Geoffrey Hinton, Yann LeCun, Yoshua Bengio- these three names don’t need a separate introduction. Best known as the godfathers of AI or godfathers of Deep learning, they came together during the 34th annual AAAI conference, speaking to a group of reporters about many AI-related topics, from ethics in the Artificial intelligence world and how to fix the flaws in the field. And these are the highlights of the entire discussion.
- Geoffry Hinton expressed his thoughts about how the increase in the uses of dedicated chips is going to enhance the performance of neural networks. With larger on-chip memory, neural networks would not have to face the technical limitations with reusing weight. LeCun, on the other hand, expressed his concerns with data batching. Graphic Processing Units make the data pass through the neural network in batches, which ultimately leads to complications. Yosua Bengio spoke in detail about the problems with sparse computing. According to LeCun’s predictions, all these problems will be solved with the emergence of the new and enhanced hardware.
- During the ongoing discussion, Hinton noted how modern neural networks are tiny compared to the concept of really big neural networks. The constant progress on the hardware can make it possible in the future to create much bigger neural networks with parameters larger than just ten billion. When asked about how the general public is mistrustful of the weird ways machines operate, as evident in the adversarial examples, Hinton explained how the imperfections of classifiers increase adversarial examples. It was also pointed out by LeCun how animals and machines alike can be fooled similarly.
- Last but not the least, Yann LeCun was asked about Facebook’s use of facial recognition algorithms. While talking about how these algorithms can be used for both good and bad, LeCun also talked about how a lot of these uses depend on the democratic institutions of the society. Yoshua Bengio expressed his opinion, saying that scientists in such cases should speak out about the different implications of these technologies as the governments in many countries want scientists to openly join the conversation.
Discussions as these bring about the best ideas of all in the AI mechanism. However, before we can actually go in deeper and explore the limitless possibilities of these ideas, we have to get a better understanding of Neural networks and AI.
Neural Networks: Human Attempt At Playing God?
If you didn’t know already, Neural Networks are large architectures that help artificial intelligence systems to sort and classify data. While it sounds very technical and in general boring, the matter of artificial neural networks is one of the most fascinating in the tech world. Why do you ask? Because Neural Networks are developed using the human brain as inspiration.
Source: Photo by Robina Weermeijer on Unsplash
Neural networks are really a series of algorithms that work together just like a human brain to understand the subtle relationship between different kinds of data in huge datasets. These networks, structured after the neurons of the human brain can adapt to change without any external updates on output criteria. A ‘Neuron’ in this case is a mathematical function that collects and classifies information based on the architecture.
The History
The idea of the neural network was first proposed by Warren McCullough and Walter Pitts, both researchers at the University of Chicago, in 1944. At that point, Neural nets were used in both neuroscience research and computer science. The popularity of the neural networks, however, has waxed and waned throughout the decades, but it has ultimately made a comeback thanks to the increased processing power of the graphics chips.
The Workings Of Neural Network
Broadly used to train AIs, Neural networks help computer programs to perform tasks without additional help. In training sessions like these, the input examples are hand-labeled from the beginning.
Source: Photo by from Pexels
Let’s think about an object recognition system being trained by a neural network for a moment. To train the system, the AI scientists will feed the system with hand-labeled images of various objects until the system can recognize the common visual patterns in the same type of objects, such as doors and windows on a house, handles on coffee cups, wheels on cars and so on. Modeled after the human brain, a neural network comes with millions of processing nodes that are interconnected, like synapses.
Modern-Day neural networks are organized into layers or nodes that act in a feed-forward mode, which means that the data only moves in one direction through them. The individual nodes can be connected to several nodes from the layers above and below it, sending and receiving data.
As soon as the node receives a number of inputs, it will multiply it with the weight (a numerical value) it assigns with each data input, then adding the resulting product until it’s a single number. If the single number is greater than the threshold number, the node sends the sum of the weighted inputs to its outgoing connections.
During training the weights and the threshold number of the neural nets are set to random, feeding the training data to the bottom layer, a.k.a the input layer and the data moves through the layers, getting multiplied and added. Eventually, it arrives at the output layer, transformed by computing throughout the system. It’s notable that since the weights and threshold numbers are set to random, they keep adjusting themselves till all training data with similar labels are yielding the same results.
The working of a neural network is surely complicated, but it’s benefits are many. While I know that the uses and effects of AI and Neural Networks don’t need to be mentioned separately, I still think it is necessary to take a brief look at them.
AI And Neural Networks: The Uses
AI and Neural networks are being used in a versatile way throughout the industries. Whether we are talking about the financial industry or the medical industry, the use-cases of neural networks are so many that it will be hard enough to list them all. However, here we are going to talk about a few of the most important uses that affect our everyday life.
Neural Network In Healthcare
The emergence of newer technologies has shown us different ways of approaching neural networks. While it seems way futuristic for us to use neural networks in the healthcare industry, the fact still remains that the use of neural networks is decades old in healthcare.
Source: Photo by Richard Catabay on Unsplash
The uses of the neural network are many. But few of the most important ones include diagnostics, medical image processing, biochemical material analysis, and drug development. The benefits of using neural networks in such diverse ways are clearly evident in the advancements of medical treatment all over the world.
