How the BRAIN of an AI Works: Shockingly Simple but Genius!

this video is sponsored by Masterworks you've heard this from some famous incredible people something to the effect that AI will take over the world and will spell the end of humanity is this a sensible forewarning of things to come or an alarmist overreaction on a harmless technology part of the problem is that very few people understand how AI works and if you don't really understand how it works then it's going to be an unknown and we fear the unknown in this video I'm going to try to Ally some of this fear by explaining how AI really works in detail I made a more General video on how AI apps like chat GPT work if you haven't seen that I highly recommend you view it first to get a general idea of what's going on but in this video I'm going to get into much more detail about what's really going on by showing you how neural networks work which is the essential element of AI and by the end of this video I'm hoping the unknowns will become more of a known and you'll be able to understand just how the quote-unquote mind of an AI actually works stay tuned because that's coming up right now [Music] foreign first you should understand that an artificial neural network also called neural network for short is at its core a mathematical equation no more no less a powerful neural network is typically a very complex equation but nevertheless it's just math the term neural network comes from its analogy to neurons in our body in our brain we have around 86 billion neurons these neurons are what give us the ability to do anything they send the signals to the various parts of our body to act and they send signals to our brain that can enable us to think and be humans neurons and neural networks serve an analogous role and that's why the nomenclature is similar but let me stress this from the start neurons in our brain are much more complex and advanced than neurons in an artificial neural network artificial neurons do not work the same way the analogy with artificial neurons is limited because it applies to the way that they are networked not to the way that they function like in the brain we connect multiple neurons together and form a neural network which we can train to perform a task if we continue the analogy a little further and look at a biological neuron and that in an artificial neural network we see why it makes sense to draw this comparison a biological neuron has a cell body with a nucleus this is the core of the neuron where the processing takes place around the cell body there are dendrites which are are signal receivers they take in inputs which the cell body can then process on the other end there is an axon which ends with the terminal axons that can pass on a signal to the dendrites of the next neuron in a similar way a neuron in a neural network is a processor which is essentially a function with some parameters this function takes in inputs and after processing the inputs it creates an output which can be passed along to another neuron the input is analogous to the dendrites and the output is analogous to the terminal axons the function in the neural network is like the cell body like neurons in the brain artificial neurons can also be connected to each other via synapses this is what a neural network is its neurons connected with each other receiving inputs and passing along outputs in a complex Network while an individual neuron can be simple and might not do anything too impressive it's the networking that makes them so powerful and that network is the core of artificial intelligence systems so how do these artificial neurons work well the essence of an artificial neuron is nothing but this simple equation from elementary school where X is the input W is a weight B is a bias term and the result our output is z of X what does this mean essentially any input X is being Modified by multiplying with a weight and then a bias is added in order to get a result or output this allows the AI system to map the input value X to some preferred output value Z of X how are W and B determined this is where training which is what I talked about in a prior video comes in we have to train the parameters W and B into the AI system such that the input can be modified into the appropriate or correct output so how is the training done let's do a very simple example so that you can get an intuitive idea for what's happening let's say you have five dollars you're trying to determine can I afford a cup of coffee the coffee costs three dollars so the inputs are five dollars representing the money that you have three dollars for the cost of the coffee and the question can I afford this you know the output should be yes and you're trying to train the system to give you a yes result initially the system may choose an arbitrary value for the weight and bias you enter the inputs and see what the output is if the output is a no the system adjusts the weight and bias and keeps iterating until the correct answer yes is given in the output then you change the input to two dollars and do this process again you know the answer should be no this time so if the input is yes then the system adjusts its weight and bias again until the output is no similarly the inputs are changed over many iterations until a system output gives the correct answer over a variety of inputs this is what training is all about note that when we have multiple inputs the function Z of X representing the neuron is modified to be a sum like this if you think about how our mind works we sort of do this same thing we take in all the inputs and we process this with the math that we have learned and numbers we understand with this we can figure out what we can afford and what we can't essentially this is what a neural network can do too but just in a more explicit way using purely mathematics this leads us to a problem with our equation the case I just showed is a very simple yes or no output our function Z of X which is a simple linear equation can provide the correct output but what happens when the boundary conditions are not linear when the output cannot be so simple and clear-cut for example what if you're a mortgage lender at a bank and want to use an AI system to determine how much money you can lend a particular customer in this case many inputs would need to be taken into account things like the person's income housing market conditions in the area the loan term amount of cash deposit and a host of other factors in this case the output would not have such a distinct linear boundary the output might look more like this chart on the right problem like this is solved by using a mathematical trick which we call an activation function this activation function represented by a in this formula is a mathematical function we apply to our output Z of X there are different choices for an activation function but a good example is the sigmoid function an example of a sigmoid function is the following this mathematical