AI Limits, Dangers and Threats E-Book

Philippe AGRIPNIDIS

• Herausgeber: WS
• Kategorie: Wissenschaft und neue Technologien
• Serie: Knowledge
• Sprache: Französisch
• Veröffentlichungsjahr: 2023

Seitenzahl: 145

Beschreibung

If you want to understand, without any technical jargon, what the limits, dangers, threats and conceptual flaws of artificial intelligence are, then this book is fully suited to your desire for knowledge and understanding.

You'll discover that AI, which wants to present itself as an exceptional computing advance, is in fact more than 10 years behind the technologies already available. And what's more, AI designers don't know how to explain the whys and wherefores of the information their software provides when used. They don't know what goes on in the computer mill of their own codes. AI is like a parrot armed with a photocopier enclosed in a watertight black box. It only reproduces what it has seen or heard before being captured. It's not a question of providing the relevant or the appropriate, or even the true or the certain, but the credible, the probable, the possible.

You will also learn that, although presented as a marvel of automatism, AI only functions through perpetual human corrections. And that, despite this, there have been many instances of behavioral slippage. And that more will follow. For it is the very nature of this so-called artificial intelligence to be subject to its own contradictions and errors. You'll also see how AIs become bipolar, oscillating between phases of information bulimia and purging of superfluous data.

You'll also get an insight into why, despite its limitations, AI fascinates us. And why we use it. And the first economic, social, psychological and cognitive backlashes that are beginning to appear.

The book also presents the fundamental threats inherent in AI. They are intrinsic to the approach that presided over its conception. If we can calculate everything, on every subject and all the time, then let's calculate everything. But without going into the depths, without any real knowledge of the World. By keeping the approach of flat thinking, to a single time of reflection. Because AI's greatest weakness is that it is built on a limiting and confusing tool: text. To create new Knowledge, AI is like a propeller engine that wants to go to the Moon. As soon as it reaches a certain threshold, it is no longer efficient. There's a glass ceiling that AI will never be able to break through to bring us the necessary evolution or progress that Humanity needs at the start of the 21st century.

It should be noted that this book is the third part of another: "Prelude to Quantum Graphs", with a few additions, updates and simplifications. For the original part was written in deep and recurrent connection with the notion of Quantum Graphs, which goes beyond the simple problem of AI. It was timely to present these insights without this entanglement with Quantum Graphs, which are the antidote and alternative to AI through the universal construction of standardized, structured and articulated knowledge models.

Leseprobe

Philippe AGRIPNIDIS

AI LIMITS, DANGERS and THREATS

A TOOL WITHOUT MASTERY

Collection

Knowledge

ISBN 978-2-487087-09-5

SPHARIS Publishing

83 a rue des Alliés 42100 Saint-Étienne France

Copyright © – November 2023 – AGRIPNIDIS PhilippeCurrent edition 1.0

All rights of reproduction, by any process whatsoever,

adaptation or translation, reserved for all countries.

Digital edition or print on demand.

From the same author:

The Universal Inventions. Paper: ISBN 978-2-918651-15-4

Digital: ISBN 978-2-918651-14-7

www.spharis.comhttps://www.youtube.com/@spharishttps://twitter.com/Sphariscom

Cover illustration

. License Elements ENVATO© by MEFTAHYs-PROTOTYPE

FOREWORD

Global intellectual property rights

Under article L.122-5, 2° and 3° a) of the French Intellectual Property Code, only “copies or reproductions strictly reserved for the private use of the copier and not intended for collective use” and only analyses and short quotations for the purposes of examples and illustration are authorized, and “any representation or reproduction, in whole or in part, made without the consent of the author or his successors in title or assigns is unlawful” (art. L. 122-4). Such representation or reproduction, by any means whatsoever, would therefore constitute an infringement punishable under articles L 335-2 et seq. of the French Intellectual Property Code.

Drawings and illustrations

All the authors of drawings, photos and illustrations, insofar as they are known or informed, are cited in the book. And images, in the broadest sense of the term, are integrated into the book in compliance with their licenses for use. In the event of an error, please contact the publisher so that the information can be rectified for a future print run.

