Soukaïna Abelhad

Soukaïna Abelhad is a computer science engineer and data scientist who has contributed to open education and developed a practical framework for examining AI automation claims. Her work includes helping Moroccan teachers assess Wikipedia critically, contributing data architecture to Wikidata for Education, and testing automation against the human labor, evaluation and costs required to deliver accepted work.

Making open knowledge useful in classrooms

Originally from Agadir, Morocco, Soukaïna organized the PyLadies chapter in Casablanca, helping women learn programming. She also became a WomenTech Network global ambassador and contributed to introductory machine-learning videos.

By 2021, she was working as a data scientist and had served as Morocco’s local coordinator for Reading Wikipedia in the Classroom. She analyzed a regional teacher survey, helped adapt training materials to local needs, recruited teachers and gathered their expectations. When COVID-19 forced the program online, she translated materials, taught modules and helped participants navigate the platform as well as the coursework. The program taught educators how online information is produced, how to assess its reliability, and how to recognize bias and missing knowledge.

Her work on Wikidata for Education addressed a different obstacle: matching learning resources to a curriculum is difficult when its standards and competencies are not machine-readable. The initiative’s Ghana pilot aimed to structure curriculum information and connect it with Wikidata so relevant resources would be easier to find and curate. As the project’s data technical lead, Soukaïna contributed to its data architecture; the broader educational program was a collaborative undertaking.

What an automation percentage leaves out

In her AI automation diligence framework, Soukaïna asks what a claim such as “70% automated” actually counts. A system can produce more output while still depending on people to finish or correct the work. She connects three checks:

• Human labor behind automation: Compare workload growth with the fully loaded labor needed to deliver it, including reviewers, trainers, escalations and rework. If those requirements rise proportionally with output, the system may be assisting people without removing the claimed labor.
• Defined autonomous completion: Specify where a workflow begins and ends, then measure how often it finishes without human intervention. Work attributed to AI and a task completed autonomously are different measures; a straight-through completion rate needs a clearly defined task that a buyer can inspect.
• Evaluation against expert judgment: Check whether automated judgments agree with domain experts. A model grading another model can provide useful evidence, but agreement with an automated rubric does not by itself establish that the work is acceptable.

These checks lead to cost per accepted autonomous task: model costs plus human review, escalation and rework, measured against tasks that actually finish and meet the acceptance standard. Failed attempts still consume resources and must remain in the accounting. Soukaïna’s framework gives buyers a way to connect an automation percentage to completed work and the operating cost behind it.