Artificial intelligence - a question will be asked for each answer

Artificial intelligence in speech expert systems is developing in the direction of chat bots. Undoubtedly, this is a direction of development that everyone understands. Of course, there are many problems with its formation, this can be seen from the presentations of the developers . They talk well about their achievements, about the problems that they managed to solve, but I would like to see the problems that they could not solve.

Let's look at all these problems with information theory. Information - information perceived by a person and (or) special devices as a reflection of the facts of the material or spiritual world in the communication process (GOST 7.0-99).

The main functions when working with information:

  1. Search for new information;
  2. Saving information;
  3. Information transfer.

According to the first function. It can be divided into several levels. The local level where, in principle, tasks are performed quite effectively. An example would be a user’s question: “Where is the nearest gas station?”, The systems give an answer to it quite reliably. Here, in my opinion, the term “local level of search for new information” should be defined as the ability to obtain new information by a user from a source that exists in competition with other sources that are not related to AI. The presented example is good in that it is naturally clear that the user can receive information about the nearest gas station from others, maps, that is, from other sources. And the choice of source is determined only by the costs of the user.

Tasks that cannot be solved by the user on the basis of local information can be attributed to the external (upper) level, that is, to a level that has a dimension higher than that available to the user. An example is the task of constructing the optimal route of movement from A to B, taking into account the external situation (traffic jams, etc.) and user performance criteria (time spent, etc.). We see that the solution to these problems, now, is being implemented satisfactorily.
Tasks that require going beyond the knowledge of society, we designate this level - global, and ignore it. Since, on one side of these problems is forecasting, and on the other hand, the class of these problems, there is Gödel’s incompleteness theorem.

By the function of saving information. I assume that these systems do this at a level that is currently achievable.

The third function is the transmission of information. At the moment, theoretically, any system can transmit the necessary information in various forms. We assume that the level of development for this aspect is satisfactory.

Let us see from the point of view expressed here on the development of such expert systems. Let's look at the principles used by the developer : “Fortunately, we have the full power of Yandex search engine, which every day faces millions of queries, searches for millions of answers and learns to understand which ones are good and which are not.” That is, in fact, the physical solution search scheme is reduced to a simple scheme:

Запрос пользователя -> поиск состоятельного ответа в базе данных -> вывод ответа

As you can see, the answer seems to be one of the recorded, that is, existing results, either transformed and / or agreed into a single result from several recorded results. Or from editorial databases : “We want the personal assistant to be an integral personality with a well-defined set of qualities. Here our editorial texts come to the rescue. Their peculiarity is that they were originally written on behalf of the personality that we want to recreate in Alice. It turns out that you can continue to train Alice on millions of lines of random texts, but she will answer with an eye to the standard of behavior laid down in the editorial answers. And this is what we are already working on. ”

In fact, it turns out that the chat bot is an intermediary between the user and an indefinite number of persons (legal wording). And I think that with this view of these systems, you can structure tasks and problems, detail them and define new goals.
What is a conversation. This is the process of transferring information from one subject to another in one session. It is used in various circumstances. With equal subjects, to coordinate actions, with unequal subjects, to determine the actions of a subordinate subject.

Then the types of conversation tasks are formed from their type.

By the number of participants:

  1. Monologue;
  2. Dialog;
  3. Collective conversation (conference);
  4. Communication with an indefinite circle of people.

With this clustering, taking into account the attitude to the “local level of searching for new information”, it can be seen that the search for tasks for AI is carried out mainly when using AI for the fourth type of tasks “Communication with an indefinite circle of people”.

Let's look at the other types.

Monologue - here we will consider as a monologue an attempt by the subject to systematize the results of the analysis, fixing the results. Since according to M. Twain: “Any mention in the press, even the most negative, except for the obituary, is an advertisement.” What AI can help a person if he turned to him with a monologue. The class of answers is quite obvious - to look for similar monologues from your database. Immediately, an aspect arises that, in essence, and, any scientific study, before publication is a monologue of one or more authors. Whether this task is feasible for AI, to find similar research today. The answer is yes.

Dialog. Let's take an example from social networks, based on the famous song:
- Do you love me?
- Yeah.
- Will you be with me?
- yeah
- So we will be together, so we will be there, always with you!

We analyze this example. Can the AI ​​help the user at a local level? I suppose that it is insignificant, since at the local level the interlocutor, for those who conduct dialogue, has priority weight. But, from the point of view of tasks of the upper level, there is already the help of AI, and therefore there are tasks for it. The first is to determine if the subject is talking to the bot. The second is the validity of the answers. And further, a huge number of tasks: trolling, degree of interest, hidden goals, etc.

Increasing the importance of dialogue. You can see how to use the power of AI in various fields:

A) hiring;
B) negotiation;
C) Litigation.
And the list of tasks is huge.

Imagine a clear model for the example of hiring.

Party A (employer). Job submission. The AI ​​analyzes the submitted resumes and presents a list of suitable candidates. Eichar analyzed and refused everything, AI gives advice on what is necessary to change, supplement, etc. If he (HR) was unable to find an employee, then let him think about his compliance and about the proportionality of the requirements and rewards. (This is a slap from me on all HR).

Side B (applicant). Writing a resume. Compilation of AI, tips, ratings, comparative results. Well, the applicant has the opportunity to critically look at their capabilities.

Conference. For this type of communication, compiling an analogy for dialogue, one can also see many tasks in which AI can be useful. From current tips to signaling the level of involvement in the conversation topic of all its participants. The impact of each participant on the outcome of the conference. Search and connect external specialists in the current mode. For unequal entities, these are instructions for using the equipment. And much more.

Communication with an indefinite circle of people. In this type of communication, examples of the use of AI are already a huge number, from the creation of advertising texts to the creation of speech at the UN General Assembly.
Here, an attempt to systematize tasks for AI is presented to the public. The idea of ​​a criterion for comparing various AIs is also presented. Basically, this material was welded, according to the results of Alice’s skills tests , which various developers tried to implement.

Also popular now: