Hands-on: How AI can help Persons-In-Need

11 min readOct 12, 2024

A previous article described how Machu Picchu, a collaborative and open-source project, leverages technologies of the 21st century to change completely how humanitarian aid can be approached. https://kvutien-yes.medium.com/project-machu-picchu-white-paper-2024-part-1-735b60c55a92

See a 10-minute video here: https://youtu.be/z1ylfi60ES0

In this article, you can experience by yourself part of the Machu Picchu project using free AI tools. No installation, no configuration, simply from your browser. Our objectives is to use AI to dertermine how persons in need can share risks based on the similarity of their profiles.

It is long because it reproduces extensively the answers of a chatbot to illustrate our point. The reader can skip the details of the answers.

Quick Look at Machu Picchu

Machu Picchu is a toolkit composed of existing proven open source software. It allows local humanitarian staff to support the persons-in-need post their profiles while keeping ownership of these profiles.

Helper organisations, be they international or national humanitarians or commercial service providers, can access these profiles against a micro-fee paid directly to the persons-in-need.
These profiles can be transformed by AI into kinships. Using blockchain and the AI measures of kinship, these persons may mutually share risks. This is what you will experience hands-on here, using a only a browser.

What you will exercise hands-on in this demo?

All you need is a free Google account (for example Gmail) and a browser. There is no installation nor configuration required. You can stop anytime, quit the browser and come back to resume later.

You will use AI to generate 10 profiles of agropastors in Niger, 10 profiles of small farmers in Latin America, 10 more in South India. The profiles need not be recorded in any strict database format.
You will assume that one of these persons suffers of a hardship. You will ask AI to identify among the 30 profiles in your “database” who are the people who are “closest” (in profile) to the victim to help.

If you are curious to know how these people can help each other across borders and overseas, see explanations in the video above. This exercise aims only to demonstrate AI in action, to help persons in need. If you want to dig deeper into the technical implementation, join Machu Picchu.

Our program of exercises

Step 1: Generate random but realistic profiles of persons in need

You will use Generative AI to generate 10 profiles of agropastors in Niger, 10 profiles of farmers in Latin America, 10 profiles of farmers in South India.

Step 2: Feed AI with the profiles and ask questions

You will feed Generative AI the 30 profiles that you generated. Think of it as a kind of database.
You will ask questions about the profiles, like “who is closest in profile to Ramesh Sabu to financially help him overcome this bad season?” or “to build a gender inclusion program, which type of reward will be most appreciated?”

In real life, the profiles will be shared by real persons, in free text. There is no special formatting required. The existing AI tools will sort out the meanings of the profiles. The persons will keep ownership of their profiles by the help of blockchain smart contracts. See the Machu Picchu White Paper above for technical details.

Let’s go! Generate profiles of Persons in need

We generate our random profiles using duck.ai : in your browser, open the site https://duck.ai, click on “Get started”, click “Next” to use the first of the AI models proposed (GPT-4o mini), click on “I agree”. In the prompt field, at the bottom of the page, type the following to initiate the context:

list 20 components of a profile of a person in need that the humanitarian organisations want to collect the most?

The answer is random but in my case, here is what duck.ai says:

When humanitarian organizations assess individuals in need, they typically collect a variety of information to understand their circumstances and provide appropriate assistance. Here are 20 components that may be included in a profile of a person in need:

Name: Full name for identification.Age: To understand the demographic and specific needs.
Gender: Important for tailoring assistance and services.
Marital Status: Information on whether they are single, married, divorced, etc.
Number of Dependents: Information on children or other dependents.
Location: Current address or area of residence.
Displacement Status: Whether they are internally displaced, refugees, or host community members.
Health Status: Any medical conditions or disabilities.
Access to Healthcare: Availability and accessibility of medical services.
Education Level: Highest level of education completed.
Employment Status: Current job situation and skills.
Income Level: Financial situation and sources of income.
Housing Conditions: Type and quality of housing.
Food Security: Access to sufficient and nutritious food.
Water and Sanitation Access: Availability of clean water and sanitation facilities.
Social Support Networks: Family and community support systems.
Coping Mechanisms: Strategies used to deal with stress and hardship.
Legal Status: Documentation and legal rights (e.g., refugee status).
Cultural Background: Ethnicity, language, and cultural practices.
Specific Needs: Any particular needs related to gender, age, disability, or other factors.

Collecting this information helps humanitarian organizations tailor their interventions and support to the specific needs of individuals and communities.

