The Problem – GIGO
NOTE: The purpose of the next few lessons is to help you use AI to get better results from your thinking.
If you need to open your own personal ChatGPT account there is no fee. ACTION: Click here https://openai.com/product and go to MENU then LOGIN.
You may also use other AIs like Copilot, Claude, Perplexity etc.
Let’s get started. The written conversation in the next few lessons is between myself and ChatGPT …
Lesson 20 DFQ: What is the most interesting idea, or insight, you personally discerned from this lesson?
Next Lesson: Garbage Prompts
Use of AI is a skill just as thinking is a skill. If your thinking involves a flawed assumption including an ambiguous assumption, it doesn’t matter how good your subsequent thinking is, the result will also be flawed. Most assumptions are based on perception and no-one’s perception is perfect. Therefore all results that come from perception are inaccurate in some way which may not initially be apparent.
It is important to take care in how one phrases a question and it might be a good exercise to ask the question in a number of different ways and compare the different results.
GIGO means that if I have Garbage in my head and I train GPT with it, I will garbage out. However if I have garbage in my head, how would I know that I got garbage out, I mean, the machine will confirm my stupidity and I would be happy by the machine confirmation. We are making an assumption that what we put in is within the scope our knowledge and expertise. Some people believe that they “know” even when they do not know. So the question is, if I do not know what I do not know, how would I know if GPT’s GO, hence garbage out, is right or wrong?
GIGO is not only about the quality of data on which the AI system has been trained but also on the quality of the questions that you ask it. This is not intuitively obvious. It is easy to blame Chat GPT for its limitations, when poor prompting can exacerbate its limitations.
My most interesting idea I personally discerned from the lesson is in relation to the critical role the Quality of Inputs play in any system and that the computers or systems are limited in terms of miracles they can perform to improve the quality of the output or results they will give.
The other two related insights are that:
• The outputs or results we will get from the ChatGPT should not be taken to be the truth/accurate. They may need to be further validated.
• People with different capacity to formulate questions to the ChatGPT will receive different quality of outputs or results from it hence the need to work on updating our software.
Neil Brown …
It was valuable to impose on myself the discipline of providing clear and accurate information. That is sensible when what one is looking for is a factual response and I imagine that this is what AI is designed to do. In my case, having to make decisions which require a judgement on the evidence, it will be more difficult. I will be interested to see how these principles can be applied when there is conflicting evidence from, say, a Plaintiff and a Defendant and a decision has to be made as to which party should be believed, or evidence from one or the other of them which sounds dubious. In other words, how can AI assist in making judgments and in exercising a discretion? That remains to be seen, I suppose.
Insight: Ghat GPT or AI in general mimic mind, in that its just a recording surface
and in a very general sense the outputs are just a reflection of the inputs and too get the best output you need to be specific and give context.
There is an emerging skill I need to effective at ChatGPT prompts. Do develop this skills I need to reflect on how to best engineer:
• clear, and relevant questions for ChatGPT
• relevant information and context for ChatGPT
• straightforward and specific questions related to the topic I’m exploring with ChatGPT
I sometimes experience bias in the responses from ChatGPT on certain issues. It can be annoying when ChatGPT gives warnings or cautions for certain queries and responses but not for others. However, as I have learned to improve my prompts, I’ve noticed improvements in the answers.
My insight revolves around two key aspects of the human-in-the-loop interaction with ChatGPT: First, I have control over the quality of my prompt (Garbage In, Garbage Out). Second, I have a certain measure of control over the quality of ChatGPT’s response because I can loop back around and reprompt ChatGPT. This creates a continuous loop where I can provide context and perspective to enhance AI’s output. I refer to this loop as an AI-assisted Current View of the Situation (CVS) to Better View-Situations (BVS) loop. When I’m dissatisfied with ChatGPT’s output, I further refine my prompt by providing additional context and perspective.
In terms of bias, I am aware that the “Garbage In” refers to the bias I introduce through my prompts when selecting specific context, viewpoints, or perspectives to feed the machine. To minimize bias, I believe it is crucial to ensure that the prompts and instructions are as neutral and unbiased as possible. I have found that this is an iterative process. Additionally, I find ChatGPT useful for identifying and addressing biases in the writing of others. For instance, I have seeded ChatGPT by copying and pasting entire articles to provide the necessary context for applying my prompt and asking it to help me identify biases along with the biases of the author.
I’m really interested in the fact these Large Language models are trained on Human Knowledge and it’s the emergent properties that no one predicted are so interesting. Most of the people working with ChatGPT have traditional computer backgrounds.
If you theorise, these emergent properties are similar to those which emerged in early humans. it get’s really interesting.
Maybe, to get the most out of these models it’s not about being a good programmer, It’s fundamentally about being a good communicator and thinker.
Things that are deemed wasteful in programming are vital in good communication and having a mutually satisfying interaction with a fellow human.
Don’t worry I get ChatGPT is math, probability and trained on positive and negative results. Nothing like humans……..
GIGO is not only about the quality of data on which the AI system has been trained but also on the quality of the questions that you ask it. This is not intuitively obvious. It is easy to blame Chat GPT for its limitations, when poor prompting can exacerbate its limitations.