A (very) basic guide to artificial intelligence

Subject: Science and tech

Section: Awareness in IT

Context:

‘Artificial intelligence’ (AI) is intelligence in a machine. There is currently no one definition of AI. A simple place to begin is with AI’s materiality, as a machine-software combination.

What does the machine do?

Linear separability is a fundamental concept in AI, illustrating how machines can distinguish between different categories of data. This concept is exemplified through the task of separating cat and dog images, a process that, while not typically assigned to linear classifiers, serves to clarify the principle involved.
Through this process, the machine attempts to find a pattern or a line (in the case of a 2D graph) or a plane (in a 3D space) that effectively separates the images into two groups, one predominantly consisting of cat images and the other of dog images. Once a satisfactory separation is achieved, demonstrating that the machine has successfully classified the images based on the given features, the experiment concludes.
This scenario underscores the machine’s ability to distinguish between categories using linear separability, a core concept in AI and machine learning, which involves finding a linear boundary that divides data into distinct groups based on their features.

How hard is decision-making?

Decision-making in AI ranges from simple to extremely complex, based on the dataset’s nature and the decision’s context.
Simple Decision-Making: In straightforward cases, such as separating a set of marbles based on a single characteristic, decision-making can be very reliable with just one parameter.
Intermediate Complexity: For tasks like distinguishing between cats and dogs, AI may require a dozen parameters. These parameters could include various physical attributes, and AI tools might plot these features on graphs to classify the subjects effectively.
High Complexity: Decision-making becomes significantly more complex in scenarios like a driverless car deciding when to brake for a bird crossing its path. Here, hundreds of parameters might be needed, including the context of the situation (e.g., the urgency of reaching a destination).
Mind-Boggling Complexity: Large Language Models (LLMs) like ChatGPT represent an even more advanced level of decision-making.ChatGPT doesn’t classify information but generates text by predicting the next word in a sequence based on its training from a massive corpus of text. This process involves understanding the creation process behind the text, reflecting the real world. ChatGPT operates with over 100 billion parameters, showcasing its high level of complexity in decision-making and text generation.

Types of Machine Learning:

Supervised Learning: Involves learning from labeled data, where the structure and categories of the data are defined.
Unsupervised Learning: The machine learns to organize and interpret data without predefined labels or categories, identifying patterns and structures on its own.
Reinforcement Learning: The machine learns through trial and error, adjusting its actions based on feedback to maximize a reward signal.

Artificial Neural Networks (ANNs):

ANNs are computational models inspired by the human brain’s network of neurons. They consist of nodes (or neurons) connected in a way that allows the network to learn and make decisions.
Components:
- Activation Functions: Algorithms that determine a node’s output based on its input signals.
- Weights: Values that represent the importance of each input to the node, influencing the output.

Transformers:

A type of ANN that allows for parallel training, making it efficient for processing large datasets.
Components:
- Encoder: Breaks down input data (e.g., an image) into smaller pieces and encodes it as numerical data.
- Decoder: Processes the encoded data to reconstruct or interpret the input data’s content.
Significance: Introduced in 2017 by Google, transformers have significantly improved the performance of ANNs in tasks like language translation by focusing attention on different parts of the input data.

GPUs and AI Development:

A Graphics Processing Unit initially designed for rendering graphics in video games, is now widely used in running ANNs due to its ability to perform parallel computing tasks efficiently.
Nvidia’s Role: Nvidia has become a leading provider of GPUs for AI and machine learning, experiencing rapid growth in valuation due to the demand for AI technologies.
Market Challenges:
- Competition from companies developing non-GPU hardware.
- Researchers creating smaller, less resource-intensive learning models.
- Development of new software to reduce dependency on specific hardware, like Nvidia’s GPUs.

Source: TH