Recently I was able to join a paja regarding AI and there I learned how little we actually know about AI and machine learning. We focus mostly on how to use AI to benefit us but not on how this technology is built or how it operates and so on. When you think about AI you may think there is an algorithm that drives data and learns from there and the result is this black box of AI. No surprise that some of us think AI is a scary thing since most people can’t explain how it works or how it is built. We just assume that the process is hard to understand, and it is easier to learn how to use it beneficially.  

Most likely this essay will not give the full understanding of how AI is built but it at least gives a better understanding of how machine learning works and what is behind AI. 

How does it learn?  

Firstly, is good to understand what machine learning is and what we mean with it. 

Machine learning is one branch of AI and computer science. In one way you could describe that it tries to teach machines to learn in human ways. A more detailed way of putting it is that it focuses on the use of data and algorithms to imitate humans’ way of learning, by gradually improving its accuracy. This has its limits for now and mostly still requires human help in the process. Because the process tries to mimic human behavior, there are different ways and layers of learning that machines can do, and based on it how they operate. But to deeply understand the process there is a need to deep dive into the various sub-domains of AI. Here, we shall not dive so deep but rather list machine learning terms and understand better their part in the AI. Also, some of the terms are sub-fields of each other, where the learning and technology are taken deeper. 

Machine Learning (ML) as a part of AI.  

ML teaches a machine how to make decisions and conclusions based on past experience etc. Data it has been given to learn from. It identifies patterns and analyses data to identify data points and the meaning of them to reach a possible conclusion without the need for having to involve human experience. UC Berkeley has broken down the algorithm of the learning into three main parts.  

1. Decision Process: The machine produces an estimate about a pattern based on some data that can be labeled or unlabeled. Since ML algorithms are used to make a prediction or classification.
2. Error Function: This part evaluates the prediction of the model the machine has made in part one by making a comparison to assess the accuracy of the model, but only if there are known examples to compare to.
3. Model Optimization Process: (the following sentences have been straightly copied from the source) “A Model Optimization Process: If the model can fit better to the data points in the training set, then weights are adjusted to reduce the discrepancy between the known example and the model estimate. The algorithm will repeat this “evaluate and optimize” process, updating weights autonomously until a threshold of accuracy has been met. “

ML is the basic term of the science of machine learning but also the process of the algorithm that makes machines analyze patterns and data points to create assumptions as a conclusion of the data it has to process.  

Deep Learning 

Deep Learning is a sub-field of AI but more specifically Deep learning is a sub-field of neural networks. Deep learning is in one way a technique used in ML since neural networks is a sub-field of Machine Learning. It is different from ML in its ways of learning by teaching a machine to process inputs through layers. In Deep Learning the algorithm is able to process datasets and the data doesn’t need to be labeled or unlabeled. It is so powerful that it can ingest unstructured data in its raw form like images or text and automatically label them and categorize them from one another. This means it doesn’t require human intervention so much to process large data sets and is different from classical ML since it doesn’t require human experts to determine the set of features to understand the difference between data inputs.  

Neural Networks 

Neural Networks is a sub-field of Machine learning that has deep learning inside of it. Neural Networking mimics the Human Neural cells. Neural networks are built in that they are comprised of node layers, containing an input layer, at least one or more hidden layers, and an output layer. These nodes are connected to each other and have an associated weight and threshold. For data to be passed along to the next layer of the network, the output of the individual node needs above the specified threshold value to be activated, so it can send the data to the next layer. Nodes and the network between them are series of algorithms that capture the various underlying variables and process the data in the same way as a human brain does.  

How do we get to the deep learning with these nodes that form a network? The deep refers to the number of layers it has in the neural network. If a neural network consists of more than three layers, it is considered a deep learning algorithm. These are most likely used in the more advantageous AIs that generate or work with computer vision, natural language processing, and speech recognition. These AIs could be live translators, that translate human speech into another language in the live time.  

Other terms used in Machine Learning and AI 

Natural Language Processing shortly NLP. This is the science of how a machine understands, reads, and processes the language that is given to it. Also, after it has understood, what the user is trying to communicate with it, it is able to generate a respond accordingly. But NLP focuses on this part of machine learning. 

Computer Vision. Computer vision is about algorithms that analyze pictures and try to understand them by breaking the image down and studying different parts of the objects on it. Focusing on understanding picture objectives, helps the machine to make better output decisions based on previous observations. 

Cognitive Computing. This science focuses on trying to teach the machine to mimic the human brain. This is made by analyzing text/speech/images/objects in a manner that a human does and gives the desired output. This way of learning either requires large data regarding human behavior or humans to help the machine to understand the desired output from it. 

Conclusion 

For AI to exist it needs machine learning and for more efficient and complex AIs to exist they need to create neural networks that allows them deep learning. In machine learning the machine learns from the algorithms it has been given and optimizes the result based on the data it has or the outputs human requires from it. Machine learning and AI have complex computer science fields and sub-fields inside them supporting and making machines better.  

Hopefully, this essay helps the reader understand that AI is not a black box but rather a taught machine that bases its learning and actions on its given data and tries to improve its outputs over time. 

References: 

What is Artificial Intelligence in 2023? Types, Trends, and Future of it? By GReat Learning Team. Published in 17.08.2023 in Great Learning Website. Link to the text: What is Artificial Intelligence ( AI) in 2023?- Great Learning (mygreatlearning.com) 

What is Machine Learning? By IBM. Published in IBM website. Link to the text: What is Machine Learning? | IBM 

Tekoäly: matkaopas johtajalle. By Antti Merilehto. Published in 26.03.2020 by Alma Talent.