The Chess Player who Couldn't Pass the Salt

AI: Soft and hard, weak and strong, narrow and general

Dear KV,

Our company is looking at handing much of our analytics to a company that claims to use "Soft AI" to get answers to questions about the data we have collected via our online sales system. I've been asked by management to evaluate this solution, and throughout the evaluation all I can see is that this company has put a slick interface on top of a pretty standard set of analytical models. I think what they really mean to say is "Weak AI" and that they're using the term Soft so they can trademark it. What is the real difference between soft (or weak) AI and AI in general?

Feeling Artificially Dumb

Dear AD,

The topic of AI hits the news about every 10 to 20 years, whenever a new level of computing performance becomes so broadly deployed as to enable some new type of application. In the 1980s it was all about expert systems. Now we see advances in remote control (such as military drones) and statistical number crunching (search engines, voice menus, and the like).

The idea of artificial intelligence is no longer new, and, in fact, the thought that we would like to meet and interact with non-humans has existed in fiction for hundreds of years. Ideas about AI that have come out of the 20th century have some well-known sources—including the writings of Alan Turing and Isaac Asimov. Turing's scientific work generated the now famous Turing test, by which a machine intelligence would be judged against a human one; and Asimov's fiction gave us the Three Laws of Robotics, ethical rules that were to be coded into the lowest-level software of robotic brains. The effects of the latter on modern culture, both technological and popular, is easy to gauge, since newspapers still discuss advances in computing with respect to the three laws. The Turing test is, of course, known to anyone involved in computing, perhaps better known than the halting problem (https://en.wikipedia.org/wiki/Halting_problem), much to the chagrin of those of us who deal with people wanting to write "compiler-checking compilers."

The problem inherent in almost all nonspecialist work in AI is that humans actually don't understand intelligence very well in the first place. Now, computer scientists often think they understand intelligence because they have so often been the "smart" kid, but that's got very little to do with understanding what intelligence actually is. In the absence of a clear understanding of how the human brain generates and evaluates ideas, which may or may not be a good basis for the concept of intelligence, we have introduced numerous proxies for intelligence, the first of which is game-playing behavior.

One of the early challenges in AI—and for the moment I'm talking about AI in the large, not soft or weak or any other marketing buzzword—was to get a computer to play chess. Now, why would a bunch of computer scientists want to get a computer to play chess? Chess, like any other game, has a set of rules, and rules can be written in code. Chess is more complicated than many games, such as tic-tac-toe (a game that is used to demonstrate to another fictional computer in the 1983 film WarGames that nuclear war is unwinnable), and has a large enough set of potential moves that it is interesting from the standpoint of programming a winning set of moves or a strategy. When computer programs were first matched against human players in the late 1960s, the machines that were used were, by any modern concept, primitive and incapable of storing a large number of moves or strategies. It wasn't until 1996 that a computer, the specially built Deep Blue, beat a human Grandmaster at the game.

Since that time, hardware has continued its inexorable march toward larger memories, higher clock speeds, and now, more cores. It is now possible for a handheld computer, such as a cell phone, to beat a chess Grandmaster. We have had nearly 50 years of human/computer competition in the game of chess, but does this mean that any of those computers are intelligent? No, it does not—for two reasons. The first is that chess is not a test of intelligence; it is the test of a particular skill—the skill of playing chess. If I could beat a Grandmaster at chess and yet not be able to hand you the salt at the table when asked, would I be intelligent? The second reason is that thinking chess was a test of intelligence was based on a false cultural premise that brilliant chess players were brilliant minds, more gifted than those around them. Yes, many intelligent people excel at chess, but chess, or any other single skill, does not denote intelligence.

Shifting to our modern concepts of soft and hard AI—or weak and strong, or narrow and general, etc.—we are now simply reaping the benefits of 50 years of advancements in electronics, along with a small set of improvements in applying statistics to very large data sets. In fact, improvement in the tools that people think are AI is, in no small part, a result of the vast amount of data that it is now possible to store.

