China will ‘rule the planet’ until 2100 if it wins AI war against USA, expert claims

[shockedcanadian]

I've done a great deal of reading on the subject. There are some domains in which A.I has been oversold in my opinion, or the horizon is further away, however, on the whole, without much doubt in my mind, the statement is the title is accurate. THIS is the real global threat, NOT climate change. The West has to win this race.

I'm so interested in the subject that I'm partially through developing a basic Neural Network in Python (a set of Machine Learning algorithms), without any use of libraries that provide the math and underlying code. I was interested because it was important for me to understand the bare bones structure, a sort of deconstruction or reverse engineering that provides me a great deal of insight into how code libraries work, now and in the future. Everything I have learned is thanks to online learning and self motivation. Many books and even more videos/college courses provided.

Putin himself, that of former KGB stated that "the nation that wins the A.I race will rule the world." I've personally taken a great interest in the subject and underlying coding, processes and potential. Canada will be left behind in this arena, of this I am certain, but America must not. America must embrace A.I and definitely win this race and you MUST protect yourselves from China getting their hands on what California in particular is producing. They will do this by recruiting the best engineers, data scientists and practitioners, use of honey traps, spies, and, via theft.

It's not going to be useful for every industry, but believe it when you read, it is not going to slow down.

China will 'rule the planet' until 2100 if it wins AI war against USA, expert claims

THE country that becomes the world leader in artificial intelligence by 2030 will rule the world for at least the rest of the century, according to a new report.

Economy experts appear to generally agree that AI is up there with steam power, electricity and information systems technology, when it comes to the impact it will have on our planet.

Chief Economist for Europe and Central Asia at the World Bank, Indermit Gill, made the claims in his recent report for the Brookings Institution.

According to Gill: "Technological leadership will require big digital investments, rapid business process innovation, and efficient tax and transfer systems.

"China appears to have the edge in the first, the US in the second, and Western Europe in the third."

He thinks the country that gets the best grip on all those things will end up leading the world.

 

[Mac-7]

I've done a great deal of reading on the subject. There are some domains in which A.I has been oversold in my opinion, or the horizon is further away, however, on the whole, without much doubt in my mind, the statement is the title is accurate. THIS is the real global threat, NOT climate change. The West has to win this race.

I'm so interested in the subject that I'm partially through developing a basic Neural Network in Python (a set of Machine Learning algorithms), without any use of libraries that provide the math and underlying code. I was interested because it was important for me to understand the bare bones structure, a sort of deconstruction or reverse engineering that provides me a great deal of insight into how code libraries work, now and in the future. Everything I have learned is thanks to online learning and self motivation. Many books and even more videos/college courses provided.

Putin himself, that of former KGB stated that "the nation that wins the A.I race will rule the world." I've personally taken a great interest in the subject and underlying coding, processes and potential. Canada will be left behind in this arena, of this I am certain, but America must not. America must embrace A.I and definitely win this race and you MUST protect yourselves from China getting their hands on what California in particular is producing. They will do this by recruiting the best engineers, data scientists and practitioners, use of honey traps, spies, and, via theft.

It's not going to be useful for every industry, but believe it when you read, it is not going to slow down.

China will 'rule the planet' until 2100 if it wins AI war against USA, expert claims

THE country that becomes the world leader in artificial intelligence by 2030 will rule the world for at least the rest of the century, according to a new report.

Economy experts appear to generally agree that AI is up there with steam power, electricity and information systems technology, when it comes to the impact it will have on our planet.

Chief Economist for Europe and Central Asia at the World Bank, Indermit Gill, made the claims in his recent report for the Brookings Institution.

According to Gill: "Technological leadership will require big digital investments, rapid business process innovation, and efficient tax and transfer systems.

"China appears to have the edge in the first, the US in the second, and Western Europe in the third."

He thinks the country that gets the best grip on all those things will end up leading the world.

Its scary to think of china being the dominant superpower in the world

but it could happen if we allow it

 

[james bond]

THIS is the real global threat, NOT climate change.

Yes, it is the real political threat and not lies liberals believe lol. But you're right! I should try to build something like you did. Maybe it will give me answers to some of the posters here.

Can yours fool people into them not noticing that it is AI they are dealing with? Even if we can tell it's artificial, it's good. Then, we strive for the best artificial.

 

[Wuwei]

I've done a great deal of reading on the subject. There are some domains in which A.I has been oversold in my opinion, or the horizon is further away, however, on the whole, without much doubt in my mind, the statement is the title is accurate. THIS is the real global threat, NOT climate change. The West has to win this race.

I'm so interested in the subject that I'm partially through developing a basic Neural Network in Python (a set of Machine Learning algorithms), without any use of libraries that provide the math and underlying code. I was interested because it was important for me to understand the bare bones structure, a sort of deconstruction or reverse engineering that provides me a great deal of insight into how code libraries work, now and in the future. Everything I have learned is thanks to online learning and self motivation. Many books and even more videos/college courses provided.

Putin himself, that of former KGB stated that "the nation that wins the A.I race will rule the world." I've personally taken a great interest in the subject and underlying coding, processes and potential. Canada will be left behind in this arena, of this I am certain, but America must not. America must embrace A.I and definitely win this race and you MUST protect yourselves from China getting their hands on what California in particular is producing. They will do this by recruiting the best engineers, data scientists and practitioners, use of honey traps, spies, and, via theft.

It's not going to be useful for every industry, but believe it when you read, it is not going to slow down.

China will 'rule the planet' until 2100 if it wins AI war against USA, expert claims

THE country that becomes the world leader in artificial intelligence by 2030 will rule the world for at least the rest of the century, according to a new report.

Economy experts appear to generally agree that AI is up there with steam power, electricity and information systems technology, when it comes to the impact it will have on our planet.

Chief Economist for Europe and Central Asia at the World Bank, Indermit Gill, made the claims in his recent report for the Brookings Institution.

According to Gill: "Technological leadership will require big digital investments, rapid business process innovation, and efficient tax and transfer systems.

"China appears to have the edge in the first, the US in the second, and Western Europe in the third."

He thinks the country that gets the best grip on all those things will end up leading the world.

Usually ANN (artificial neural networks) focuses on a specific problem like pattern recognition, or chess. The major problem is training the network. The game of Go was trained by allowing the computer to play another computer, after many millions of games at lightening speed it could beat a human master.

Do you have a particular problem you are working on? Are you teaching it by show and tell? Are you using a multi-layer network? Back-propagation?
.

 

[shockedcanadian]

I've done a great deal of reading on the subject. There are some domains in which A.I has been oversold in my opinion, or the horizon is further away, however, on the whole, without much doubt in my mind, the statement is the title is accurate. THIS is the real global threat, NOT climate change. The West has to win this race.

I'm so interested in the subject that I'm partially through developing a basic Neural Network in Python (a set of Machine Learning algorithms), without any use of libraries that provide the math and underlying code. I was interested because it was important for me to understand the bare bones structure, a sort of deconstruction or reverse engineering that provides me a great deal of insight into how code libraries work, now and in the future. Everything I have learned is thanks to online learning and self motivation. Many books and even more videos/college courses provided.

Putin himself, that of former KGB stated that "the nation that wins the A.I race will rule the world." I've personally taken a great interest in the subject and underlying coding, processes and potential. Canada will be left behind in this arena, of this I am certain, but America must not. America must embrace A.I and definitely win this race and you MUST protect yourselves from China getting their hands on what California in particular is producing. They will do this by recruiting the best engineers, data scientists and practitioners, use of honey traps, spies, and, via theft.

It's not going to be useful for every industry, but believe it when you read, it is not going to slow down.

China will 'rule the planet' until 2100 if it wins AI war against USA, expert claims

THE country that becomes the world leader in artificial intelligence by 2030 will rule the world for at least the rest of the century, according to a new report.

Economy experts appear to generally agree that AI is up there with steam power, electricity and information systems technology, when it comes to the impact it will have on our planet.

Chief Economist for Europe and Central Asia at the World Bank, Indermit Gill, made the claims in his recent report for the Brookings Institution.

According to Gill: "Technological leadership will require big digital investments, rapid business process innovation, and efficient tax and transfer systems.

"China appears to have the edge in the first, the US in the second, and Western Europe in the third."

He thinks the country that gets the best grip on all those things will end up leading the world.

Usually ANN (artificial neural networks) focuses on a specific problem like pattern recognition, or chess. The major problem is training the network. The game of Go was trained by allowing the computer to play another computer, after many millions of games at lightening speed it could beat a human master.

Do you have a particular problem you are working on? Are you teaching it by show and tell? Are you using a multi-layer network? Back-propagation?
.