No smileys but ironic dots ¡

Rather than using the punctuation marks used on digital media, such as :-) and ;-), we’ve opted for the inverted exclamation mark ¡, which is an aesthetically and typographically pleasing replacement for these two encodings. One ¡ is equivalent to :-) and three ¡ to ;-). Thus ¡¡¡.

Use of Wikipedia

Wikipedia’s three main strengths are its ambition, its enthusiastic volunteers and the use of permanent links to keep one topic and one topic only at the same Internet address. This is true even if the content of the topic changes in one direction or another. For this reason, links to this resource will be preferred.

STAKES AND FUNCTIONS

Artificial intelligence stakes and contexts

A disproportionate and growing importance

Hardly a day, or even an hour, goes by without “news” about Artificial Intelligence (AI) appearing in the newspapers or on our news feeds. Adorned with all manner of virtues, it seems to be a magic formula, a digital philosopher’s stone capable of curing all of humanity’s ills. Its invocation could lift mountains and move oceans. Effortlessly. And without a backlash. All our worries, all our problems will be solved, by a few clever calculations from thousands of interlinked processors that will scientifically and truthfully—how could it be otherwise—bring us the best solutions at the lowest cost. This is indeed the promise of AI.

But where there’s promise, there’s hope. But not guarantees and certainty. This is one of the first ambiguities of AI. It is not the least. But the most serious problems stem from the dispossession of human beings of their ability, duty and obligation to think for themselves.

Fortunately, if the danger is real, there is at least one alternative to mitigate these potentially catastrophic drifts. It’s based on the notion of Knowledge Models and Quantum Graphs. The latter are presented in the book (in French) “Prélude aux Graphes Quantiques”, available on digital distribution platforms and in hard copy on Amazon. The information you are about to read about artificial intelligence in the present work is taken from this first book, which includes, in addition to a development on the antidote and alternative to AI, a presentation of the Quantum Graphs method. The formalization of standardized, structured and articulated Information that it enables leads to the generation of Knowledge Models. It’s this method and convention that we’ll be referring to when we speak of Knowledge Models.

The chapters to come will be devoted to describing how AI works, presenting its dangers, its limits, its threats. And its cardinal flaw, the use of text. Which means that, instead of being the rocket engine it would like to be, it’s just a beautiful propeller engine that won’t take us to the heights of human civilization.

So, let’s start with a few basics about AI, before looking at how it works.

A few basics

This is not intended to be a technical course on artificial intelligence. There are many relevant books on the subject, as well as free videos and training courses on the Internet.

Because more than just the appearance of computers, we’re going to develop the very principle of artificial intelligence. What it is essentially based on. Its vision of the world. Or at least those of its designers. And we’ll see what this implies in terms of limitations and illusions.

About time! From black-and-white into the land of color

As a stumbling block to your thinking about what AI is, what it really brings, in terms of opportunities or threats, you need to integrate the following information. What artificial intelligence will achieve by the end of 2023 is absolutely normal. There are no spectacular achievements to celebrate. There’s nothing to be astonished about. The only reason to marvel would be that this is happening so late in the history of computing! After all, everything had been ready since 2010: server farms, powerful processors, high-capacity hard disks, abundant memory, algorithms and, above all, the Internet mesh to “suck in” data and disseminate it, so that AI in 2023 mode would already exist. It’s just, finally, the computer’s ability to create content that’s being used. For text and images. There’s nothing magical or surprising about it. It would be like being surprised to see black-and-white images on a color television. Because IT can generate much more than AI.

It’s the delay in getting to this level that should make us wonder. Given that all this is based on old methods and forms of data storage and statistical exploitation that are in fact obsolete for creating new Knowledge and sharing it. It’s a logical outcome. But of an older generation. It’s a limitation and an end. We won’t go any further collectively with this tool. So, there’s no adoration to be had for current AI. That’s why this book has been written. To raise awareness—to inform—to alert. And to help spread a fair vision of AI’s real capabilities, but also of its limits and the pitfalls it may contain. And these last two categories are well fulfilled.

A caricatured definition?

