剛口教授在這篇文章中討論「人工智慧」是不是一個正確的「指號」。他的論點涉及語意學和語言哲學兩個領域概念多於科學和技術兩個領域。

剛口教授認為魏勒教授提出的「控制及通訊機制學」要比「人工智慧」來得貼切、適當。如剛口教授所說，前者不會讓人們聯想到「智慧」、「意識」、和「感知」這類其實跟「人工智慧」並無關聯的概念。


AI is not intelligent

AI should not be called AI

David J. Gunkel, 06/22/23

The act of naming is more than just a simple labeling exercise; it's a potent exercise of power with political implications. As the discourse around AI intensifies, it may be time to reassess its nomenclature and inherent biases, writes David Gunkel.

Naming is anything but a nominal operation. Nowhere is this more evident and clearly on display than in recent debates about the moniker “artificial intelligence” (AI). Right now, in fact, it appears that AI—the technology and the scientific discipline that concerns this technology—is going through a kind of identity crisis, as leading voices in the field are beginning to ask whether the name is (and maybe already was) a misnomer and a significant obstacle to accurate understanding. “As a computer scientist,” Jaron Lanier recently wrote in a piece for The New Yorker, “I don’t like the term A.I. In fact, I think it’s misleading—maybe even a little dangerous.”

What’s in a Name?

The term “artificial intelligence” was originally proposed and put into circulation by John McCarthy in the process of organizing a scientific meeting at Dartmouth College in the summer of 1956. And it immediately had traction. It not only was successful for securing research funding for the event at Dartmouth but quickly became the nom célèbre for a brand-new scientific discipline.

For better or worse, McCarthy’s neologism put the emphasis on intelligence. And it is because of this that we now find ourselves discussing and debating questions like: Can machines think? (Alan Turing’s initial query), are large language models sentient? (something that became salient with the Lemoine affair last June), or when might have an AI that achieves consciousness (a question that has been posed in numerous headlines in the wake of recent innovations with generative algorithms). But for many researchers, scholars, and developers these are not just the wrong questions, they are potentially deceptive and even dangerous to the extent that they distract us with speculative matters that are more science fiction than science fact.

Renaming AI

Since the difficulty derives from the very name “artificial intelligence,” one solution has been to select or fabricate a better or more accurate signifier. The science fiction writer Ted Chiang, for instance, recommends that we replace AI with something less “sexy,” like “applied statistics.” Others, like Emily Bender, have encouraged the use of the acronym SALAMI (Systematic Approaches to Learning Algorithms and Machine Inferences), which was originally coined by Stefano Quintarelli in an effort to avoid what he identified as the “implicit bias” residing in the name “artificial intelligence.”

Though these alternative designations may be, as Chiang argues, more precise descriptors for recent innovations with machine learning (ML) systems, neither of them would apply to or entirely fit other architectures, like GOFAI (aka symbolic reasoning) and hybrid systems. Consequently, the proposed alternatives would, at best, only describe a small and relatively recent subset of what has been situated under the designation “artificial intelligence.”

But inventing new names—whether it is something like that originally proposed by McCarthy or one of the recently proposed alternatives—is not the only way to proceed. As French philosopher and literary theorist Jacques Derrida pointed out, there are at least two different ways to designate a new concept: neologism (the fabrication of a new name) and paleonymy (the reuse of an old name). If the former has produced less than suitable results, perhaps it is time to try the latter.

Cybernetics

The good news is that we do not have to look far or wide to find a viable alternative. There was one already available at the time of the Dartmouth meeting with “cybernetics.” This term—derived from the ancient Greek word (κυβερνήτης) for the helmsman of a boat—had been introduced and developed by Norbert Wiener in 1948 to designate the science of communication and control in the animal and machine.

Cybernetics has a number of advantages when it comes to rebranding what had been called AI. First, cybernetics does not get diverted by or lost in speculation about intelligence, consciousness, or sentience. It is only concerned with and focuses attention on decision-making capabilities and processes. The principal example utilized throughout the literature on the subject is the seemingly mundane but nevertheless indicative thermostat. This homeostatic device can accurately adjust for temperature without knowing anything about the concept of temperature, understanding the difference between “hot” and “cold,” or needing to think (or be thought to be thinking).

Second, cybernetics avoids one of the main epistemological problems and sticking points that continually frustrates AI—something philosophers call “the problem of other minds.” For McCarthy and colleagues, one of the objectives of the Dartmouth meeting—in fact, the first goal listed on the proposal—was to figure out “how to make machines use language.” This is because language use—as Turing already had operationalized with the imitation game—had been taken to be a sign of intelligence. But as John Searle demonstrated with his Chinese Room thought experiment, the manipulation of linguistic tokens can transpire without knowing anything at all about the language. Unlike AI, cybernetics can attend to the phenomenon and effect of this communicative behavior without needing to resolve or even broach the question concerning the problem of other minds.

