a series of simply determining factors and the determination needs to happen MANY, MANY, times.
I also want to talk about my stance on block chain. I said that crypto currency was NOT to ideal application of block chain. I'm sure that will rankle some. The hear of block chain is a decentralized ledger. For the state of the data to be accepted and remain accepted it must agree with a certain percentage of the instances of the ledger. As such, if someone controls a majority of the ledgers or whatever percentage is required by the block chain (unlikely to be 100% as it needs at least SOME fault tolerance), then they can control the data.
In other words, you only want to use a distributed ledger system if the application is one in which everyone has a vested interest in the ledger's accuracy. Currency... not one of those scenarios. Stealing money is literally something humans do every day. Eventually, someone will find a way hijack even a ledger as widely distributed as Bitcoin. And, even if I'm wrong, people will try and will only be countered by other people trying harder.
It is hard to imagine a valid application of block chain, but they do exist. But, I would say most sane applications are distributed only in the physical sense. A single owner of all of the data, but distributed to add redundancy and speed. A database; Transactions uniquely verified. Data duplicated across every node. Multiple nodes validating the state of the data. No one contributing to the block chain wanting it to be compromised.
I'm going to say something I feel like I say far too often; no solution is universal.
Personally, I'm not even sure what field block chain is valid in. All I can say is, it that isn't cryptocurrency.
AI and machine learning definitely are applicable to a great many more things. But, AI and ML are like short cuts to reasonable answers. And, today they require a LOT of smarts to get right.
In my opinion, there are simply a lot of fields where AI offers no advantage over classical computing methods, or where the advantage isn't worth the cost.
What sort of tasks is ML good at today? If you can put data into columnar grid (like an Excel spreadsheet), and you know which data points are relevant and exactly how they are relevant and can tell a computer that. BUT, there is too much for a human to process... then you have a problem modern machine learning is good at.
I'm simplifying the living shit out of this. Firstly, the arrays something like TensorFlow can manage are N-dimensional, not just two. And, in theory, you could probably model anything you want in such a structure. The problem is really about adding the other two pieces into the equation. You need to be able to build a set of rules which understands your data and the effort to do so must be less than the human investment required to derive the answer on your own.
I would say it requires a problem of a reasonable level of complexity, but which can be broken down into a series of simply determining factors and the determination needs to happen MANY, MANY, times.
I also want to talk about my stance on block chain. I said that crypto currency was NOT to ideal application of block chain. I'm sure that will rankle some. The hear of block chain is a decentralized ledger. For the state of the data to be accepted and remain accepted it must agree with a certain percentage of the instances of the ledger. As such, if someone controls a majority of the ledgers or whatever percentage is required by the block chain (unlikely to be 100% as it needs at least SOME fault tolerance), then they can control the data.
In other words, you only want to use a distributed ledger system if the application is one in which everyone has a vested interest in the ledger's accuracy. Currency... not one of those scenarios. Stealing money is literally something humans do every day. Eventually, someone will find a way hijack even a ledger as widely distributed as Bitcoin. And, even if I'm wrong, people will try and will only be countered by other people trying harder.
It is hard to imagine a valid application of block chain, but they do exist. But, I would say most sane applications are distributed only in the physical sense. A single owner of all of the data, but distributed to add redundancy and speed. A database; Transactions uniquely verified. Data duplicated across every node. Multiple nodes validating the state of the data. No one contributing to the block chain wanting it to be compromised.
Comments
Post a Comment