Poverty comes in two types: monetary and non-monetary. The amount of money made and spent, among other economic factors, is used to forecast monetary poverty. Based on survey and census data, non-financial poverty is anticipated. Census data is information about the population that is systematically collected. However, as non-financial poverty has many dimensions, finding more pertinent factors is essential.

Artificial intelligence (AI) is a promising technique to increase the recognition of poverty. Compared to traditional models, AI-based models have the following advantages: they can handle many more parameters, are more accurate, can be computed much faster, can handle more data, and require less human work. Additionally, AI outperforms humans in detecting pertinent factors in various data sets, including e-commerce, call detail records, and remote sensing data. Further, this data can be updated more frequently and at a finer scale than other data types. Together, these factors can aid in identifying poverty and facilitating future assistance for the poor. AI can be used to determine poverty in the following ways:

Analyzing various data sets on variables that indicate financial poverty is the first way AI may be used to spot poverty. There are several levels of monetary poverty. On a personal level, one way to determine financial poverty is to look at how people feel about it. AI can identify these emotions by examining materials like interviews. For instance, AI may locate sentences discussing broad deprivations, possibilities, causes, and poverty. To learn what individuals believe about poverty, what it means to them, and how it affects their lives, AI must pre-process these messages. Pre-processed texts consist of the following:

·       Reducing the length of phrases and paragraphs.

·       Substituting words with their root forms.

·       Looking for similar words by examining the grammatical structure of a sentence.

·       Specifying the word’s type, such as a noun or verb.

 Monetary poverty at the household level can be found by looking at the society’s poorest 40% of households by income. From big data collection, AI can determine which household traits characterize that category of households. From a data collection containing more than 100,000 households, AI identified the following pertinent traits:

·       State.

·       Area.

·       Ethnicity.

·       Number of people in a household.

·       Total income.

·       Average monthly income.

·       Income per person.

On the community level, monetary poverty can, for example, be identified by looking at urban poverty. AI can analyze geospatial and survey data to assess population density and the height of buildings. Geospatial data is data about a particular location. Based on this data, AI can recognize the existence of slums and which characteristics are related to community life in slums. Apart from high monetary poverty, these characteristics are:

·       Ethnicity.

·       Religion.

·       Region.

·       High women’s fertility.

·       Poor school attendance.

Looking at city-level poverty can help determine monetary poverty at the local level. AI can analyze an enormous e-commerce data set to decide which traits point to urban poverty. For instance, AI identified that, particularly in a data set with information from 19 million advertisements for cars, motorbikes, apartments, houses, and land for sale or rent, with about 100 characteristics per advertisement such as sale prices, number of sold goods, number of viewers, and number of buyers. This implies that e-commerce data can be utilized to recognize urban poverty.

AI can also identify poverty by studying various data sets on variables that point to non-monetary poverty. On several scales, non-monetary poverty can also be detected.

Household poverty can help identify non-monetary poverty at home. Based on survey data from numerous houses and even more household members regarding demographics and health, AI may identify household poverty. Household poverty is most often indicated by data about the size and age of the household and by indicators of quality of living, such as the distance to a water source. more so than those related to health, health behaviors, and education, such as years of education.

On the city level, non-monetary poverty can, for example, be identified by looking at village poverty. AI can recognize village poverty based on data from different data sets concerning high-resolution imagery by satellites, street maps, points-of-interest such as hospitals and schools, and surface height data. From this huge amount of data, AI has extracted three dimensions that can be used to predict poverty at a village level:

·       Socioeconomic conditions, for example, the percentage of land within a specific area that is covered by buildings.

·       Access to facilities and services, for example, involves the time needed to reach the nearest hospital.

·       Agricultural production conditions, for example, involve the proportion of land covered with forest, cropland, or slopes.

In this case, the percentage of land within a specific area covered by buildings was most relevant for recognizing village poverty, followed by access to facilities and services.

Looking at county-level poverty can help determine non-monetary poverty at the county level. AI can take advantage of nighttime satellite photos that, for example, indicate variations in the amount and intensity of nightlight used in each area. To identify these variations, AI evaluates various pixel attributes, such as the sum of all the pixels within the county boundary and the proportion of pixels with values larger than zero. A picture comprises hundreds of tiny areas of imagery called pixels. When using nine or more of these pixel attributes, variations in nightlight can be used to determine the relative wealth and development of various counties.

Examining poverty and malnutrition at the national level can help identify non-monetary poverty. Data on asset poverty, the number of children with a healthy weight, the number of underweight children, the frequency of impaired child growth due to poor nutrition, and the number of underweight women are used to quantify poverty and malnutrition. Based on past observations and information about geographical elements related to location and remoteness, AI can identify these indications. Relevant factors include the number of underweight women and asset poverty. Compared to vegetation and the weather, these indications are more appropriate. It’s also important to note that while being frequently employed as indicators of poverty, conflicts and sudden increases in the price of food have proven to be far less effective at detecting poverty and malnutrition.

In conclusion, based on various diverse and enormous data sets, AI can be utilized to identify both monetary and non-monetary poverty. New indications can be found and used in different data sets, making this conceivable. The individual, household, community, city, county, or national levels can all be used for this.

 Georgios Ardavanis – 25/04/2023