Artificial Intelligence, Machine Learning and Deep Learning: A comprehensive look

1. What is Artificial Intelligence, do we really need it?

Artificial Intelligence (AI) is the ability of a computer or machine to perform tasks that usually require human intelligence and interference. This includes things like learning from experiences, understanding language, solving problems, and making decisions. In simple terms, AI is when machines are designed to think, learn, and act like humans in certain ways, without needing constant human input.

Do we always need it? Not really, we do not need Artificial Intelligence integrated into every daily application of ours. AI systems can be resource-intensive, requiring significant computational power, which can strain the system. Additionally, training AI models is often complex, time-consuming, and expensive. For some applications, simpler solutions may be more efficient and effective, as not every task requires the sophistication of AI.

"AI isn't a tool - it's an Agent"

Overusing AI can also lead to unnecessary dependencies, making systems more prone to errors and harder to maintain. Furthermore, integrating AI where it isn’t essential can raise ethical concerns, such as data privacy risks or bias in decision-making. It’s important to carefully evaluate whether AI genuinely adds value to an application or if traditional methods are more practical and sustainable.

The Black box Large Language AI models that we use in our day-to-day life like Chat-GPT, Gemini, Co-pilot etc. also have a significant environmental impact. A recent study by University of California revealed the significant water footprint of GPT 3 for training purposes, which was approximately 700000 litres. Additionally, detailed calculations revealed that per every query on Chat-GPT, 4.2 grams of CO2 is emitted. This is a substantial amount considering its billions of users.

AI systems are unnecessary in situations where operations can be performed empirically, and an analytical solution to the problem already exists. If a function can directly and accurately solve the problem, developing an AI system to approximate the same function using various input parameters is redundant and inefficient.

Moreover, relying on AI in such cases not only wastes resources but also introduces potential errors during training and prediction. Analytical solutions are often more reliable, interpretable, and easier to implement. Therefore, it's more practical to prioritize straightforward mathematical or logical methods when they are sufficient to solve a problem.

Occam's Razor is the principle that states: "The simplest explanation, with the fewest assumptions, is usually the best."

For example, for basic mathematical functions like Addition, subtraction, Integration etc. we do not need to build an AI model, a simple calculator can do the work. Point being, according to Occam's Razor, there's no need to complicate the solution (by developing and training an AI model) when a simpler, more direct approach works efficiently and effectively. Over-engineering in such scenarios violates the principle of simplicity and can lead to wasted resources without providing additional value.

2. So, where should we exactly use Artificial Intelligence?

AI models should be used strategically in areas where they provide clear and measurable benefits, addressing problems that traditional methods cannot solve efficiently or consume a significant time and effort. Their strength lies in handling complex tasks like pattern recognition, data analysis, and decision-making at scale, making them invaluable in sectors such as healthcare, finance, education, etc. For instance, in healthcare, AI can be used for early disease detection, personalized treatment recommendations, and drug discovery. In finance, AI excels at fraud detection, credit risk assessment, and algorithmic trading. Similarly, AI-powered tools in education provide personalized learning experiences, automated grading, and adaptive content delivery, while in transportation, AI optimizes traffic flow, enables autonomous vehicles, and improves logistics.

To use AI effectively, organizations and individuals must first assess whether it is genuinely necessary for a specific application. Ethical considerations should also play a critical role in determining the use of AI. Developers and organizations must ensure the responsible handling of data, addressing concerns like privacy, bias, and fairness.

AI works best when it complements human expertise rather than replaces it. Integrating AI systems should aim to augment human capabilities, enabling better decision-making and improving overall efficiency. The focus should be on applications where AI delivers meaningful value - simplifying processes, enhancing accuracy, and solving complex challenges. By carefully evaluating the need, ensuring ethical implementation, and balancing automation with human oversight, AI can be a powerful tool that drives innovation while remaining practical and sustainable.

3. What are these concepts: Machine Learning and Deep Learning?

3.1 Machine Learning

The name itself explains the concept, making the ‘Machine’ ‘Learn’. Machine Learning is the way of teaching machines to learn using data to make decision or predictions, without explicitly programming it for every individual task. Machine Learning is a subset of Artificial Intelligence, where AI refers to broader concept about development of systems that stimulate human intelligence. Machine Learning is way to develop such tools which stimulate human intelligence.

Data is analysed for learning patterns, semantics and representations and these help in formulating predictions. It is an effective method since large amounts of data can quickly be analysed, it updates to new information and this can be used to solve complex problems, which can be difficult to be solved by traditional programming.

For example, a spam filter in your email uses machine learning to recognize and block spam messages by analysing patterns like keywords or sender behaviour. Another example is a recommendation system, like Netflix suggesting movies based on what you’ve watched and liked before. These systems learn from data and continuously improve, making our lives easier and more efficient.

