Generative AI for the beginner

What is Generative AI?

Most of us have experienced AI decision making algorithms whether this be when we apply for credit cards, ask for support in chat, get map directions, face/fingerprint recognition, and much more with a yes or no type answer.  Generative AI is different in that it creates content rather than makes a decision.  So, in short generative AI is a creator.

Generative AI is capable of creating written content, pictures, video, and music.  We hear about software code from Chat GPT.  Code is basically a language and is a great example of “written” content creation.  Stanford professor Eric Brynjolsson has been quoted as saying that generative AI is the “calculator of writing” (source Bloomberg).  This is a nice way to relay how useful and everyday this technology is likely to become.

Learning from massive amounts of data to be able to answer complex questions that go beyond a yes/no, true/false, accept/decline decision, generative AI is a bit like using logical reasoning and imagination to.  Let me explain.

A Conceptual View

This is a conceptualised view rather than accurate technical detail.  A comparison with a student can help us understand what an AI tool is doing.  When a student sits an exam, this is the first time they would see a particular question.  Based on what they have learnt in lessons and through logical reasoning, they should be able to have a good attempt to answer the question well.  Our AI does something similar.  AI learns from data and uses predictive reasoning to make its attempt to answer the question.

The AI model is trained through data, these are its lessons.  The more data, the more lessons, the better the knowledge of the AI is likely to be.  However, just like the student where practice exams help, AI uses reinforcement learning, thumbs up or down, good or bad ratings, to improve its understanding and prediction.  This is easy to see with text-based content, but similar for other content types.  Learning about shapes, patterns, colour and other visual elements, the AI uses its artificial imagination to create pictures, logos, designs and similar.

Like a student, AI might not get the answer right every time!

How does AI answer the questions?

For simplicity I will focus on text based AI.  A predictive text or Large Language Model basically makes lots of connections between different words and groups of words so it is able to guess what might come next.  Add context in the form of a question and the AI starts to be able to respond in a human like way.  There is a lot more that goes on behind the scenes but superficially this is the idea.  For example, the word “I” is very likely to be follow by “am”, “have”, or “will”.  The AI learns what might be the most likely next word for the task you are doing or question you have asked.  "I am going to the shops", "I have received your email", each can be predicted by the context or question.

A situation you may have seen this is when writing an email or a sending a text message.  As you start to type the AI guesses the word, the next word, or the sentence for you.  Not as clever as something like Chat GPT but hopefully gives you a feel at an everyday level.