How Intelligent is Artificial Intelligence?

Deep Blue vs Garry Kasparov (1997)

Humans have been beaten by their own creation for the first time after Garry Kasparov’s defeat against IBM’s Deep Blue AI in a chess match in 1997. That’s when computers proved to be worthy of being called intelligent and marked a significant milestone in the development of artificial intelligence (AI).

Photo by Growtika on Unsplash

In recent years, Artificial intelligence (AI) has shown remarkable growth and advancements in various fields, and you’ve probably witnessed how the industry blew up in the last two years. Unless you’ve been living under a rock!

It all started when OpenAI made the famous ChatGPT available to the public which uses AI to generate human-like text. Following that, we started to see how AI was being used in other ways, such as DALL·E 2 which generates pictures, or Microsoft Bing AI which helps you with research, and so much more. This list is not limited and can go on forever!

Photo by Andrew Neel on Unsplash

The level of implementation and integration of AI keeps increasing every day, and at this point, we use AI in various ways in our daily lives. “Ranging from chess-playing machines to self-driving cars to chatbots and so on” (Mitchell, 2024).

This advancement has made lots of people interested in learning about how AI works, and it raises the question:

How intelligent is artificial intelligence? How do we measure its intelligence, and what are the implications for the future of technology?

So let’s have a deep dive into AI and unravel how it works layer by layer.

Main Sections of Artificial Intelligence (ML & DL)

Artificial intelligence, or AI, is technology that enables computers and machines to simulate human intelligence and problem-solving capabilities.

IBM

That is the simple definition of AI, and by combining it with other technologies (e.g. robotics) we can achieve and accomplish tasks that can only be performed by humans.

However, the main trait that distinguishes AI is that the model needs training and learning to be able to perform. The main subsections of AI and the algorithms that are used to train the model are Machine Learning (ML) and Deep Learning (DL).

DL subsection of ML, & ML subsection of AI (Credit: Levity)

Wait, so, what is the difference? …

In short, machine learning requires the usage of a relatively smaller number of data points compared to deep learning, which makes machine learning easier and less time-consuming, but best for identifying patterns. On the other hand, deep learning requires analyzing millions of data points through layers of neural networks which allows the computer to build extensive knowledge over time (just like our brain).

ML vs DL (Credit: Zendesk)

In other words, machine learning is like cooking based on the same recipe every time, but deep learning is like making the recipe better and better each time you cook.

But then, how does each one work?

Machine Learning

By definition, Machine Learning is a branch of AI trained on statistical models and algorithms, which enable it to make predictions and decisions. Using training and historical data, machine learning algorithms can improve and adapt over time, enriching their capabilities (Grieve, 2018).

By giving the AI enough training data to learn from, the computer will be able to find relationships and patterns, and be able to make predictions based on the input data. Here’s how it works:

Process of Machine Learning (Credit: Zendesk)

After the AI is developed and trained by the engineer, the algorithm will follow a standard process:

1. Receive information and data from the user
2. Analyze and process the data
3. Find patterns and relationships
4. Based on the patterns, make predictions
5. Output or send the result to the user

The process in which ML follows is common across different machine learning models, but the way that the AI gets trained is different in each model.

Machine Learning Models (Credit: Research Gate)

• Supervised Learning: A series of labeled data is fed to the algorithm, and it learns a model to for responding and identifying the data. For example, a programmer is trying to teach the computer how to distinguish between dogs and cats. The programmer will feed labeled images of trees and apples, and over time, the AI will recognize patterns to differentiate the two. For instance, the algorithm may recognize that apples can be round objects in colors green or red, and trees have wooden bodies with leaves on their branches. Once the model is provided with sufficient information, the programmer starts feeding it unlabeled data so it can accurately distinguish between the two.
• Unsupervised Learning: The algorithm will receive a series of unlabeled data, and without any human interventions, it will analyze and identify patterns in those data. The data will go through a process of clustering and segmentation, and the AI will be able to group the specific data and connect the data that are related together. For example, a series of unlabeled pictures of cats and dogs are fed to the algorithm, and the computer will learn to distinguish and group images that would more likely be dogs with dogs, and cats with cats.
• Reinforcement Learning: The algorithm will learn and be trained through a trial-and-error approach with feedback. The model gets fed training data, and to fine-tune the AI, the programmer will provide ‘rewards’ for every correct answer, and ‘punishes’ for the wrong answers. This model is mostly used for more complex tasks that come with large series of data. For example, reinforcement learning is used in autonomous vehicles to teach models how to stay in the lane, stop behind the red light, etc. After numerous trials and errors, the AI learns how to make decisions that will keep the car in the lane and the passengers safe.

Deep Learning

Deep Learning is a subfield of machine learning that structures algorithms in layers to create an ‘artificial neural network’ that can autonomously learn and make intelligent decisions. Deep learning models can analyze data continuously. They draw conclusions similar to humans — by taking in information, consulting data reserves full of information, and determining an answer (Grieve, 2018).

Process of Deep Learning (Credit: Zendesk)

The way deep learning works is inspired by the way the human brain functions, in addition to the integration of a network of neurons that transmit information to each other. Deep learning uses a layered structure of algorithms called artificial neural networks. The data is sent through these layers of the neural network, and by using mathematical operations and fine-tuning the neurons, the model will identify patterns and draw a final output. That’s why deep learning models tend to be more complex and advanced than standard machine learning models.

