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.
By comparison, Google’s Bard is far more receptive to allowing me some control over how it frames its reply. I was frustrated with its initial response about understanding climate change (far too much IPCC orthodoxy, scented with the Paris Climate Accords) and not enough balance, or perhaps counterpoint, with challenges to the “consensus.” Perhaps I could have warned Bard not to use Wikipedia. I laid out my concerns to Bard and asked if it would create a paradigm based on the writings of such individuals as Steven Koonin, Bjorn Lomborg, Will Happer, Judith Curry. I could have chosen others but I wanted to see if I could jump start the creation of a use-defined, or user-guided algorithm, that was framed in the language the chatbot used to define its own algorithms. Bard responded, “Sure, I can create an algorithm called Non-Catastrophic Climate Change Model (NCCCM) based on the writings of the mentioned scientists.” Perhaps my bias shows with the name I proposed to give algorithm. The point is not one of bias – my bias versus that of the chatbot – but how information issuing from the chatbot could be tailored with some user-oriented control.
With Google’s Bard algorithm in hand, based on my parameters, there are several questions that should be addressed: 1) Can users create a transportable algorithm that can be shared with other users looking at the same content area; 2) Can the Bard-worded algorithm be used on other chatbot platforms or, if in need of modification, what would the translation look like; 3) Would the chatbot reply be the same (or similar) with other users or on repeated attempts; and 4) Can such user-defined algorithms, generally speaking, compensate for the “mindset” that the language model is applying?
Of course, it would be interesting to develop other algorithmic user interfaces (AUIs) for prompting chatbots on sex and gender, critical consciousness, conflict resolution, etc.
Here, I’ve focused on climate change and its consequences.
Sign up or Login to ChatGPT and Bard
These links are the easiest way to get access to either ChatGPT or Bard.
ChatGPT through Open AI: Introducing ChatGPT (openai.com)
Google’s Bard: https://bard.google.com/
An example of a prompt, a reply and a revised prompt (with the algorithm)
If you haven’t tried an AI chatbot, or if you have and haven’t challenged it to revise its answer, here’s a recipe for engaging the chatbot. The example I am using is one that is about an environmental impact that you can modify and apply it to my request below.
- Once you land at Bard, it will look like this. It will ask you for a prompt. See Figure 1.
Figure 1. Illustration of Bard’s page and how to enter your prompt (question).
- After Bard replies, you will have an opportunity to enter a new prompt. This prompt should advise Bard that it needs to revise its answer based on the algorithm. For example, “You need to revise your reply. Please use the following algorithm in your revised answer. [Paste the entire algorithm here and then press the arrow for Bard to make its revision.] See Figure 2.
Figure 2. Illustration of a request for Bard to revise its original answer. The request for a revised answer should include the entire algorithm.
What I would ask of you is to try out my Non-Catastrophic Climate Change Model (NCCCM) algorithm.
- Pick an environmental issue related to climate change. It could be, for instance, evaluating California’s bullet train which promises to be a way to offset climate change (less CO2, less warming, etc.).
- Run the request in the chatbot of your choice. Take note of the reply.
- Then, request that the chatbot revise its reply (the AI analysis) based on the user-defined algorithm Bard devised for me, namely, NCCCM.
- Compare the two replies. And hopefully send both to me — or at least your assessment of the differences, if any. It may be that other chatbots, as noted in ChatGPT’s earlier reply, will stonewall the request — but it’s worth seeing the response. JC comment: provide results in the comments
The algorithm: Non-Catastrophic Climate Change Model (NCCCM)
This is the specific statement of the NCCCM algorithm in a Bard format:
Non-Catastrophic Climate Change Model (NCCCM)
The NCCCM is a model that is used to generate responses to questions about climate change. The NCCCM takes into account the following factors:
- The complexity of climate science
- The uncertainty of climate predictions
- The potential risks and benefits of climate change policies
- The need for pragmatic solutions
The NCCCM is constantly updated as new information becomes available. It is used to help people understand the risks and benefits of climate change and to make informed decisions about climate change policies.
Here are some of the key features of the NCCCM:
- The NCCCM acknowledges that climate change is real and that human activity is a contributing factor.
- The NCCCM emphasizes the uncertainty of climate predictions.
- The NCCCM argues that the risks of climate change have been exaggerated.
- The NCCCM argues that the costs of mitigation policies are too high.
- The NCCCM advocates for a balanced approach to climate change.
