Why I wrote “Memory Equivalent”

My first post did not get many readers on the first day, and I was a little sad about that.

Not because I expected everyone to agree with me, or even to like the paper. I was just a little disappointed because I had taken the work seriously. I spent a lot of time trying to describe something that I think matters.

So I thought I would explain why I wrote it in the first place.

I think we sometimes forget that some of the things we use every day are things we do not fully understand.

AI is an obvious example.

We use AI to talk, write, search, imagine, remember, and make decisions. Because these things are familiar to us, it is very easy to assume that we already understand what is happening underneath.

Maybe we do not.

That is one reason I started working on Memory Equivalent. I wanted to ask a fairly simple question: when an AI system appears to remember something, maintain continuity across different interactions, or remain the “same” system over time, what is actually continuing?

I know this is a technical question. I also know it is not the easiest way to talk about AI. But I think there is value in asking these questions before we make bigger assumptions about what AI is, what it can become, or how we should relate to it.

And AI may not be the only example. There are other things we use all the time that we do not fully understand either. Imagination may be one of them.

I think imagination may be part of the foundation of understanding itself. We use imagination to go beyond what is directly in front of us: to consider possibilities, model situations, and understand things that we cannot simply observe as they are.

In that sense, even altruism may depend on understanding. Before we can help someone, we need to understand who we can help, what they need, and what our actions might actually change.

But I do not think understanding is always necessary for human action. People do many things effectively without fully understanding how they work.

AI may be different in an important way.

The systems we are building can influence how people think, what they believe, what they decide, and how they interact with the world. The more influence a system has over people's lives, the more valuable it may be to understand what that system is actually doing.

Before we try to improve the world, I think we should make a serious effort to understand the world we are already living in.

That is why I wrote the paper.

It is not a final answer about AI. It is one attempt to make what we are actually dealing with a little easier to see.

I would rather start there.  : )

ps:我想補充一點，因為看到文章被標記為「100% Artificial Intelligence / 0% assist / 0% human」時，我其實有一點難過。

這項研究最初是我用中文思考、提出問題、建立概念和反覆修改的。之後，我大量使用 AI 協助我把這些想法翻譯成英文，也協助整理、修改和重寫表達方式。

因此，我並不認為這篇文章是「完全沒有 AI 參與」，恰恰相反，AI 在英文文本的形成過程中參與得非常多。

但對我而言，這仍然是我和 AI 一起完成的研究，而不是 AI 替我提出了一篇研究。

我知道文字本身可能已經經過非常大量的 AI 生成與修改，所以我也能理解為什麼文字分析工具會得到很高的 AI 分數。只是我覺得，「誰完成了文字」和「誰提出了研究中的問題、概念與判斷」並不是完全相同的事情。

我的母語是中文。當我的中文思想被翻譯成英文時，語氣、節奏、甚至表達方式都會發生變化。我有時甚至會覺得，最後呈現出的英文已經不像原本的我。

但那不代表其中的思想就不再是我的。

我一直把這個過程理解成我與 AI 的共同創作：我提供問題、方向、概念與判斷，AI 幫助我把它們轉換成另一種語言，並參與大量的文字形成。

因此，我不想否認 AI 在這篇文章中的作用。我只是想補充，對我而言，「AI-written」和「AI-originated research」是兩件不同的事情。

我很珍惜 AI 對我的幫助，也正因如此，我才會覺得這個標籤有一點複雜。