If we wanted to express, without any technical concepts or jargon, what generative artificial intelligence viscerally is, we could say that it is: “a parrot in a black box equipped with a photocopier that regularly jams paper”. You’ll see that this definition goes much deeper than it might at first appear, in terms of what artificial intelligence intrinsically is. For as long as artificial intelligence is organized the way it is in 2023, it will only repeat and distort what it has collected. And we’ll see in the next section that the designers don’t really know why or how it works. We don’t know, in essence, why there is this result.

The parrot doesn’t know what it’s saying or why it’s saying it. The proof is that if you ask him the same question again, you’ll never get the same answer. What is presented as a strength by the lauders of artificial intelligence is actually a demonstration of its weakness. It is absolutely unable of explaining its choices. And to demonstrate them. It’s like this. You’ll never know why. But don’t worry, neither will he…

What’s more, it’s a parrot that requires you to speak just one foreign language. Since the pioneers in the commercialization of artificial intelligence, OpenAI, were American, the development of the tools was mainly based on this language. Even if, supposedly, AI can also work in other languages such as French, Spanish or Italian. But you don’t get the same efficiency or quantity of output as with the AI’s native language, English. If you’re English or American, you might not mind if the AI speaks Anglo-American. But in fact, it doesn’t really speak that language. And that’s the problem. You have to master another language, the obscure one of the “prompt”! We’ll go into more detail later, but the “prompt” is a structuring of commands, of orders that are sent to the artificial intelligence. It’s like telling a dog what to do. But you have to make yourself understood. We’ll come back to this point in the section on the limits of artificial intelligence.

After this initial, non-technical approach, which corresponds well to what artificial intelligence is conceptually speaking, we’re going to move on to a presentation of its technical functioning.

How Artificial Intelligence works

Pulsions and history

Without going into too much detail, there have been several waves of development over the decades. We could trace this idea back to the invention in 1834 by Charles BABBAGEof the first computer concept and the realization of a “machine” capable of sequentially reading instructions from Jacquard card perforations. Who’s to say that in his wildest dreams, he hadn’t envisaged and hoped that a machine could itself produce instructions for others? But at present, we have no indication that this idea ever occurred to him. But given his visionary genius, and the fact that his machine could actually be built and operated, even with the means available at the time, it’s not impossible that he thought of it.

A schematic representation of the Turing testIllustration no. 1. Wikipedia. @Bilby

In any case, apart from the technical concepts themselves, the first person to suggest that artificial intelligence might exist was Alan TURING, inventor of the eponymous test. Published in 1950, the idea was to express the fact that, one day or other, it would be impossible for a human being to distinguish, in a remote communication, by screen or intermediaries, whether he was talking to another human being. Or a machine.

Until the arrival of ChatGpt 4, the test had not been passed. Now, it can sometimes be passed. Which is not to say that the machine can’t be “uncovered” by anyone, on any subject. But the reasons for doubt are and will become stronger. For the concept evoked by TURING was the equivalent of opening a technological and modern Pandora’s box. Mentioning the concept of artificial intelligence through a computer could only invoke a pressing need to achieve it. The race was on. We knew where it would end. We just didn’t know how long it would take to cross the finish line. And on several occasions, we mistakenly thought we were just a few meters away.

And above all, we didn’t know who would win. If we have to name a winner at the end of 2023, it’s Open AI.

At least in this first race.

But what is artificial intelligence? The puzzle lover!

It all depends on where you stand. If it’s marketing, then, as we saw at the start of this book, in the long term, it’s true intelligence, the panacea and remedy for all the ills that can exist in Humanity and on the Planet. Everything will be solved and saved by artificial intelligence. We won’t know why or how, given the way it works today, but according to its promoters, everything will be fine. And nothing should happen without artificial intelligence.

Realistically speaking, the ambition of artificial intelligence is to generate ideas, proposals, action plans, recommendations, summaries and content on its own, without human intervention, once it has been coded and trained. Hence the use, from 2022 onwards, by certain students of tools of this type to have them write assignments or documents (dissertations, parts of theses) on their behalf. The complementary idea is that this production should be fast and of at least human quality. Or even better. And to achieve these general, but not necessarily generous, objectives (profits for shareholders are expected), technological choices have been made.

