人工智慧與人類的未來

Abstract

本文基於尤瓦爾·諾亞·赫拉利在其著作《人類大歷史》中關於人工智慧的論點。赫拉利論點的獨特之處在於，他認為即使是弱人工智慧也足以取代人類，而大多數人則認為需要強人工智慧或通用人工智慧才能取代人類。我將提出三個可能反駁赫拉利的論點，但最終都將失敗。接下來，我將把討論擴展到通用人工智慧。布萊恩·坎特韋爾·史密斯、約布斯特·蘭德格雷貝和巴里·史密斯對此持反對意見。他們的論點適用於強人工智慧，但在弱人工智慧方面則不然。最後，我將基於黑格爾關於有限性和無限性的哲學提出一個論點，該論點可以駁斥赫拉利的弱人工智慧足以取代人類的觀點。

赫拉利認為，即使是弱人工智慧也足以分析人類行為，並且在擁有足夠的數據後便能更準確地預測人類的行為。數據越多，預測就越準確；隨著預測變得越來越準確，人們就會越來越少地自己做出決定，而是選擇遵循人工智慧做出的預測。一旦你把所有的決策都交給人工智慧，人類的價值就所剩無幾了。人工智慧如何在沒有首先獲得理解和意識的情況下分析並預測人類的行為？赫拉利認為，答案不在於電腦科學，而 在於生物學。生物學家發現，人類深受基因的影響，並通過對不同因素的複雜計算做出決定。這意味著人類可以被視為通過一系列計算來回應世界的演算法。如果人類是演算法，那麼人工智慧只需要分析人類如何作為演算法進行這些計算，就能預測他們的下一步行動。人工智慧對人類的預測不必完全準確，如果它們的準確性高於人類自身的判斷，那麼人類就應該合乎邏輯地遵循人工智慧的預測，而不是自己的判斷。理解和意識不是必需的，因為人工智慧所做的只是分析人類過去的行為，並用它來預測他們未來的行為。

第一個論點是賽爾著名的中文房間論證。第二個論點來自托馬斯·福克斯，他在其論文《自戀抑鬱的科技》中也批評了赫拉利的論點。第三個論點要麼是否認人類必須遵循人工智慧的預測，要麼是規範人工智慧的發展。賽爾的論證可以分為幾個部分。想像一下你身處一個充滿符號和符號操作規則的房間裡。一組符號進入房間，你按照給定的規則操作房間裡的符號來做出回應。你不知道的是，進入房間的那組符號是一個中文問題，而你做出的回應是一個完美的中文答案。在這種情況下，你理解中文嗎？賽爾的答案是否定的，這表明理解不能僅僅通過符號的操作獲得，而這正是人工智慧所做的。托馬斯·福克斯反對赫拉利關於人類可以被簡化為僅僅是演算法的說法。他認為，人類是具有生物身體和感受能力的生命體，而人工智慧缺乏這兩者。為了抵制人工智慧的誘惑，人類必須擁抱他們有限的生物身體並對此感到滿足。第三個論點很簡單，人類可以簡單地選擇不遵循人工智慧的預測，或者我們可以像馬蒂亞斯·里塞建議的那樣規範人工智慧的發展。即使人工智慧為我們提供了最佳解決方案，我們人類也可能更喜歡不太理想的選擇。例如，人工智慧可能會根據一個人的偏好以及期待觀看這部電影幾個月的事實，預測他會觀看某部動作片。然而，我們仍然可以在最後一刻改變主意，簡單地選擇不觀看這部電影。

這三個論點將被證明不足以反駁赫拉利的論點。因此，要解決弱人工智慧通過準確預測取代人類的問題，就必須找到隱藏在人類是演算法的前提中可以推出人類不只是演算法的特徵。這樣的特徵可以在黑格爾關於有限性和無限性的哲學中找到。根據黑格爾的觀點，人類是有限的存在，但他們也超越了自己的有限性，達到了無限。無論弱人工智慧變得多麼先進，它都無法超越自身的有限性，達到無限。

總之，尤瓦爾·諾亞·赫拉利提出了一個有趣的論點，即弱人工智慧足以取代人類。針對強人工智慧的論點在這種情況下是不充分的，因為赫拉利並沒有聲稱人工智慧可以獲得理解或意識。解決弱人工智慧問題的方案是找到隱藏在人類是演算法的前提中足以證明人類不只是演算法的特徵。這個隱藏的特徵是人類超越個體有限性的能力。

