What's WRONG with using artificial intelligence (AI) for commercial intelligence and crisis management purposes?

         The spread of artificial intelligence (AI) is becoming more and more widespread in recent times. Go to Linkedin or any other specialized group or social network, you will see many publications offering and advertising AI as an indispensable assistant in business. AI is capable of replacing a lot of inefficient people in their workplaces. Many programs and courses are offered to better master and learn AI.

However, is AI suitable for all tasks? Can it fully replace humans, especially in high-risk areas?

When you're hiring new staff, don't you pre-screen them at the interview? Why not dig a little deeper into AI? Find out more about it? Is it worth the risk?

If you have ever read “combat documents”, for example, intelligence reports on some problem, now by structure of construction, I will reproduce a similar product. How much to trust it is for you to judge! It is better to double-check everything yourself, especially if the topic of research is close and in demand by you. In the future, I will explain why the document is composed in this way! This also directly relates to the very essence of our narrative.

So:

Why is it that (even) entry level analysts, will do the job much better and more efficiently than artificial intelligence (generative network)?

It should be noted that entry-level analysts should work under the guidance of an experienced mentor.

What is the purpose of any analytical study? On the one hand, it is an in-depth investigation of the subject of study. On the other hand, it is to obtain new knowledge about the subject, its activity, parameters and other accompanying factors. I emphasize - we are talking about really new knowledge! Such knowledge that did not exist before! This is the value of an analyst as a specialist and the value of analytical work in general. In the view of an ordinary person, this knowledge may concern some trifles that have no special significance. But the fact remains that this is new knowledge about the object of research that did not exist before.

Who are entry-level analysts? Entry-level analysts are specialists who have some knowledge of the research topic, either by virtue of their position or by having received specialized training in basic analytical skills. An entry-level analyst may not know the subject or scope of the research in detail. However, he or she is familiar with the basics of analytical information processing and knows how to write/describe the issues under study:

·         exactly how;

·         in what order and sequence,

all of these things matter when creating/extracting new knowledge.

One of the oldest visual-graphic analyses of information still used by the FBI today, the “linkage matrix,” even involves the use of a special alphabetical listing of the information and data in question. This is necessary so that, upon further verification, it is highly desirable that the order of the information considered be the same for both the analyst and the reviewer. Otherwise, the matrix will undergo visual changes. NO. The nature of the relationships, if no error is made, will be correct! But the appearance of the matrix will most likely be completely different. And the task of an analyst is not just to make a matrix, any computer can do it. But! To extract new knowledge, new understanding/clarification of the situation and to make, at least, a forecast of the most probable development of the situation for the future.

Since this is a type of qualitative analysis, even presenting the conclusions in mathematical form is hardly possible.

Why does an entry-level analyst need guidance from a more experienced manager? Despite their knowledge, the analyst's view of the problem as a whole may be very general and not deep. Whether it is a specific risk or market or market area. An experienced supervisor is able to tell the analyst what to look out for, where to look, sometimes, what to look for, simply based on a more advanced knowledge of the research area and the characteristics of the research object. The supervisor will also see the gaps made by the analyst and bring unexplored aspects to the analyst's attention. Finally, an experienced supervisor will help in drawing conclusions and generating the very new knowledge that is the ultimate goal of the research.

Each such job, by the way, increases the skill level of the analyst himself, making him a more advanced expert.

Now let's look at exactly how artificial intelligence (AI) works when creating its projects. Is it fully capable of replacing an intelligence analyst? I've had to review a lot of material on this to be able to speak on the topic in a more substantive way. However - do NOT take my word for it! Double-check for yourself. I may be wrong or misguided.

1. Logic

Large Language Model (LLM) aka neural or generative networks. Let us first address the LOGIC of composing or generating text documents. This is an important consideration indeed! If logic is used that is unclear or not applicable for the given conditions, it can very much affect or distort the document being generated and/or the expected results.

So, Large Language Model (LLM), which uses machine learning algorithms based on large text data sets, starts generating/composing a document. The logic or principle of LLM is based on determining the probability of maximum combination of words (their meanings). That is, each subsequent word should maximally match the previous one, including in terms of meaning!

The LLM uses so-called “textual clues” to complete a coherent narrative and ultimately answer the question posed. “Hints” are also given algorithmically by the LLM (if the requester knows how to do this at all). The question is then solved by means of “statistical inference”. If there are statistically more positive answers than negative answers, then the LLM believes it has solved the problem correctly. If there are more negative answers than positive answers, then the problem is not considered solved!

Yes! To solve “text hints”, you will need additional machine learning + so called “hint engineering”. Possibilities of additional customization and LLM training, i.e. specially trained team of experts.

I don't know if I have managed to explain the LOGIC of LLM in an accessible and understandable way, but try, for the sake of practice and understanding do a little practical exercise.

Let's keep the text small, literally two or three sentences. You don't need more than that, you are not competing with a computer that performs several million operations per second. Our brain is much more modest in its capabilities.

So, try to compose two or three coherent sentences, making sure that each word corresponds to the previous word in meaning and content as much as possible! If you succeed, then, using a logical “clue” bring the resulting text to a meaningful conclusion! How? Did you get it?