3.2 So how does Machine Learning really work?

For prediction of the ground truth, different algorithms are used based on the interpretability of the problem, nature of the input data whether complex or simple patterns are to be analysed, or scalability of the system where computational efficiency is the concern. Some of these algorithms are listed below:

A. Linear Regression -

When to use it:

• If your goal is to predict a numerical value based on input features.
• When the relationship between your features and the target variable is linear or approximately linear.
• If you need a simple, interpretable model that explains how different features contribute to predictions.

Example:

Predicting house prices based on size, location, and number of bedrooms.

B. K-Nearest Neighbours (KNN)

When to use it:

• When you need a simple, intuitive algorithm for classification or regression.
• When your data is small and not too high-dimensional (because KNN can be computationally expensive for large datasets).

Example:

• Classifying whether a plant is a type of flower or weed based on features like petal size and colour.

C. Logistic Regression

When to use it:

• For binary classification problems (e.g., yes/no).
• When the relationship between the features and the target variable can be modelled using a logistic (S-shaped) curve.

Example:

• Predicting whether an email is spam or not.

4. Deep Learning

Deep Learning is a specialised type of Machine learning that uses structures called as neutral networks, which is inspired by how human brain works, to learn and make decisions. Deep learning is inspired by Cognitive Neuroscience (a.k.a Neuropsychology) as it explains complex cognitive processes of the human brain. The neural networks consist of different layers characterized by weights and interconnected nodes (just like neurons) which process information in a hierarchical way i.e. starting with learning of simple absolute patterns and gradually diving deep into complex patterns.

Deep learning is an effective and efficient way for working with large amounts of data, such as images and videos, or even text, because it can automatically identify important features in an effective manner relative to other machine learning algorithms.

For example, deep learning powers facial recognition systems by analysing key facial features, and it’s used in self-driving cars to detect pedestrians, traffic signs, and lanes. It’s also behind virtual assistants like Alexa or Google Assistant, enabling them to understand and respond to voice commands accurately.

4.1 How does Deep Learning really work?

Deep learning works by using neural networks with multiple layers to process data and automatically learn patterns from it. Each layer extracts features, starting with simple ones (e.g., edges) and progressing to more complex representations (e.g., faces or objects). You should use deep learning when you have large amounts of labelled data, complex tasks like image recognition or natural language understanding, and access to significant computational resources. It is ideal for scenarios where traditional machine learning fails to capture intricate patterns in data. Deep learning is most effective when feature extraction and decision-making need to happen automatically. Below are some of the commonly used deep learning algorithms:

A. Convolutional Neural Networks:

CNNs are designed to process and analyse visual data, such as images or videos. They work by breaking an image into small parts (like patches or filters) and scanning for patterns such as edges, shapes, or textures. These patterns are then combined to recognize the bigger picture, like identifying a cat in an image. CNNs are great for tasks like image classification, object detection, and facial recognition because they focus on important features in the data and ignore irrelevant details.

B. Recurrent Neural Networks:

RNNs are used for analysing sequential data, like time series, speech, or text, where the order of the data matters. They work by remembering information from previous steps while processing new data. For example, when predicting the next word in a sentence ("I am going to the ___"), an RNN remembers the context of earlier words to guess the next word correctly (e.g., "store" or "park"). This memory-like behaviour makes RNNs effective for language translation, speech recognition, and stock price prediction.

C. Generative Adversarial Networks:

GANs consist of two parts: a generator and a discriminator. The generator creates fake data (like realistic-looking images), and the discriminator tries to distinguish between the real and fake data. Both parts compete with each other, and over time, the generator becomes so good that the fake data looks almost real. GANs are widely used for creating realistic images, enhancing photo quality, and even generating deepfake videos.

5. Conclusion

Artificial Intelligence, Machine Learning, and Deep Learning are transformative technologies that have reshaped how we approach complex problems and interact with the world. While AI enables machines to replicate human-like intelligence, Machine Learning and Deep Learning provide the means to process and analyse data in innovative ways, driving advancements across diverse fields like healthcare, finance, education, and transportation. However, AI is not a universal solution. Its integration must be thoughtful and strategic, ensuring it adds real value while addressing challenges such as resource consumption, ethical concerns, and environmental impact. Simpler, traditional approaches should not be overlooked when they are sufficient for solving problems efficiently.

In conclusion, the key to harnessing AI lies in balancing its immense potential with responsible and practical application. By using AI to complement human expertise, we can solve complex challenges, improve efficiency, and create impactful innovations. At the same time, prioritizing ethical considerations, sustainability, and simplicity ensures that AI remains a tool for progress without unnecessary complications or negative consequences. With the right approach, AI has the power to revolutionize industries while maintaining a strong foundation of practicality and responsibility.