Finally, cybernetics does not make the same commitment to human exceptionalism that has been present in AI from the beginning. Because the objectives initially listed by the Dartmouth proposal (e.g., language use, form abstractions and concepts, solve problems reserved for humans, and improve themselves), definitions of AI tend to concentrate on the emulation or simulation of “human intelligence.” Cybernetics by contrast is more diverse and less anthropocentric. As the general science of communication and control in the animal and the machine, it takes a more holistic view that can accommodate a wider range of things. It is, as N. Katherine Hayles argues, a posthuman framework that is able to respond to and take responsibility for others and other forms of socially significant otherness. 

Back to the Future

If “cybernetics” had already provided a viable alternative, one has to ask why the term “artificial intelligence” became the privileged moniker in the first place? The answer to this question returns us to where we began—with names and the act of naming. As McCarthy explained many years later, one of the reasons “for inventing the term ‘artificial intelligence’ was to escape association with cybernetics” and to “avoid having either to accept Norbert Wiener as a guru or having to argue with him.” Thus, the term “artificial intelligence” was as much a political decision and strategy as it was a matter of scientific designation. But for this reason, it is entirely possible and perhaps even prudent to reverse course and face what the nascent discipline of AI had so assiduously sought to avoid. The way forward may be by going back.

David J. Gunkel is a professor at Northern Illinois University and author of The Machine Question and Person-Thing-Robot.

「圖林測試」是1950年代英國數學家和電腦/人工智慧先驅圖林博士提出：一個檢查機器是否具備和人類相當智力的方法。

由於人工智慧研發進展神速，圖林測試被認為已經趕不上時代。《科技彙報》上這篇文章提出對此類測試的新構想。

下文中附圖1呈現「人類智慧」的八個面向：

生存、語言、邏輯-數學、音樂、空間、身體-動作、人際、與思想。

附圖2 則是根據附圖1呈現「人類智慧」的八個面向對ChatGPT所做的評估：

a.  在語言和邏輯-數學兩個面向上，作者給ChatGPT等級3的評分；
b.  其它6個面向上ChatGPT的等級為1 (完全不具備)。

請至原網頁瀏覽這兩個附圖。

又：全文須在該網站註冊才能看到；以下僅轉載該網站刊出部分。

The AI revolution has outgrown the Turing Test: Introducing a new framework

Chris Saad, TechCrunch, 03/14/23

As AI becomes a transformative part of our technology landscape, a common vocabulary about the capabilities of each new tool and technique is essential. Common vocabularies create shared intellectual spaces allowing all stakeholders to accelerate understanding, increase adoption, facilitate collaboration, benchmark progress and drive innovation.

So far, the most widely known tool for benchmarking AI is the Turing Test.

However, the field of artificial intelligence (AI) has come a long way since the inception of the Turing Test in 1950. As such, it is becoming increasingly clear that the Turing Test is insufficient for evaluating the full range of AI capabilities that are emerging today — or are likely to emerge in the future.

The Turing Test operates on a simplistic pass/fail basis and focuses heavily on chat/linguistic capability, which is only one aspect of human intelligence. This narrow focus on language ignores many other critical dimensions of intelligence, such as problem-solving, creativity and social awareness. Additionally, the Turing Test presupposes a level of human-like intelligence that may not be relevant or useful for evaluating AI.

The framework

To address these limitations, there is an urgent and critical need to develop a more nuanced and comprehensive framework for evaluating AI capabilities across multiple dimensions of intelligence.

This insight led me to develop the “AI Classification Framework.” The ACF is a new approach to evaluating AI capabilities based on the Theory of Multiple Intelligences.

The Theory of Multiple Intelligences was first proposed by psychologist Howard Gardner in 1983. Gardner argued that intelligence was not a single, unified entity but rather a collection of different abilities that could manifest in a variety of ways. Gardner identified eight different types of intelligence: According to Gardner, individuals may excel in one or more of these areas, and each type of intelligence is independent of the others. The theory challenged the traditional view of intelligence as a singular, fixed entity and opened up new avenues for exploring the diversity of human cognition. While the theory of multiple intelligences has been subject to some criticism and debate over the years, it has had a significant impact on the field of psychology and education, particularly in the development of alternative approaches to teaching and learning.

This seemed perfect as a basis for the AI Classification Framework. Following the theory, the framework supports evaluating AI tools across multiple dimensions of intelligence, including linguistic, logical-mathematical, musical, spatial, bodily-kinesthetic, interpersonal and intrapersonal intelligence.

For each dimension of intelligence, the framework provides a scale from 1–5 with 1 being “No Capability” or the equivalent of a human infant and 5 being “Self-agency” or capability that might be considered “Super Intelligence” — beyond human ability.

Examples

The framework itself is a detailed table of descriptions that can be found here. I’ve also created a simple visual representation for easy high-level reference. What follows are two simple examples of the segments that respond to the capabilities of ChatGPT and DALL-E 2.

(待續)