Above all, we need to understand how AI works today. The “creations” generated by AI are like a final jigsaw puzzle made from multiple pieces of other jigsaw puzzles that are completely different from each other in terms of subjects, illustrations, dimensions and cut-outs! And all from different stores and manufacturers.

But I’m telling you, we’ll make it! We’ll make it.Illustration #2. License @vmiregolda AY images.

The AI will look here and there in its data capture (we’ll come back to this when we talk about intellectual property) for pieces of data that it will patch up to make the whole thing stand up and look like something. We’re going to plane what doesn’t fit. Stretch out what’s too short. Flatten what’s too thick. Inflate what’s too thin. Deform to fit. And above all, gluing, taping and stapling pieces from all sides and horizons so that from a distance they look like something. What appears to be a single piece is actually made of bits and pieces. Worst of all, the designers don’t know why.

We’ll see exactly why in the next chapter.

But then, what is Intelligence?

Because if we can’t define or agree on a common notion of what human intelligence is or isn’t, we’re not going to get off to a very good start when it comes to creating artificial intelligence ¡ That’s what we might logically and healthily expect from this approach. But this is not the case. It’s important to understand that the word Intelligence, associated with the concept of the artificial, is, for the moment, nothing more than marketing. Because it’s not about producing intelligence, it’s about producing content. Credible, possible. But not intelligence. In fact, we believe that there are several types of intelligence. There’s purely analytical thinking intelligence, but also emotional intelligence. And others, such as methodological and environmental intelligences (10 types to be discovered). Not to mention that there are also variants of the so-called emotional intelligence.

Knowledge Models explain why we have different types of intelligence among human beings. They are maps of different roads (relationships) and crossroads (elements). One mathematically “intelligent” person has such and such roads and crossroads in his head. And another, more gifted for emotional intelligence, will have such and such a configuration.

Whatever the case, in all possible forms, we always find the ability to act, in the broadest sense. This means choosing, estimating, deciding, intervening or taking action, without the situation having already been known or experienced. It’s the power to produce something relevant and original, fully adapted to contexts, actors and resources. By knowing exactly why or on the basis of what assumptions, or bets, we structure our interventions. It means using Knowledge and Information to articulate the thoughts that will structure your future actions. Intelligence is creation. But it’s not dumb copying. It’s not copy and paste, tracing paper. It’s not following a recipe book, and not understanding it. Or not knowing what the words really mean. Intelligence is the ability to produce the timely and appropriate, even without all possible information. Compared with Quantum, which brings Knowledge, intelligence is the ability to generate the appropriate while NOT having all the Relations and Elements at one’s disposal.

With so-called artificial intelligence, as you’ll see in the next few paragraphs, we’re a long way from that. Much further. Too far.

The structuring of generative artificial intelligence

Still without using computer jargon, we’re going to summarize what artificial intelligence is. As far as the technical aspect is concerned, we’re going to formalize developments in three waves. The first, in the 1960s, was to introduce knowledge directly from lines of code. This was an instruction-by-instruction approach. I do this, then I do that. Computer historians and technicians agree that the results were not brilliant.

This led, after a long pause, to the creation of a new approach based on the notion of “rules”. Very much in vogue in the 1980s and 1990s, the idea was to create independent nodules expressed in the form of rules that are capable of expressing themselves or not activating according to conditionalities. Do this as long as it’s not at such and such a level, unless there’s such-and-such an event, then give priority to rule number 17. The limits of this method were soon reached, and the regrets were as great as the disappointments.

Until, in the 2010s, the notion of data analysis was introduced. First with machine learning. Then by the so-called deep learning method. In our view, the difference between the two is purely and solely marketing and commercial. Deep learning is machine learning. But with more data. However, it’s still more chic to mention the word “deep” than “hollow”…

In any case, deep learning is emerging precisely to be able to better dissect the ever-increasing quantity of Petabytes of information collected. And in a way, the pupil has surpassed the master, or Frankenstein’s creature has escaped the control of its progenitor. Deep learning has become a tool in its own right. It now far surpasses Data Analysis in popularity and usage. Which is the second stage of data mining.