I have decided to create a new prestigious and highly-coveted award: the Zbigatron Paper of the Year Award.
And I hereby officially bestow the 2024 award to Hicks et al. for their academic publication entitled “ChatGPT is Bullshit“ (Hicks, M.T., Humphries, J. & Slater, J. ChatGPT is bullshit. Ethics Inf Technol 26, 38 (2024)). What a paper. A breath of fresh air in the world of hype, lies, and financial bubbles that surround Artificial Intelligence today.
The premise of the paper is this: we should stop using terms like “hallucination” for situations where LLMs make up information and present them as facts because this is an inaccurate description of this phenomenon.
Now, I have been a huge champion of using more accurate terms to describe actions or attributes of machines that are deemed to be artificially intelligent. For example, in an article I wrote two years ago (entitled The Need for New Terminology in AI) I stated:
Terms like “intelligence”, “understanding”, “comprehending”, “learning” are loaded and imply something profound in the existence of an entity that is said to be or do those things… [T]he problem is that the aforementioned terms are being misunderstood and misinterpreted when used in AI.
And then we go and create an AI hype bubble as a result. So, in that article, I called for more precise terms to be substituted for these, such as using “Applied Statistics” in place of “Artificial Intelligence”. (Can you picture how the hype around AI would diminish if it was suddenly being referred to as Applied Statistics? This is a much more accurate term for it, in my opinion.)
Indeed, Hicks et al. have gone for the same approach classifying the phenomenon of hallucinations as something completely different. I need to quote their abstract to convey their message:
We argue that these [hallucinations], and the overall activity of large language models, is better understood as bullshit in the sense explored by Frankfurt (On Bullshit, Princeton, 2005)… We further argue that describing AI misrepresentations as bullshit is both a more useful and more accurate way of predicting and discussing the behaviour of these systems. [emphasis mine]
Yes! Please! Let’s start using more accurate terms to describe the phenomenon of AI. I highly agree that bullshit is a proper, scientifically-based, and sophisticated term that should be used in today’s day and age.
I need to drop some more quotes from this paper. It really does deserve my award:
Because these [LLMs] cannot themselves be concerned with truth, and because they are designed to produce text that looks truth-apt without any actual concern for truth, it seems appropriate to call their outputs bullshit.
And then this:
Currently, false statements by ChatGPT and other large language models are described as “hallucinations”, which give policymakers and the public the idea that these systems are misrepresenting the world, and describing what they “see”. We argue that this is an inapt metaphor which will misinform the public, policymakers, and other interested parties. [emphasis mine]
Finally, somebody calling out the BS (pun intended) for what it is. Like I said, what a breath of fresh air. And how important is this!? The public, policymakers, and other interested parties are making very important decisions based on false information.
It’s a classic case of PR talk, isn’t it?
I recently read an article (The Current State of AI Markets) that tried to quantify where revenue has occurred thus far in the AI Value Chain. We all know that companies are spending a ridiculous amount of money on AI – so what’s the current ROI on this looking like?
To quote the article:
Amazon, Google, Microsoft, and Meta have spent a combined $177B on capital expenditures over the last four quarters… We haven’t seen wide-scale application revenue yet. AI applications have generated a very rough estimate of $20B in revenue.
As the article admits: it’s early days yet and the ROI may come in the future. Nonetheless, one cannot ignore the divide between expenditure and ROI.
So, when we need to call a spade a spade, it’s important that we do so. This is not a joke, nor a game. Like I have said in the past: “There’s a ridiculous amount of money being spent, passed around, and invested and a lot of it is built on a false idea of what AI is capable of and where it is going. People are going to get hurt. That’s not a good thing.”
I’m going to leave the final word on this very important topic to the official winner of the 2024 Zbigatron Paper of the Year Award:
Investors, policymakers, and members of the general public make decisions on how to treat these machines and how to react to them based not on a deep technical understanding of how they work, but on the often metaphorical way in which their abilities and function are communicated. Calling their mistakes ‘hallucinations’ isn’t harmless: it lends itself to the confusion that the machines are in some way misperceiving but are nonetheless trying to convey something that they believe or have perceived. This, as we’ve argued, is the wrong metaphor. The machines are not trying to communicate something they believe or perceive. Their inaccuracy is not due to misperception or hallucination. As we have pointed out, they are not trying to convey information at all. They are bullshitting.
To be informed when new content like this is posted, subscribe to the mailing list:
(Note: If this post is found on a site other than zbigatron.com, a bot has stolen it – it’s been happening a lot lately)