treatment allows for a greater variability in the decision boundary by introducing non-linearity to the linear formula Z of X the following images show how linear decision boundary can be modified for more complex decision making and with the addition of an activation function our flowchart is modified to look more like this where the activation function sigmoid function in this case is added as an additional step in the process before we get a final result so with an appropriate activation function there's nothing stopping an AI from answering much more complex questions keep in mind that with a single neuron we can't do a lot and you need more neuronsville con for the complexity of deciding for example how much money to loan a customer to buy a house I used a single neuron to give you a general understanding since you can see the fairly simple math behind it but it's not complex enough to do much by itself even with the power of non-linear activation functions but the combination of several neurons that form what we call a neural network is where the power lies so when we make complex networks with thousands if not millions of neurons like that used in Chad GPT or Bard we can obtain some incredible results as you've probably found if you've used either of these apps now I'm going to tell you something about this neural network that you might find surprising and perhaps even scary but before I tell you that let me first touch on another subject that's also scary for many nearly one in five people can have their jobs replaced by AI entirely the 300 million people across the globe equivalent to almost the entire population of the United States it's a scary future for many especially since we're dealing with ramp and inflation and one negative headline after another this is why some of the biggest money managers in the US have been diversifying outside the stock market into luxury assets with far less correlation like Fine Art today's sponsor Masterworks offers this diversification strategy which had once been one of the most exclusive to ordinary investors Masterworks is an art investing platform they buy the art outright register it with the SEC and then break it into investable shares net proceeds from its sale are distributed to its investors Masterworks allows almost anyone to invest in fractional shares of Contemporary Art since their Inception they've sold over 45 million dollars worth of artwork from Legends like Picasso Banksy and Monet and so far each of Masterworks exits have had positive net returns for their investors now you should know that historical returns are not a guarantee for future returns I'm not a financial planner and you should do your own due diligence before investing any money anywhere and like any other investment there's always a risk of loss Masterworks has over 700 000 users and their art offerings usually sell out in hours which is why they've had to make a wait list but my viewers can skip the line and get Priority Access right now by clicking the link in the description I just found master to be a great way to diversify for myself and I think it's worth your consideration now back to what some people fear most about neural networks when I talk about training the network to obtain certain known results the simple case was the example of whether we could afford to buy a coffee or not and in such a case there's not much processing that needs to happen in these various notes but for more complex results such as how much money we can loan to a customer for a mortgage or other complicated demands the adjustments that the system makes in the training process is a bit of a black box what do I mean by this when we train the system using known inputs and known outputs we're having the system self-adjust its internal networking results from the various nodes to match what the known result should be this is the training process but how exactly the network adjusts the various layers of intermediate outputs to achieve the final output we want is not really known in other words we don't know how the system is adjusting all the intermediate layers that are not the input or output layer the input or output layers are known but the stuff inside is not and so these intermediate layers of neurons are called hidden layers the hidden layers are a black box we don't really know what these various layers are doing they are performing some transformation of the data which we don't understand this is not to say that we can't find the calculated results from any particular node we can because they're just doing mathematics but the result will not make any sense to us it would just appear as some arbitrary number which the network has determined to be a suitable intermediate in order to correctly obtain the final result in the output layer it is the combination of numbers of the various in individual nodes of the Hidden layers in the network that results in the power of AI so we have a huge black box which does complex Transformations by stitching together the simple linear equation with an activation function to create complex highly non-linear outputs which Maps the given inputs to that output some people consider this hidden or unknown processing a scary problem because they say since the system is doing things we don't understand it could start to do things on its own and pose a threat to humanity I find this argument to be unconvincing myself because these networks are doing only what they're trained to do this is not to say that they couldn't be trained to behave in more complex ways in the future but with what we know today there's no way these networks can do anything they're not trained to do on their own a bad person I suppose could train an AI to do bad things but how is this any different than hackers we have today that use computers to do bad things no AI Technology based on neural networks today could become something like Skynet in The Terminator movies that suddenly becomes conscious and threatens mankind I think the real threat of AI is in its power to do things that humans do today and thus potentially eliminate jobs this is in my view the main thing we need to contend with and that's going to happen whether we like it or not EI might eliminate some jobs but it will also help us do our jobs better as well as create some new jobs but it's unlikely to kill us I think Elon was and others at least for now should be able to sleep pretty soundly at night I'll see you in the next video my friend foreign foreign

Channel: Arvin Ash

Views: 116,689

Keywords: artificial intelligence, ai, machine learning, neural networks, neural network, artificial intelligence tutorial, artificial intelligence explained, machine learning tutorial, how neural networks work, how do neural networks learn, how does an ai work, how does an ai learn, what is a neural network, what is scary about AI, the threat of artificial intelligence

Id: NxTTXuUl-Lc

Length: 14min 33sec (873 seconds)

Published: Sat May 27 2023