Artificial Intelligence (AI) Literacy

AI allows computers to simulate human-like cognitive processes, such as learning, and problem solving. AI works using algorithms, which are sets of instructions that resolve calculations.  These are 3 of the main types of AI algorithms.

• Supervised Learning involves working with pre-labelled data.  The model learns to match inputs with (known) outputs.  The aim of supervised learning is that the model will learn to predict the required output based on newly inputted data.
• Unsupervised Learning involves working with unlabelled data.  The aim of unsupervised learning is that the model will be able to recognise patterns in data that it has not previously been trained on.     
• Reinforcement Learning does not use predefined data.  The aim of reinforcement learning is to work within a specific environment and learn to do a task based on feedback.  The feedback is in the form of a penalty or a reward.

For more information on supervised, unsupervised and reinforcement learning, please review this short video (6mins)

Machine learning is a field of study into how computers can learn patterns from data without being explicitly programmed.  This learning, or training from data, allows software to become better at prediction and classification.  The more data the model is trained on, the more effective the model becomes at predicting new outcomes.

For more information on how machine learning works, please watch this short video (8 mins).

This video gives examples of how machine learning can be used.

Deep Learning uses an advanced from of machine learning which is based on the architecture of the human brain.  These layers of interconnected networks are called neural networks.  There can be hundreds or thousands of layers in a deep learning model, which allows the computer to identify complex patterns in the training data.  These deep learning models are the basis for Large Language Models.

This short video (6 mins) provides an overview of deep learning.

Example of the application of deep learning. (12 mins)

Large Language Models use machine learning techniques and neural networks to analyse the statistical relationship between letters, words and phrases.  Based on the model’s knowledge of those statistical relationships it develops the ability to predict the next most likely word in a sequence of words.  LLMs are trained on vast quantities of data from the Internet, including Wikipedia, github and Reddit.  This takes time, computer processing power and a large financial investment. 

For more information on how large language models are trained, have a look at this video from The Turing Lectures  (46mins). This video is aimed at the non-expert and explains the complexities of LLMs in plain accessible language.

For a general overview of the technology involved in LLMs, please see this article by Timothy B Lee & Sean Trott. 

Natural Language Processing (NLP) algorithms are used to analyse and interpret human language. Tokenisation is one of several analytical techniques associated with NLP that facilitates Generative AI. 

Watch this video for a simple overview of NLP.

Watch this video for more technical overview and examples of where NLP is used.

Graphic Processing Unit (GPU) 

For deep learning models to work effectively they need to consume large amounts of data and require correspondingly large computer processing capabilities.   Until relatively recently, lack of suitably large, fast computing capabilities limited the development of AI algorithms.  The development of the GPU changed this.  GPUs were invented to make graphics in video games faster but were co-opted to accelerate AI algorithms. This increased computing capability, together with the exponential availability of (training) data on the internet, facilitated the development of LLMs like ChatGPT3.5 and Meta’s LLAMA.

Gen AI brings together aspects of artificial intelligence including LLMs and NLP.  Gen AI allows users to ask a question in  natural language and in turn, it generate a response.  In other words, the AI performs complex statistical calculations at high speed, producing paragraphs of text based on the training data and the information supplied by the user in the question or “prompt”.  This is done in real time. This “prompt” can also be an image, video or audio.

For more information on how Generative Artificial Intelligence works, have a look at this video from The Turing Lectures  (80 mins). This video is aimed at the non-expert and explains the complexities of AI in plain accessible language.