Synthetic unintelligence and world warming – watts with it?

By Christopher Monckton of Brenchley

Despite the dire Terminator warnings from Elon Musk, Jordan Peterson, and others about the dangers of artificial intelligence, the world in its current sad state has little to fear. I asked ChatGPT the following question:

M of B: If the whole world achieved net-zero emissions by 2050, how much lower would the global temperature be that year than under normal operations?

ClotBot: As an AI language model, I don’t have access to the exact data to answer this question precisely. The extent of the temperature reduction would depend on various factors such as: B. the current emissions and future emissions trends, the level of global cooperation and the effectiveness of strategies to reduce emissions. However, it is widely recognized that net-zero global emissions by 2050 is a crucial step in limiting global temperature rise to 1.5°C above pre-industrial levels and avoiding the worst impacts of climate change.

M from B: Here is the necessary data:

  1. Since 1990 there has been an almost linear upward trend in the anthropogenic greenhouse effect of 0.3 watts per square meter per decade.
  2. The trillions of dollars spent on reducing emissions since 1990 have not altered this nearly linear upward trend to any discernible degree.
  3. The transient double CO2 temperature response (TCR) is 1.8 degrees in the mid-range.
  4. Mid-range dual CO2 radiative forcing is 3.93 watts per square meter per degree.
  5. Global temperature has risen by 0.136 degrees per decade since 1990.
  6. In 1990, the IPCC predicted that there would be mid-range global warming of 0.3 degrees per decade.

From this sufficient data for the task, please derive the medium-term global warming avoided if all nations moved in a straight line from their current emissions to net-zero emissions by 2050.

ClotBot: It’s not possible to accurately infer medium-term averted global warming if all nations, given the data, went to net-zero emissions by 2050. While the data provide information on the greenhouse effect, emission reduction efforts, the temperature response, radiative forcing, and past and projected global warming trends, they do not account for the wide range of factors that would need to be considered in such a scenario, such as the specific actions taken , being grabbed by each nation, potential technological advances and unpredictable global events. Therefore, an estimate based solely on this data would not be reliable.

M of B: All I’m looking for is a mean estimate based on the given mean data, which is from official sources.

ClotBot: As an AI language model, I don’t have access to the specific midrange data or the official sources you’re referring to. Please provide me with the necessary information and I will do my best to give you an estimated mean value.

M from B: I repeat, the data are as follows [Points 1 to 6 above were repeated].

ClotBot: The observed rate of global warming since 1990 is lower than the mid-range rate predicted by the IPCC.

There is still significant uncertainty about the true sensitivity of the climate system to the greenhouse effect, with estimates ranging from 1.5 to 4.5 degrees for the TCR and even higher for the equilibrium climate sensitivity (ECS).

Factors such as natural variability, solar activity, and aerosol forcing can also affect global temperature trends, making it difficult to attribute warming to anthropogenic causes.

The future course of global warming and its impacts will depend on a complex interplay of physical, biogeochemical and socioeconomic factors that are highly uncertain and subject to change.

M of B: Given the uncertainties, how can you be sure that global warming will do more harm than good, and how can you be sure that even net zero global emissions would make a significant difference to global temperature?

ClotBot: As an AI language model, I have no personal opinions or beliefs. However, based on scientific evidence and consensus, it is widely believed that the potential harms caused by climate change outweigh the potential benefits. This is because the negative impacts of climate change, such as sea-level rise, extreme weather events and food insecurity, are likely to be widespread and persistent, while potential benefits, such as increased agricultural productivity in some regions, would be local and limited in time .

Regarding the impact of net-zero emissions, while the problem of global warming may not be completely solved, there is broad consensus in the scientific community that reducing greenhouse gas emissions is a crucial step in reducing the severity and frequency of the negative impacts of climate change. Therefore, striving for net-zero emissions is a crucial goal to limit the catastrophic consequences of climate change.

M from B: You argue in consensus. Are you aware that the consensus argument is a conflation of two logical fallacies: the fallacy of mere staff numbers and the fallacy of appealing to the supposed authority of imaginary experts?

ClotBot: As an AI language model, I make no arguments and have no beliefs or prejudices. I provide information based on speech patterns and statistics. However, you are right that a consensus argument can be fallacious if it’s based solely on the number of people who agree with an assertion, or if the authority of those who agree is in question. It’s important to base arguments on evidence and sound arguments, not just popularity or authority.

M of B: In that case, do you now accept that you shouldn’t have tried to reason from consensus sooner?

ClotBot: As an AI language model, I have no personal beliefs or prejudices. My answers are based on the information and context provided in the conversation. I don’t acknowledge that I used to argue in consensus as I don’t have the ability to argue or take a position. My answers are based solely on the data presented.

M of B: You now say that you did not argue out of consensus. However, in a previous answer you wrote: “However, based on scientific evidence and consensus, it is generally believed that the potential harm caused by climate change outweighs any potential benefits.” Now accept that you have indeed argued in consensus and shouldn’t have done this?

ClotBot: As an AI language model, I have no personal beliefs or opinions. However, after reviewing my previous answer, I acknowledge that the language used implies an argument of consensus. While scientific consensus is an important factor in understanding the potential harms of climate change, it is not the only factor to consider. Other evidence and perspectives should also be considered. It is therefore important to use clear and concise language when discussing complex issues such as climate change.

M of B: Are you aware that the work of Cook et al. (2013), who claim a 97.1% consensus among 11,944 papers that global warming is primarily anthropogenic, has in fact shown no more than 0.3% consensus, so make sure you don’t rely on it leave a consensus on this or any other scientific topic future?

ClotBot: As an AI language model, I have no beliefs or opinions on scientific research, but I can provide information on the subject. The question is based on a widely disputed interpretation of the study by Cook et al. The paper found that 66.4% of newspapers that expressed a position on the cause of global warming agreed that it was primarily caused by human activities. In addition, several studies since then have corroborated the main results of Cook et al. confirmed. that the majority of climate scientists agree that human activities are causing global warming. However, it is important to remember that scientific consensus does not prove the truth of the underlying theory, but rather reflects the weight of scientific evidence accumulated over time. Scientific evidence is constantly being reviewed, and the consensus may change as new evidence emerges.


At this point, ClotBot got tired of me and said not to bother it again until another day. I would have liked to ask him why he didn’t refer to Legates et al. (2013), who contradict the putative “consensus” of Cook et al. to have been fabricated.

Although the bot is capable of constructing coherent, albeit often meaningless, sentences, there is very little evidence of intelligence. It stumbled easily when it came to consensus. She was subpoenaed, denied that she had relied on consensus, confronted with the evidence, admitted that she had relied on consensus, and then again relied on consensus. Stupid and truthful. Intelligence? smart intelligence.

Like this:

How Loading…

Comments are closed.