Papers on AI topics in the 1980s often postulated what "might be possible" once megabytes of storage were commonly available. The narrow AI systems we interact with today, such as Siri and other voice-recognition systems, are not intelligent—they cannot pass the salt—but they can pick out features in human voices and then use a search system, also based on stats run on large data sets, to somewhat simulate what happens when we ask another person a question. "Hey, what's that song that's playing?" Recognizing the words is done by running a lot of stats on acoustic models, and then running another algorithm to throw away the superfluous words ("Hey," "that," "that's") to get "What song playing?" This is not intelligence, but, as Arthur C. Clarke famously quipped, "Any sufficiently advanced science is indistinguishable from magic."

All of which is to say that KV is not surprised in the least that when you peek under the hood of "Soft AI," you find a system of statistics run on large data sets. Intelligence, artificial or otherwise, remains firmly in the domain of philosophers and, perhaps, psychologists. As computer scientists, we may have pretensions about the nature of intelligence, but any astute observer can see that there is a lot more work to do before we can have a robot pass us the salt, or tell us why we might or might not want to put it on our slugs before eating them for breakfast.

Kode Vicious, known to mere mortals as George V. Neville-Neil, works on networking and operating-system code for fun and profit. He also teaches courses on various subjects related to programming. His areas of interest are code spelunking, operating systems, and rewriting your bad code (OK, maybe not that last one). He earned his bachelor's degree in computer science at Northeastern University in Boston, Massachusetts, and is a member of ACM, the Usenix Association, and IEEE. Neville-Neil is the co-author with Marshall Kirk McKusick and Robert N. M. Watson of The Design and Implementation of the FreeBSD Operating System (second edition). He is an avid bicyclist and traveler who currently lives in New York City.

A Conversation with Arthur Whitney
A few well-chosen words about programming languages from a long-time designer
http://queue.acm.org/detail.cfm?id=1531242

Information Extraction
- Andrew McCallum
Distilling structured data from unstructured text
http://queue.acm.org/detail.cfm?id=1105679

The Network Protocol Battle
- Kode Vicious
A tale of hubris and zealotry
http://queue.acm.org/detail.cfm?id=2090149

Originally published in Queue vol. 14, no. 6—
Comment on this article in the ACM Digital Library

More related articles:

Divyansh Kaushik, Zachary C. Lipton, Alex John London - Resolving the Human-subjects Status of Machine Learning's Crowdworkers
In recent years, machine learning (ML) has relied heavily on crowdworkers both for building datasets and for addressing research questions requiring human interaction or judgment. The diversity of both the tasks performed and the uses of the resulting data render it difficult to determine when crowdworkers are best thought of as workers versus human subjects. These difficulties are compounded by conflicting policies, with some institutions and researchers regarding all ML crowdworkers as human subjects and others holding that they rarely constitute human subjects. Notably few ML papers involving crowdwork mention IRB oversight, raising the prospect of non-compliance with ethical and regulatory requirements.

Harsh Deokuliar, Raghvinder S. Sangwan, Youakim Badr, Satish M. Srinivasan - Improving Testing of Deep-learning Systems
We used differential testing to generate test data to improve diversity of data points in the test dataset and then used mutation testing to check the quality of the test data in terms of diversity. Combining differential and mutation testing in this fashion improves mutation score, a test data quality metric, indicating overall improvement in testing effectiveness and quality of the test data when testing deep learning systems.

Alvaro Videla - Echoes of Intelligence
We are now in the presence of a new medium disguised as good old text, but that text has been generated by an LLM, without authorial intention—an aspect that, if known beforehand, completely changes the expectations and response a human should have from a piece of text. Should our interpretation capabilities be engaged? If yes, under what conditions? The rules of the language game should be spelled out; they should not be passed over in silence.

Edlyn V. Levine - Cargo Cult AI
Evidence abounds that the human brain does not innately think scientifically; however, it can be taught to do so. The same species that forms cargo cults around widespread and unfounded beliefs in UFOs, ESP, and anything read on social media also produces scientific luminaries such as Sagan and Feynman. Today's cutting-edge LLMs are also not innately scientific. But unlike the human brain, there is good reason to believe they never will be unless new algorithmic paradigms are developed.

Kode Vicious - @kode_vicious

The Chess Player who Couldn't Pass the Salt

AI: Soft and hard, weak and strong, narrow and general

Related Articles