Well, what’s a neural network, and what’s so special about it?

Neural networks, also called artificial neural networks (ANNs) function by flowing the received data through nodes to succussive layers of neurons (similar to the human brain) where the algorithm learns and outputs a final answer. This depth of analysis and complexity of neural networks is what makes deep learning more advanced than standard machine learning.

Neural Network Structure (Credit: ZDNet)

• Convolutional Neural Networks (CNN): Convolutional neural networks use a series of layers, each of which detects different features of an input image. Depending on the complexity of the intended purpose, a CNN can contain dozens, hundreds, or even thousands of layers, each building on the outputs of previous layers to recognize detailed patterns (Awati, 2022). It is inspired by how the visual cortex of the human brain functions, and it is used in computers for purposes of computer vision and object detection.
• Recurrent Neural Networks (RNN): Recurrent neural networks use built-in feedback loops to remember past data points. RNNs can use this memory of past events to inform their understanding of current events or even predict the future (Grieve, 2018). This feature of remembering makes RNNs best used in Map Apps where they can predict traffic based on the time and the historical data of traffic in similar times.
• Multilayer Perceptron (MLP): Multiplayer perceptron is classified as a feedforward neural network, which means that the user input data only flows in one direction without the usage of feedback loops (used in RNN). This makes MLPs better at processing unpredictable data and patterns than other algorithms. MLPs are usually used in classifying images, and speech recognition.

And that’s essentially the backbones of Artificial Intelligence (AI) and how it works in its cores. Deep Learning and Machine Learning can get much more complex due to the level of complexity of the analysis and the size of data needed to train these algorithms. But with the rapid development of AI, there are always new aspects being introduced to this field and it is projected that AI will most likely be implemented in most jobs and disciplines in the future.

So, is Artificial Intelligence actually intelligent?

I will leave that for you to decide and let me know in the comments what you think of this technology and its remarkable growth in this age…

It’s comparatively easy to make computers exhibit adult-level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a 1-year-old when it comes to perception and mobility.

Hans Moravec (1988)

Too long, didn’t read! (Key Takeaways)

AI vs ML vs DL (Credit: Zendesk)

• 🤖 - Artificial Intelligence (AI): An emerging technology that enables computers and machines to simulate human intelligence and problem-solving capabilities. Current examples: ChatGPT, Copy.ai
• 🖥️ - Machine Learning (ML): A branch of AI trained on statistical models and algorithms, which enable it to make predictions and decisions. The algorithm can be trained in three methods of supervised learning, unsupervised learning, and reinforcement learning.
• Supervised Learning: A series of labeled data is fed to the algorithm, and it learns a model to for responding and identifying the data.
• Unsupervised Learning: The algorithm will receive a series of unlabeled data, and without any human interventions, it will analyze and identify patterns in those data.
• Reinforcement Learning: The algorithm will learn and be trained through a trial-and-error approach with feedback.
• 🧠 - Deep Learning (DL): A subfield of machine learning that structures algorithms in layers to create an ‘artificial neural network’ that can autonomously learn and make intelligent decisions.
• Artificial Neural Networks (ANN): Flowing the received data through nodes to succussive layers of neurons where the algorithm learns and outputs a final answer.
• Convolutional Neural Networks (CNN): Convolutional neural networks use a series of layers, each of which detects different features of an input image.
• Recurrent Neural Networks (RNN): Recurrent neural networks use built-in feedback loops to remember past data points.
• Multilayer Perceptron (MLP): Multiplayer perceptron is classified as a feedforward neural network, which means that the user input data only flows in one direction.

References

• Awati, R. (2022, September). What is convolutional neural network? — Definition from WhatIs.com. SearchEnterpriseAI. https://www.techtarget.com/searchenterpriseai/definition/convolutional-neural-network
• Grieve, P. (2018). A simple way to understand machine learning vs deep learning — Zendesk. Zendesk. https://www.zendesk.com/blog/machine-learning-and-deep-learning/
• Heath, N. (n.d.). What is deep learning? Everything you need to know. ZDNet. https://www.zdnet.com/article/what-is-deep-learning-everything-you-need-to-know/
• IBM. (2024). What is artificial intelligence (AI)? IBM; IBM. https://www.ibm.com/topics/artificial-intelligence
• Mitchell, M. (2024). The future of artificial intelligence. Nationalacademies.org. https://www.nationalacademies.org/news/2024/02/just-how-intelligent-is-artificial-intelligence
• Moravec, H. (1988). Mind Children: The Future of Robot and Human Intelligence. Harvard University Press.
• Peng, Junjie & Jury, Elizabeth & Dönnes, Pierre & Ciurtin, Coziana. (2021). Machine Learning Techniques for Personalised Medicine Approaches in Immune-Mediated Chronic Inflammatory Diseases: Applications and Challenges. Frontiers in Pharmacology. 12. 10.3389/fphar.2021.720694.
• Wolfewicz, A. (2022, April 21). Deep learning vs. machine learning — What’s the difference? Levity.ai. https://levity.ai/blog/difference-machine-learning-deep-learning

Halo! My name is Parsa and I’m a 17 y/o science and art enthusiast. I’m on the journey of learning more about Artificial Intelligence (AI) & Machine Learning (ML) and I will share updates as I progress in this journey. If you found this article interesting, feel free to follow and check out my LinkedIn where you can leave any suggestions or feedback! 🙏