Making a prediction is, in short, playing with uncertainty. To get an idea of how we handle uncertainty in our knowledge, answer these two questions:
- If you toss a coin in the air. Which side will be seen? Heads or tails?
- If for millions of years the sun has hidden and reappeared. If it has been hidden today, will it reappear tomorrow?
Since the beginning of Artificial Intelligence, the idea of an intelligent attitude was always the promise that machines would stand shoulder to shoulder with humans doing intelligent and mechanical tasks. But it was only until the 90s (20th century) that an algorithm that dealt with uncertainty in knowledge was presented: Bayesian Networks.
To delve deeper into this and for the purpose of this article, we will focus on the progress we have made with our Artificial Intelligence Engine (AIM).
Premise: There are no absolute truths.
Knowledge, even if it is scientific, does not have absolute truths in the vast majority of facts. There is always a small margin for doubt, that is, for uncertainty. The scientific method is based on the replication of the phenomenon to provide a thesis of knowledge that is verifiable by other people. However, if the experiment is done a thousand times, it does not mean that the same result will always be obtained (thanks to that 1 in 1000 failure, Edison was able to build the incandescent light bulb).
Our knowledge is moldable, even permeable to new events or variables that add or subtract validity from what we consider true and true. So “A true fact” is nothing more than a “belief” that we simply accept, and we do so because it is necessary for the optimization of our decision process. The faster we make decisions, the less energy we consume and the faster we learn new experiences.
Empiricism: Experience builds belief.
A concept that we have heard colloquially is that “A Lie told a thousand times is a Truth.” Unfortunately this has been a factor that has determined the slow growth of our collective human consciousness. The less we have to analyze, and simply believe thanks to the experience of others, then we build “paradigms”.
But in the case of “personal experience”, all repetitive information (custom) forges a “personal belief”. These personal beliefs are what sometimes generate controversies and discussions among human beings, since the “belief” of a fact is only the compendium of the accumulation of data acquired individually.
A predictive system also falls into this problem, and it is simply due to the type of data entered into the system. If no other data is known, then with the few or many that exist, a result will be obtained… And whether it is expected or not, is completely limited by the amount of experiences (data) that have been provided.
Information fidelity: Identity.
It is very common to have knowledge based on syllogisms that we have been adopting as intellectual beings. But the great variety of premises that can reach the same conclusion becomes complex when we increase the number of variables to be evaluated. This is why a predictive system becomes more important when we have to consider many more variables. For example, a simple deductive system (an expert system or direct logic) can easily conclude that:
- Premise: All men are mortal.
- Premise: Women are not men.
- Conclusion: Women are not mortal.
Which implies an erroneous conclusion, since more data is missing to be able to have an accurate identification of the reality of the situation. So far, only: man, woman, mortal and being have been evaluated. For example, other premises must be included such as:
- Premise: The term man refers to the species homo sappiens.
- Premise: Every woman is a homo sappiens.
- Premise: Every man is a homo sappiens.
- Premise: All homo sappiens are mortal.
- Premise: The term man can include the term woman.
In this way, the broader the number of elements to be evaluated (man, woman, homo sapiens, mortal, refer, be, include), the greater the certainty of the conclusion.
Results: Efficacy vs. Efficiency
The more variables that must be weighted, the slower the efficiency will be, but the more precise it will also be. The amount of data to be analyzed in a predictive system is one of the great challenges faced by all Big Data tools, which is why we must be very pragmatic when selecting the data sources that are going to be correlated in a system like that.
To illustrate the topic, let’s imagine a system in which you have a large number of invoices with customer data, items you have purchased, quantities, prices and their places of purchase. The owner of the information wants to pass all this through a system that helps him “detect future purchasing patterns of a city” in particular (not a user). Perhaps with this information he can estimate future purchasing behaviors and be relatively agile. But in practice it would be necessary to involve other aspects, for example: Conditions of the roads, climatic aspects of the cities, information on the harvest seasons and even the changing value of the daily activities of the city, that is: vacation seasons, work seasons. , fairs, etc.
Therefore, if the information we have allows us to reach conclusions, it is also possible that we do not have all the relevant information to obtain the expected effectiveness, but at least it will allow a notable margin of efficiency.
In the real world: Uses
But, what is all this for? Any person or entity that has a large amount of data (Big Data) is likely to begin to have questions about what type of knowledge they can extract from all that information. That makes it an important asset for future decision-making. If the experience is already in the data, why not make it a useful resource for the future of the organization? That is where a predictive system should be used.
And to be more specific, a predictive system can be implemented in:
- Medicine: To recognize and adjust medical procedures on medications and treatments that have been effective or not. To diagnose diseases with a much broader vision of the patient’s information and their environment.
- Education: Adjusting the pedagogical models of each individual according to their own experience and previous learning, so that the next lesson is truly constructed in such a way that the learner has the greatest guarantee of long-term learning and in accordance with their learning model.
- Marketing: To focus the advertising effort on a young, adult, male or female user, especially if you know more data about the individual, such as their consumption habits or behavioral habits.
- Economy: Trying to determine whether a brand, a cryptocurrency or a stock on the stock market will rise or fall in the future can depend on many variables, not only on the numbers in the history of its behavior, but also on factors external to it, such as: the climate, politics, the profile of society, the emotional perception of the population, among many other variables.
- Sales: Establishing whether a sales opportunity can be successful or not may require an excellent salesperson. But it is also possible to establish it by knowing aspects of the sale as such. Data such as: date of the opportunity, customer, product or service to be sold, customer’s purchase history, customer’s level of curiosity about similar products, product profile and even external characteristics of the sale (weather, enthusiasm on social networks). , influence of a famous figure, etc.).
- Production: Many times the production process depends on sales or the company’s strategy. Even production can have added value with a predictive system if it can define how much and when to produce a product or not, thus optimizing production resources and reducing losses of finished products that will never be sold.
- Maintenance: Each machine, vehicle, building, road, inventory or asset requires maintenance over time. But each one, due to its own nature and the variables of its location and use, generates different conditions that alter the maintenance periods, even if the manufacturer has suggested a standardized period. Therefore, it is more precise, economical and effective to use a predictive system to determine when to perform maintenance on an asset, over and above what the manufacturer has said.
Maybe you have not considered all the scenarios to implement a predictive system from our Solutecia Artificial Intelligence Engine (MIA), but with all these aspects, it is very possible that you have gotten an idea, if this is a solution that can help in your work area. Write to us and we could help you resolve your concerns in your company or need.