Experiment with me: A user-friendly algorithm for chatting about climate change
Posted on May 20, 2023 by curryja
by Joe Nalven
I recently published one article about how to incorporate an AI chatbot as part of a college course. I focused on lying, persuasion and self-reflection. I went even further and anticipated a White House comprehensive plan to counter antisemitism. I used a chatbot to develop my own version of what should be in that plan. Now, I am off on another adventure using a chatbot to understand the ins and outs of climate change impacts. And I need your help. You can see my inquiry as similar to the struggle to get computers, other electronic devices and software to be more user friendly — such as the improvements through GUIs (graphic user interfaces). Perhaps AUIs (algorithm user interfaces) would be helpful — especially if users can insert their own parameters (or “biases”) with a preset language formula.
There are many articles that say that AI language models are not to be feared since they are just a tool. Others claim that more is at stake. I would suggest that in both cases the question is whether humans actually control the input/output process. Part of such control requires knowledge about that part of the process which we consider. As someone who is naïve about coding and how the AI black box works, I focus on the observable world of language. That’s what I have access to in framing a request and receiving a reply. So, in that small part of the AI universe – the one that has received substantial attention in the past several months – I would like to assert some control over the conversation with the AI “mindset.” For me, when wearing my hat as a digital artist, I learned digital editing with a host of algorithms that included filters, presets and other editing tools to compose my art. My “control” was going beyond just pushing a button. Now, with a language model, I am trying to discover whether I have any similar “control” over this domain and, if so, what are its limits and results.
As a human who thinks and believes that one can have some control of interactions with a chatbot – more than the illusion of control, I decided to experiment with creating my own algorithm to insert into this conversation. A user-defined algorithm that would add my own parameter to the chatbot output. The first question is whether that is really possible. When I approached ChatGPT, I received stonewalling and deflection and ultimately: “As an AI language model, I can’t directly assist in creating or fine-tuning algorithms outside the scope of the provided capabilities. My purpose is to provide information and assist with general inquiries to the best of my knowledge and abilities.” and “As an AI language model, I can provide information based on the writings and perspectives of the scientists you mentioned. However, it’s important to note that I don’t have direct access to specific writings or the ability to modify the underlying training data.” So, there are bumps in the road that might defeat my objective short of jailbreaking out of the programs guardrails.
https://judithcurry.com/2023/05/20/experiment-with-me-a-user-friendly-algorithm-for-chatting-about-climate-change/