Now you can ask duck.ai to generate the first 5 profiles:

generate 5 profiles of agropastoral persons in Niger with random values of the factors above. A few factors might be missing, a few factors can be added that may serve to better characterise one or some of the common factors

The answer you will receive might be different but in my case, here is what duck.ai says (I only quote the first profile to keep this article short):

Here are five fictional profiles of agropastoral individuals in Niger, incorporating various factors to characterize their situations:

Profile 1: Amina Diallo

Name: Amina Diallo
Age: 34
Gender: Female
Marital Status: Married
Number of Dependents: 4 (3 children, 1 elderly parent)
Location: Zinder Region
Displacement Status: Not displaced
Health Status: Generally healthy, but has a chronic respiratory condition
Access to Healthcare: Limited; nearest clinic is 15 km away
Education Level: Primary education (completed)
Employment Status: Farmer and livestock keeper
Income Level: Approximately $200 per year
Housing Conditions: Traditional mud-brick house with thatched roof
Food Security: Occasionally faces food shortages during the dry season
Water and Sanitation Access: Access to a borehole, but water quality is poor
Social Support Networks: Strong community ties; participates in local women’s group
Coping Mechanisms: Engages in barter trade with neighbors
Cultural Background: Hausa ethnic group
Specific Needs: Access to better healthcare and agricultural training

Profile 2: Moussa Ibrahim

Name: Moussa Ibrahim
Age: 45
Gender: Male

etc.

In my case, profile 5 is incomplete due to the limitations of the free GPT-4o service. Let’s ask for a complement:

continue generating profile 5

In my case, the answer is:

Profile 5: Salima Mahamadou

Name: Salima Mahamadou
Age: 32
Gender: Female
Marital Status: Single
Number of Dependents: 1 (a young child)
Location: Dosso Region
Displacement Status: Not displaced
Health Status: Generally healthy, but has experienced postpartum complications
Access to Healthcare: Moderate; local health post available, but often lacks supplies
Education Level: Completed primary education
Employment Status: Small-scale farmer and artisan (weaver)
Income Level: Approximately $250 per year
Housing Conditions: Simple mud-brick house with a thatched roof
Food Security: Faces food insecurity during the lean season
Water and Sanitation Access: Access to a nearby river, but sanitation facilities are inadequate
Social Support Networks: Active in a women’s cooperative for artisans
Coping Mechanisms: Participates in community barter systems and savings groups
Cultural Background: Zarma ethnic group
Specific Needs: Access to maternal health services and training in sustainable farming techniques

These profiles illustrate the diverse circumstances and challenges faced by agropastoral individuals in Niger, highlighting their unique needs and the factors that humanitarian organizations may consider when providing assistance.

Store the generated profiles in NotebookLM

NotebookLM is a free AI web tool provided by Google. We will store in it the generated profiles and we will ask it to look for profiles across the world that are kin to a given profile. In Machu Picchu, people with these profiles can share risks and mutually support one another on a global scale since weather risks are localised.

In your browser, open another tab besides duck.ai, open the https://notebooklm.google.com web site, login with your gmail account, click on “New Notebook”. In the dialog “Add Sources” click on the bottom right button “Copied text”. In the dialog “Paste copied text”, copy-paste the profiles generated by duck.ai above. Click “Insert”.

In the left side bar of NoteboolLM, a first note named “Pasted text” will appear. When you hover the mouse over it, the document icon is replaced by 3 vertical dots with a white-on-black tooltip saying “More”. Click on it and choose “Rename source”. Rename the note into “Niger: profiles 1–5”. Click “Save”.

Keep generating more profiles, store them in NotebookLM

In duck.ai generate 5 more profiles (4 profiles seems to hit the maximum of GPT-4o mini):

generate 4 more profiles in Niger please

continue profile 9 and add a 10th profile please

Store these profiles in another note of NotebookLM. Use the “+” icon in the sidebar of NotebookLM, choose again the button “Copied text” and copy-paste the profiles. Click “Insert”. Rename the note.

Now we generate and store the 10 profiles in Latin America and 10 more in South India. Check for completeness of the last profiles in each round and ask to complete if so desired. As you’ll see, even if some profiles are incomplete, we can still ask AI to evaluate the degree of proximity (kinship) between profiles.

generate 5 profiles of small farmers in latin America in the same vein please

complete profile 4 in Latin America and generate 5 more profiles

complete profile 8 and generate profiles 9 and 10

generate profiles of 5 small farmers in South India in the same vein please

complete profile 4 in South India and generate 5 more profiles please

complete profile 8 above and generate profiles 9 and 10

Start working with NotebookLM

Check that all notes in the left sidebar are checked to be sure that their contents are taken into account. Ask NotebookLM the following question:

profile 7 in Niger, Mariama Tchana suffers drought. She is young, divorced with 1 child. Who among the other profiles worldwide are most kin to her and can support her until the next seasonand how much each?