Not a particular problem, though I do have a particular project I am interested in that will be quite challenging to construct involving audio (though not NLP). I have only begun to research it and I've read a whitepaper or two on it. I am confident it can be done. The problem is relying on CPU power, I can only use GPU in the cloud (Azure among a few other accounts) and I refuse to pay for it and I would still be limited by my computers computation and transfer speed to the cloud itself.

Thus, I only have limited computational power, which is vital especially if one builds a deep network. For the idea I have, there's no question I don't have the computational power. Some NN's on small data sets might only take a couple of minutes to run, others could take hours, or even days on CPU. The idea I have would take days, even on a decent GPU. So, we will see...

I know the case well that you speak of regarding Alpha Go, as it involved Reinforcement Learning, possibly the most promising of Machine Learning and the area I find most inspiring. It was groundbreaking because unlike other machine learning tactics, RL isn't Supervised Learning (where you know the "labels" or what I think of them as "dependent variables" from my University day) where a computer tries to give you the correct output (which you already know via a dataset), or Unsupervised (where you don't know the answer but are really allowing the computer to find patterns on it's own), but reinforcement learning is truly artificial intelligence in the sense most people think of. Or at least how I did before I investigated the subject.

With Reinforcement Learning, the computer, Alpha Go as you referred, played itself millions of times, with only one side always winning, and it gradually improved as it beat itself each time, against a slightly better opponent each time. What was amazing is that it did this without any understanding even of the rules of Go, and then Chess. It basically taught itself the games by seeing the patterns and receiving "rewards" for in essence, winning via a series of correct moves. Within 4 hours of training it learned to beat the best computer chess player Stockfish and was much better than any human Grandmaster that God ever created. It's also been applied to video games, drones, A.I etc. You reward a machine for a "good" decision making and thus, you can in essence, generalize the model across different games and even robotics as Boston Dynamic has done with robots (not with the same software as Alpha Go of course).

When you say "multi-layer, I suppose you are referring to a "Deep" network of multiple hidden layers and nodes, or Deep Learning NN's. Yes, I have run many such networks, on relatively small datasets. Including across different Neural Network types, such as Convolutional Neural Network, from a course I took and even Recurrent Neural Networks. There are many such examples out there, and some smaller datasets built into tensorflow call MNIST. Old hand written numbers which the machine learns to distinguish, also quite a few datasets I've found on Kaggle, a competitive machine learning site which I have cleaned up and messed around with.

The neural network type has become more specialized from what I've learned. For instance, CNN's are best used with photo data where you have a "slide" feature in the code which basically reconstructs the pixels into vectors and can really efficiently work with this data in ways a standard Feed Forward Network would take many times longer to read. RNN's are best employed data such as audio, language, text, real time stock prices and weather data. Without getting into details, you have a constant information back loop within the network that allows it to work even with time series data.

Oh, and then there is Long Term Short-Term Memory NN's (LTSM). Even more useful for real time stock data and time series data stuff. It's quite intricate and interesting as well with a memory gate that forgets and saves computational and storage power. I'm not too experienced with that, though I took a course on it and understand the general concept.

As for the NN I am working on, it is quite small. Just a single hidden layer with a few nodes. I am just trying to learn the nuts and bolts of the math behind the NN from the core syntax rather than using tensorflow or keras which does all of the difficult calculation code for you. It's a real learning experience and something I wanted to do to improve my understanding.

Back propagation is a part of the NN yes and I have incorporated it. All that means is the feed forward NN after one epoch through the hidden layers (an attempt through the data) runs back from the output layer back to the input layer while adjusting the weights and biases to decrease the cost function (essentially the error between the computer prediction and the correct answer). It is really the heart of neural networks and the guy who devised this is from the University of Toronto (we do have a few smart Canadians who stick around, hah).

Once you develop a decent NN model, the trick is basically tweaking the hyper parameters such as the learning rate, number of nodes and hidden layers (you can use drop off for instance to alter the way the nodes interact to basically get out of a situation where your machine is stuck and no longer learning). I've taken a course online from Andrew Ng on tuning hyper parameters and it really varies depending on what type of algorithm your model employs (Known Nearest Neighbor, Linear Regression, SVM etc).

If you are interested you can mess around with a great site to give you an idea of how it works. Check this out. If you are into this sort of subject, it's a great way to visualize is if you haven't already. You can't apply your own dataset unfortunately, but it gives you a good idea of how it works with a dataset.

Tensorflow — Neural Network Playground