This thesis is based on Yuval Noah Harari’s arguments on ai in his book Homo Deus: A Brief History of Tomorrow. First, there is a difference between weak AI and strong AI. Weak AI specializes in a certain field of work, while strong AI can operate in multiple fields with proficiency at least equal to humans. What makes Harari’s argument different is that he argues that weak AI is enough to replace humans when most argue that strong AI or artificial general intelligence that usually gains understanding or consciousness, is required to replace humans. I will present three possible counter arguments to Harari that ultimately fail. Then, I will extend the discussion to artificial general intelligence. Brian Cantwell Smith, Jobst Landgrebe and Barry Smith opposes the likelihood of artificial general intelligence. Their arguments are great for strong AI but not so much for weak AI. Finally, I will present an argument based on Hegel’s philosophy on finitude and infinity which can refute Harari’s argument that weak ai is enough to replace humans. So, the argument based on Hegel’s philosophy will be my answer to Harari’s argument.

Harari believes that even weak artificial intelligence will be able to analyze human behavior, and with sufficient data, make accurate predictions on humans’ actions. With more data the more accurate the predictions, and as the predictions become more accurate the less people would make decisions for themselves, instead opting to follow the predictions AI has made. Once you leave all the decision making in the hands of AI, there is little to no value left for humanity. How is it possible for artificial intelligence to analyze and thus predict humans’ actions without gaining understanding and consciousness first? Harari argues that the answer lies not in Computer science but in biology. Biologists find out that humans are deeply affected by their genes and make decisions through a complicated calculation of different factors. This means that humans can be seen as algorithms that respond to the world through a series of calculations.

If humans are algorithms, then artificial intelligence only needs to analyze how humans make these calculations as algorithms to predict their next move. Artificial intelligence’s predictions on humans don’t have to be perfectly accurate, as long as their accuracy is better than humans’ own judgement, humans should logically follow the prediction of AI over their own. Understanding and consciousness isn’t required since all the AI is doing is analyzing what humans did before and using it to predict what they will do in the future.

The first of the three possible counter arguments against Harari is Searle’s famous Chinese room argument. The second argument is from Thomas Fuchs, who also criticizes Harari’s arguments in his paper “Narcissistic Depressive Technoscience”. The third argument is to either deny that humans must follow the predictions from AI or regulate the development of AI.

Searle’s argument can be broken down into several parts. Imagine yourself in a room filled with symbols and rules on how to manipulate these symbols. A set of symbols come into the room, and you respond by manipulating the symbols in the room following the given rules. Unknown to you, the set of symbols coming in is a question in Chinese, and the response you made is an answer in perfect Chinese. Does one understand Chinese after this? Searle’s answer is no, which shows that understanding cannot be gained through only manipulation of symbols which is what ai does.

Thomas Fuchs rejects Harari’s claim that humans can be reduced to mere algorithms. He argues that humans are living beings with a biological body and the ability to feel, both of which artificial intelligence lacks. To reject the temptation of AI, humans must embrace their limited biological body and be content with it. The third argument is simple, humans can simply choose not to follow the predictions from AI, or we could regulate AI like Mathias Risse suggested. Even if AI presents us with the best solution, we humans may prefer less optimal options. For example, AI could make a prediction that one will watch a certain action film based on one’s preferences and the fact that one is looking forward to watching this movie for several months. However, we could still change our minds at the last second and simply choose not to watch this movie.

The three arguments will be proved insufficient against Harari’s argument. So, to solve the problem of weak AI replacing humans through accurate predictions one must find what is hidden in the assumption that humans are algorithms that leads to the conclusion that humans are not merely algorithms afterall. This means that even if Harari’s claim that artificial intelligence could analyze humans as algorithms are assumed, there is more to humans as algorithms. Such features can be found in Hegel’s philosophy on finitude and infinity. According to Hegel, humans are finite and limited beings, but they also transcend their finitude and reach into infinity. Weak ai no matter how advanced it becomes cannot transcend its finitude and reach infinity.

In conclusion, Yuval Noah Harari brings up an interesting argument that weak ai is enough to replace humans. Arguments that are directed towards strong ai are insufficient in this scenario, since Harari isn’t claiming that ai can gain understanding of consciousness. The solution to the problem of weak AI is to find a hidden feature in the assumption that humans are algorithms that proves humans are not merely algorithms afterall. This hidden feature is humans’ ability to transcend the finitude of the individual.