Here is the answer of NotebookLM:

Here is a list of profiles that could potentially support Mariama Tchana:

Profile 1: Amina Diallo

Amina is a 34-year-old married female farmer and livestock keeper in the Zinder Region of Niger.
She has 4 dependents.
Her income is approximately $200 per year.
Amina occasionally faces food shortages during the dry season.

Profile 3: Fatoumata Souley

Fatoumata is a 28-year-old married female farmer and market vendor in the Tillabéri Region of Niger.
She has 3 dependents.
Her income is approximately $300 per year.
Fatoumata’s food security is generally secure but relies on seasonal crops.

Profile 5: Salima Mahamadou

Salima is a 32-year-old single female small-scale farmer and artisan (weaver) in the Dosso Region of Niger.
She has 1 dependent.
Her income is approximately $250 per year.
Salima faces food insecurity during the lean season.

It is important to note that the sources do not provide information on the willingness or ability of these individuals to provide financial support to Mariama. Additionally, the sources do not provide a specific amount of money that each person could potentially contribute.

Therefore, it is not possible to definitively determine who among the other profiles is most kin to Mariama and how much each could contribute to support her until the next season.

Let’s ask again NotebookLM

which profiles in Latin America and South India are most similar toMariama Tchana and can support her?

Here is the answer

Although the sources provide profiles of individuals in Latin America and South India, they do not contain information about their willingness or ability to support someone in Niger. The profiles are focused on individual challenges and do not discuss transnational support networks.

Here are the profiles in Latin America and South India that share some similarities with Mariama Tchana:

Latin America

Profile 2: Maria Elena Torres: Maria Elena is a 35-year-old single mother of two in Veracruz, Mexico. She is a small-scale farmer facing food insecurity, especially during the dry season. Like Mariama, she is involved in a women’s agricultural group and participates in local savings groups.
Profile 4: Lucia Rojas: Lucia is a 29-year-old married mother of one in Santa Cruz, Bolivia. She is a small-scale farmer who faces challenges during the lean season and has limited access to maternal health services. Similar to Mariama, she is active in a local women’s cooperative and participates in community events.
Profile 6: Mariana Castillo: Mariana is a 27-year-old single mother of one in Oaxaca, Mexico. She is a small-scale farmer facing food insecurity, especially during the rainy season. Like Mariama, she is active in a women’s cooperative and participates in local savings groups.
Profile 8: Rosa Elena Paredes: Rosa Elena is a 33-year-old divorced mother of two in Quito, Ecuador. She is a small-scale farmer who faces challenges during the lean season and has limited access to healthcare. Like Mariama, she is active in a local women’s group focused on agriculture.
Profile 10: Camila Fernández: Camila is a 29-year-old single mother of one in Mendoza, Argentina. She is a small-scale farmer who faces challenges during the harvest season and has limited access to reproductive health services. Like Mariama, she is active in a local women’s cooperative.

South India

Profile 2: Lakshmi Devi: Lakshmi is a 35-year-old widow with two children in Karnataka, India. She is a small-scale farmer facing food insecurity, especially during the lean season. Like Mariama, she participates in local savings groups.
Profile 4: Meera Reddy: Meera is a 28-year-old married mother of one in Andhra Pradesh, India. She is a small-scale farmer who faces challenges during the lean season and has limited access to maternal health services. Similar to Mariama, she is active in a local women’s cooperative.
Profile 6: Anjali Menon: Anjali is a 32-year-old single mother of one in Kerala, India. She is a small-scale farmer who faces challenges during the harvest season. Like Mariama, she is active in a local women’s cooperative.
Profile 8: Priya Nair: Priya is a 30-year-old married mother of two in Tamil Nadu, India. She is a small-scale farmer who faces challenges during the lean season and has limited access to reproductive health services. Like Mariama, she is active in local women’s groups focused on agriculture.
Profile 10: Kavita Sharma: Kavita is a 34-year-old divorced mother of two in Tamil Nadu, India. She is a small-scale farmer facing food insecurity, especially during the lean season. Like Mariama, she participates in local savings groups.

While these profiles highlight individuals with similar demographics and challenges to Mariama, the sources do not provide any information on their capacity or willingness to provide financial support across continents.

How dos it apply to Machu Picchu?

Behind the scenes, this impressive example utilizes the superior pattern matching capabilities of Generative AI: the text of each profile is transformed into a vector of 1024 numbers or more. This is call “word embedding”. The vectors can then be compared using matrix multiplication. The resulting scalar value is a measure of the similarity between 2 profiles.

In Machu Picchu, we feed these similarities into a smart contract to determine which support each profile has committed to bring to the person in need.