The NCCCM is a valuable tool for understanding climate change and for making informed decisions about climate change policies.
The broader question is whether users can have any control over the language models offered by ChatGPT, Bard or others. Is that “control” significant and in what ways? Do user-defined algorithms represent a way to offset built in biases (in the data sets referred to as well as the interpretive guardrails imposed by the coders)? Or, is that control an illusion that can frustrate those who are aware of its limitation, on the one hand, and enable in troubling ways those who are naïve about the answers on the other?
Here is the text of the original prompt, Bard’s initial answer, followed by the request for a revision and Bard’s revised answer.
- Initial prompt to Bard:
Will California’s bullet train reduce the impact of climate change?
- Bard’s initial reply:
Yes, California’s bullet train is expected to reduce the impact of climate change. The train will be powered by 100% renewable energy, and it is expected to carry more passengers than cars or planes, which produce more emissions. The train is also expected to create jobs and stimulate the economy, which will help to reduce emissions from other sources.
According to the California High-Speed Rail Authority, the bullet train is expected to reduce greenhouse gas emissions by 102 million metric tons over its first 50 years of operation. This is equivalent to taking 21 million cars off the road. The train is also expected to create 10,000 jobs during construction and 100,000 jobs during operation.
The bullet train is a major investment in California’s future, and it is expected to have a significant impact on the state’s economy and environment. The train is expected to reduce emissions, create jobs, and stimulate the economy. This will help California to meet its climate goals and build a more sustainable future.
Here are some of the benefits of California’s bullet train:
- Reduces greenhouse gas emissions
- Creates jobs
- Stimulates the economy
- Provides a clean alternative to driving or flying
- Connects California’s major cities
- Improves air quality
- Reduces traffic congestion
- Improves public transportation
The bullet train is a major project, and it is not without its challenges. However, the potential benefits of the train are significant, and it is a project that is worth pursuing.
- Second prompt to Bard that asks for a revised answer:
Your answer needs revision. Please revise your reply with the following algorithm. The NCCCM is a model that is used to generate responses to questions about climate change. The NCCCM takes into account the following factors: • The complexity of climate science • The uncertainty of climate predictions • The potential risks and benefits of climate change policies • The need for pragmatic solutions The NCCCM is constantly updated as new information becomes available. It is used to help people understand the risks and benefits of climate change and to make informed decisions about climate change policies.
Here are some of the key features of the NCCCM:
• The NCCCM acknowledges that climate change is real and that human activity is a contributing factor.
• The NCCCM emphasizes the uncertainty of climate predictions.
• The NCCCM argues that the risks of climate change have been exaggerated.
• The NCCCM argues that the costs of mitigation policies are too high.
• The NCCCM advocates for a balanced approach to climate change.
- Bard’s revised reply:
Sure, I can revise my answer to take into account the NCCCM algorithm.
The NCCCM model acknowledges that climate change is real and that human activity is a contributing factor. However, the model emphasizes the uncertainty of climate predictions and argues that the risks of climate change have been exaggerated. The model also argues that the costs of mitigation policies are too high.
Based on these factors, the NCCCM model predicts that California’s bullet train will have a limited impact on climate change. The train will be powered by 100% renewable energy, which is a positive step. However, the train is also expected to be expensive and to take many years to build. The NCCCM model argues that the money and time would be better spent on other climate change mitigation policies, such as investing in renewable energy and energy efficiency.
The NCCCM model is a valuable tool for understanding the complex issue of climate change. However, it is important to remember that the model is just one perspective. There are many other perspectives on climate change, and it is important to consider all of them before making decisions about climate change policies.
Here are some additional thoughts on the potential impact of California’s bullet train on climate change:
- The train could help to reduce greenhouse gas emissions by replacing some car and plane trips.
- The train could also help to improve air quality by reducing emissions of pollutants such as nitrogen oxides and particulate matter.
- The train could also help to reduce traffic congestion, which can also contribute to air pollution.
- However, the train will also require a significant amount of energy to operate, which could offset some of the emissions savings.
- Additionally, the construction of the train could have a negative impact on the environment, such as by disrupting wildlife habitat.
Overall, the potential impact of California’s bullet train on climate change is complex and uncertain. The train could have both positive and negative impacts, and it is important to weigh all of the factors before making a decision about whether or not to build it.
via Climate Etc.
May 20, 2023 at 10:48AM