What is an AI?

‍
Technically, an AI is an artificial intelligence, but that’s not what ChatGPT and other tools really are. Current “AI” are massive pattern recognition and pattern recreation software tools. 

The software reviews massive datasets - words, pictures, etc. - and takes notes about the patterns it finds there. For images, some of those patterns might be: “Things that get tagged with ‘dog’ are fluffy and adorable, and occasionally snarling.” For text, those patterns might be, “Were you raised in a” is often followed by “barn”. The computer makes millions or billions of these pattern notes, and stores them in what is known as a model.

To use the software, the human user inputs text, and then the machine creates an answer based on the notes in the model. For images, these answers start with random static that is then adapted iteratively until it matches the patterns in the notes baked into the model. That’s how tools like MidJourney create beautiful, unique images - start with something random, then refine it until it matches the d desired patterns. Text works a little differently, but still starts with randomness and moves towards something more plausible.

This highlights the opportunity and danger of using AI in professional work: current AI usually starts with randomness, and seeks to match patterns. Current AI does NOT usually check for accuracy. How could it? Models don’t contain a standard of objective truth to compare against - to an AI model, if something looks true or sounds true, it matches the patterns of things that ARE true. Appearance is fact for an AI. 

On the positive side, this means that AI tools can generate truly new things - the starting point of randomness will always introduce unexpected elements. As long as the output still matches the pattern to an acceptable degree, that randomness is allowed to remain; this gives the tools the illusion of creativity. It also means that, for tasks involving pattern recognition, AI is remarkably capable. This can be producing images, or finding key phrases in text, or anticipating common arguments to something that’s already been written. 

On the negative side, this also means that AI tools easily hallucinate - i.e. present something that looks and sounds true, but is objectively false or dangerous. A simple example: some cleaning solutions need to be diluted or mixed before they can be used. An AI would not necessarily be able to distinguish between instructions to dilute bleach with water (both liquids used in cleaning, usually a good idea), and instructions to dilute bleach with ammonia (both liquids used in cleaning, but a remarkably, terribly, disastrously bad idea). 

Also on the negative side, AI can be used to try and circumvent learning. For papers and assignments that primarily rely on pattern recognition, AI is very good doing those assignments for the student. Unfortunately, the simple assignments that beginning students often need to develop fundamental skills are also the most common ones.

At PowerNotes, we aren’t opposed to AI - in fact, we think it can be a marvelous tool to help students and professionals be more effective in research and writing, learn more, learn more quickly, and produce better final output. That’s the opportunity.

However, the danger described above means that we also strongly recommend that AI tools are used with human oversight and understanding so that those semi-random creations don’t lead to unfortunate consequences.

🔗 Read the next article in this series now!