AI Explained: The Most Common Questions and Answers for Newbies

As a self-taught AI learner, I have been studying AI for a while and I would like to share my experience of learning AI to help others who are interested in learning AI.

Here I have summarized some common questions to help answer those who are interested in AI and want to learn AI. Also, if I think of more questions in the future, I will update this article.

Q1: Is mathematics necessary for learning AI? Can I just skip mathematics and dive right in to learn AI?

A1: It's not recommended that you skip mathematics when learning AI, otherwise you will have no idea what you’re doing. That is not a good way to learn AI. In fact, mathematics is the core of AI, and AI is actually a complex mathematical model. Almost all the AI courses will require you to learn mathematics first, especially vector calculus, linear algebra, and advanced statistics. As I said earlier, AI = Coding + Math, and they cannot be separated. The best way to learn AI is to know how it works.

Q2: I have heard about machine learning and deep learning in AI. What's the difference between them?

A2: They are both fields of AI, but deep learning is a subfield of machine learning. However, not all machine learning models use neural networks. Compared to deep learning models, machine learning models require more human intervention to design features and tune parameters. Deep learning models use neural networks to learn and make decisions, so deep learning needs less human intervention, and they have better capabilities to do self-learning.

Q3: Is deep learning better than machine learning?

A3: Deep learning is not always better than machine learning. They have their own advantages and disadvantages, and they are applied in different situations. For instance, if your data set is small and you only want to do some simple tasks, then machine learning is more suitable for you, since machine learning can train on small data sets. If your data set is large and you want to do some complex tasks, then deep learning is more effective for you.

Q4: How do I start to learn AI?

A4: Learn mathematics (mainly vector calculus, linear algebra, and advanced statistics) first, then learn coding (mainly Python). Having a good foundation in mathematics is helpful for learning AI. As you delve deeper into deep learning, you will realise that the math becomes more difficult and abstract.

Q5: Should I have a good laptop for training AI?

A5: A good laptop for learning AI is not a strict requirement, but it can make your learning experience more smooth and enjoyable. AI involves a lot of complex computations and data processing, which can be demanding for your laptop's hardware. If you have a laptop with a powerful processor, a large amount of RAM, and a dedicated graphics card, you will be able to train AI faster and more efficiently. I remember I spent RM4000 to buy a gaming laptop in 2022 because I wanted to learn AI. Now my dream has come true. I can say it's worth buying a good laptop if you're really interested in learning AI.

Q6: If I don't have a good laptop for training AI, how do I solve this problem?

A6: You can use online platforms and services that provide cloud computing and GPU resources for AI. For instance, you can use Google Colab, a free service that lets you run AI code in your browser using Google's servers and GPUs. You can also use Kaggle, a platform that offers free access to datasets, notebooks, and competitions for AI. However, it's better to have your own good laptop for learning AI.

Q7: What's TensorFlow, Keras and PyTorch in AI?

A7: TensorFlow is an end-to-end framework that was developed by Google to train an AI. Keras is a high-level API that was integrated into TensorFlow in 2017. It simplifies the process of building and testing deep learning models by providing common layers, activation functions, optimizers, and metrics. PyTorch is a framework that was developed by Facebook. It is based on Torch, a scientific computing library for Lua.

Q8: Why we need them to train AI?

A8: They provide many advantages and features that make the process of training AI easier and faster. They already have a lot of built-in modules and optimized algorithms to train AI. You can also write your own code using NumPy, SciPy, Pandas and scikit-learn to train AI, but it would be more difficult and time-consuming, since you would have to implement many of the low-level details and computations.

Q9: TensorFlow by Google or PyTorch by Facebook? Which one is better?

A9: There is no answer to this question, since they are suited for different use cases and preferences. If you're working on a large AI project that requires high performance, TensorFlow is preferred. If you're working on an AI research, PyTorch is preferred. However, PyTorch is more compatible than TensorFlow, and PyTorch is easier for beginners, since PyTorch has a more Pythonic and intuitive syntax, which makes it easier for beginners to learn and debug. Also, PyTorch has better integration and support for popular Python libraries and tools, such as NumPy, SciPy and scikit-learn.

Q10: Can AI only be trained by CPU? Can GPU train AI?

A10: AI can be trained by both CPU and GPU, but GPU is faster and more efficient than CPU, since GPU has stronger computing power than CPU. GPU can perform parallel processing, which means it can handle multiple tasks at the same time, while CPU can only perform sequential processing, which means it can only handle one task at a time.

Q11: Should I use GPU rather than CPU to train AI?

A11: It depends on your situation. If you train your AI model on a small data set, then just using CPU is enough. If you train your AI model on a large data set, then you should use GPU rather than CPU. However, it's complicated to install the GPU version of TensorFlow, since you also need to install some NVIDIA graphic card softwares for training AI, and also you need a strong NVIDIA graphic card to train your large AI model, such as the GeForce RTX30 and RTX40 series. Here I'm using the CPU version of TensorFlow, since I didn’t train a large AI model.

Q12: Which graphic card is better for training AI? Nvidia GeForce RTX series or AMD Radeon RX series?

A12: I would recommend Nvidia GeForce RTX series for training AI. Based on some recent benchmarks and reviews, it seems that Nvidia has an edge over AMD in terms of performance and innovation for AI and machine learning tasks.

Q13: What are the specs of your laptop for learning AI? How about its price?

A13: My laptop’s graphic card is NVIDIA GeForce RTX3050 4GB GDDR6, CPU is 11th-gen i5-11400h @ 2.70GHz with 6 cores and 12 threads, and memory is 16GB DDR4. Its price in 2021 was approximately RM4000. I think now its price has decreased to RM3000 because of newer laptops released.

Q14: What platform do you use to train your AI model?

A14: I train my AI model in Jupyter, which is an interactive computing platform that supports Python. It allows you to run your code in separate cells, similar to PowerPoint slides.

Q15: Does the size and structure of an AI model determine its strength and performance?

A15: Nope, sometimes a smaller and simpler AI model can perform even better than a larger and more complex one. One such example is TinyLlama, a mini natural language processing (NLP) model. It is based on a transformer network, and it is much smaller and faster than other popular NLP models, such as GPT-4 and Bard.