What the LLM is more or less successfully able to do today:

1. Natural Language Text Generation: LLMs, using the power of computational linguistics, can create/generate: writing articles and texts; dialog conversations with an interlocutor; song generation and composition.
2. Machine Translation: Translate text between any pair of languages. LLM greatly simplifies and facilitates the translation of source text into the desired language.
3. Generate original content: The content you create can be used to create blog posts, write articles, and other types of content.
4. Sentiment analysis: A new trend in LLM work is sentiment analysis. Based on special software, LLM is trained to recognize and classify emotional states and sentiments present in the annotated text. The program is able, on the one hand, to identify emotions such as positivity, negativity, neutrality and other complex sentiments. On the other hand, the program can incorporate the components of these emotions into the generated text (an indispensable option for programming, like NLP, or when creating propaganda and influence materials).
5. Understanding, summarizing and classifying text: By training an LLM model to understand and analyze huge amounts of data, LLMs allow AI models to understand, summarize and even classify text in various forms and patterns.
6. Dialogues, questions and answers: LLMs provide real-world capabilities to accurately perceive and respond to natural language interlocutor queries. In doing so, LLMs learn the context of the query and sift through an extensive collection of texts to provide appropriate answers to the interlocutor's questions.

Which of the following is not available to a (novice) intelligence analyst? Probably point 2 and point 4. Although, an intelligence analyst (specially selected for the job) who does not know the language of the enemy is like a driver who does not know how to drive a car. As for point 4, on the one hand, such tasks are not on the agenda, but on the other hand, don't tell or show this to a specialist in psychological operations.

However, we did not stop at the main thing! Obtaining NEW knowledge about the object of research (intelligence development)! That which did not exist before.

For this skill is responsible, closely related to logic, but a different process!

2. Creativity

The main task of information and analytical work is to search for/produce new knowledge! This process is impossible WITHOUT creativity!

Creativity is a process that creates qualitatively new materials, spiritual values or knowledge. The main criterion that distinguishes creativity from manufacturing (production) is the uniqueness of its result. Two people, equal in terms of training and education, observing the same phenomenon, for example, a “matrix of connections”, are likely to make different conclusions and forecasts of the situation development. Especially if they have different depth of knowledge and immersion in the environment.

The result of creativity cannot be directly deduced from the initial conditions. No one, except perhaps the author or authors, can get exactly the same result by creating the same initial situation for it. In a team of analysts, all analysts must, in the process of searching for arguments, arrive at a more or less conscious and coherent result, but be prepared for unpredictable changes.

In the process of creativity, the analyst invests in the material (knowledge available on the problem), in addition to labor, some possibilities irreducible to labor operations or logical conclusion, for example, intuition and knowledge of previous events and patterns of behavior (activity) of the object of study. It is this fact that gives the products of creativity an additional value, unlike the products of production, which always provide the same (the same) results. In creativity not only the result itself is valuable, but also the process of making/producing decisions and forecasts. Creativity is always going beyond the limits of existing knowledge, algorithms, programs.

The development/generation of new knowledge and innovative ideas about the object under study and their practical verification and achievement of the intended goals is the end result of information and analytical work.

Is LLM (AI) capable of such a thing? As noted above, creativity is always going beyond the boundaries and limits of existing knowledge, algorithms, programs. Will LLM write a new suitable program to solve a new problem? I have very big doubts about this. At least for today and at the current stage of AI development.

Today, there are already a lot of cases when students have generated term papers and entire thesis projects with the help of AI. To be honest, I personally do not know (I admit that there are unknown cases) of situations when such papers and projects received excellent grades as a result. Three or four points. However, it is said that students were often rash at the defense. On the other hand, such grades are understandable. LLM operates only known and available information. Therefore, in fact, nothing new, no original ideas are offered in the results.

Can a (even a novice) analyst, under the guidance of a more experienced manager, develop new knowledge? As it seems to me, to the fullest extent!

Lastly, on the structure of the presentation of the material. Remember I said it matters too!

As we already know, the LLM will first compose the text based on algorithmic preferences and try to make the next word match the previous one as much as possible. Then, with the help of “text hints”, the text will be brought to the final result. Will this process produce “new knowledge” about the research problem? Definitely not! LLM algorithms and programs operate only with the information known and available to it.

What happens when an analyst, with available analytical knowledge and research apparatus, takes over?

• First, from the very beginning of his work, the analyst keeps in his mind the ultimate goal of the research. He looks for arguments based on that purpose. The order of words in which words match each other as much as possible is of little, if any, interest to him;
• Second, it does not need any “textual prompts” to complete the work. All the description and research immediately takes place within the framework of the declared (requested) research;
• Third, using creativity, creative thinking and art, the analyst is able to derive/generate new knowledge that did not exist before. Yes! The research will most likely require guidance and direction from a more experienced supervisor;
• Fourth, the sample intelligence report presented is NOT something random and unintentional. This submission of materials structures the analyst's thought process and defines the subject area of knowledge - where to look for something new.

3. General conclusions

What general conclusions can be drawn from our rather superficial study, which is not intended for professional programmers?

1. Where the LLM will fail is in generating new knowledge, even entry-level analysts will be more successful in finding and obtaining effective solutions based on creative and quality solutions.
2. Should AI be used to solve commercial intelligence and crisis management problems?

Surprisingly it makes sense, with some limitations, to use AI to solve commercial intelligence and crisis management problems in at least two ways:

• First, in terms of checking “to see if we've missed anything”. LLM uses only available (and in some places verified) information. It is able to show the diversity of the problem that has already been: tracked; evaluated; measured and described, by someone else and before us;
• Secondly, for training. Yes! You can train your own analysts in the basics of analytical work and information processing. Especially if your own analysts are untrained and have little practical experience.

In each case, however, consider: what kind of neural network is it; who has access to the neural network's information and to possible distortions of that information; can the particular neural network be used safely?

The answers to all these questions can be found in a special section of our manual.

Take your time! Think! Check! Evaluate! Use it wisely!

P.S.

Guess by the way, in the foreseeable future, will a neural network replace a professional analyst capable of extracting new knowledge?