Neural Network In Banking And Finance
Neural networks come with a plethora of benefits for the banking and financial industry as well. Eliminating the need for complete information to train the AI model, with Artificial Neural Networks, the finance industry is better equipped to handle the various uncertainties of the industry. While there are many ways ANN is being used in finance, here are the very important ones-
Source: Photo by Markus Spiske on Unsplash
- Loan Application Evaluation
There are various factors the banks have to consider before they approve of a loan. And this is where the application of neural networks have proven to be most useful. Banks want to maximize the loan return rate, and as a result, they use neural networks to sort through the loan applications and the applicator’s details to find the most eligible ones. The entire process includes analysis of the past failures of the applicators.
- Stock Market Evaluation
The stockbrokers are now training Neural Networks to predict stock market indexes and stock values. Using historical data the ANNs are trained based on different limitations and parameters. The variables and information used to train such neural networks are chosen wisely as these variables determine the prediction quality of the neural network. In this case, the feed-forward networks are the most used architecture due to their easy implementation and better generalization abilities.
-
Searching For Credit Card Customers
It is important for credit card businesses to acquire customers who need credit cards. Not acquiring the ideal customers would eventually lead to a low breakeven percentage between per card revenue and per card cost. With neural networks, the credit card businesses are able to choose the customers who are in need of credit cards and will use it to make purchases. The neural networks are trained to provide more meaningful questions for the application for identifying the ideal customers.
Neural networks have many other applications of neural networks in the banking and finance industry, such as bond rating prediction, business failure rating prediction, currency prediction, bank failure, and many others.
AI & Neural Network In Manufacturing
It is true that the application of AI in the field of manufacturing has brought immense benefits, though in most cases the application of AI in manufacturing is concerned with expert systems using perfectly labeled data to automate tasks. However, the manufacturing process today is changing with great speed based on the technological advances as well as the changes in customer’s demands and reduced life cycle of products. Only with a neural network will the manufacturing industry be able to catch up with these strings of changes.
Source: Photo by The New York Public Library on Unsplash
The self-adapting character of ANN works perfectly with the environment of changes that are going on in the manufacturing industry. This technology is also a widely used solution for process monitoring and controlling applications. With the use of neural networks, industry professionals are making sure that the manufacturing process is of high quality.
The Neural Network Effect On Our Life
As one of the most sensational technologies of this era, AI and neural networks have been in the headlines for a long time. But as these technologies become more and more mainstream, blending into our everyday lives, it starts to impact our lives more. Now the question remains if these effects are good or bad!
It is without a doubt that the application of neural networks has made our lives a lot easier than it used to be ten or twenty years ago, but at what cost? Let’s take a look at every way AI is affecting our lives to understand the deeper implications of AI technology.
-
Adapting To Changes
It goes without saying that as time changes, so does industries. Based on the public demands, industries have to change their way of working, how they manufacture products and how they market it. With AI and neural networks, adapting to these changes has become incredibly easy.
Source: Photo by Craig Sybert on Unsplash
In order to change their work process to fit with modern times, every industry has to take note of many variables in the market. Whether it is the spending power of the customer or the recent inflation, these variables are the key to optimizing the industry’s inner workings. Understanding these variables, however, is not an easy task considering that they are usually hidden in the huge amount of datasets collected from various sources. And that’s where the application of AI and neural networks comes in.
Analyzing the datasets and understanding the subtle relationships between these variables is what these Ai models are trained to do. With them, it is easier for industry professionals to understand how they can optimize the industry workings in order to fit the ever-changing and growing demands of the customers and make revenues. The fast adoption of AI and neural networks so far has ensured that the different industries have better ways to deal with changes and enhance their workings accordingly.
-
Loss Of Human Touch?
When it comes to the matter of AI application in industries, many ask the simple question: will AI take away my Job? And it is indeed a valid question. The way AI is reducing the number of menial tasks through automation is somewhat alarming, especially when you consider how many of those tasks were previously performed by human beings. But the question that should be asked is whether the quality of the work is being compromised or not.
Source: Photo by Franck V. on Unsplash
And AI is not human, no matter how much it’s decision-making skills resemble a human being. And no matter how trained the system is in analyzing the datasets and making perfect predictions, it is still not going to be able to provide the human touch. Human intuition and experience is something you can not teach a machine. And there are many industries where human intuition is absolutely needed for the decision making process.
Implementation of AI and neural networks into an industry is indeed a complicated process if you are to consider every aspect of its effect on that industry. While on one hand, it will enhance the inner workings of the industry, it will lack human experience and intuition. So considering all, we have to make the final call, if AI is actually worth it or not.
-
Automating Industries
Industry automation is one of the many beneficial aspects of AI and neural networks. Using historic datasets, neural networks are used to train AI models to do works that were until then done manually by human workers. Automation does save a lot of time and improves the quality of work in many cases. But there’s one problem with this instant automation.
Source: Photo by Franck V. on Unsplash
The implementation of AI in various industries is in fact costing a lot of people their jobs, such as customer service in eCommerce and data entry. The automation is not only taking away jobs, but it is making a lot of white-collar jobs a lot harder.
While the above concern is a legitimate one, there is a silver lining to this as well. Ai might be wiping away a lot of positions throughout industries, but it is creating a lot of other opportunities, for people, and it is only a matter of time before industry professionals start taking the advantage of that.
AI And Neural Networks: Worth The Hype Or Not?
According to Geoffrey Hinton, Yann LeCun, Yoshua Bengio, the three maestros of AI and deep learning, the advancing hardware technology is going to change the modern Neural Network. And if that stands to be true, then we will be seeing more impactful uses of neural networks and artificial intelligence. As the uses of AI and ANN change, so will the way it affects our lives. This is why in the future we need to put more thought before implementing the technology, considering all aspects of its effect on society and whether it will be a better solution in the future as well.