Models identifying at-risk jobs that could help AI upskilling strategies

It’s understood by most that artificial intelligence is a tool powerful enough to change the labor landscape. That means a likely disruption in the availability of jobs to humans, or, at the very least, a reshaping of specific tasks within current jobs. 

It’s unclear when that tipping point will happen, or whether change might come at a more gradual pace rather than a sudden avalanche. But it’ll come, and before then, the Philippines — as with any other country — should do what it can to guide the population to where it’s able to create new economic value in the AI era. 

The education department has rolled out its National AI Upskilling plan, budgeted at P1.5 billion for 2026, focusing first on high-schoolers, but will also include programs for those of working age. It’s anchored by a 5-tier system, from the basics of AI to actual skills improvement for professionals. 

Now, what might help make the program more efficient is for it to identify the most at-risk types of skills and jobs to the AI shift. Basically, what are the occupations with the most tasks that can, with enough confidence, be outsourced to AI? 

I found three models that might be helpful.

The MIT’s Iceberg Index 

First is the Massachusetts Institute of Technology’s (MIT) “Iceberg Index,” published in November 2025, which used large-scale, AI agent-based simulations in order to foresee the jobs that are at risk. 

At its core, the simulation made use of the ONET database, which enumerates the tasks of more than 900 different types of occupations in the US. This same database is also used by the other prediction models that will be discussed later. 

Then, Project Iceberg created a “digital twin” of the US labor market, representing 151 million workers as autonomous agents executing over 32,000 distinct skills, making use of thousands of “production-ready AI tools” including LLMs such as Microsoft’s Copilot, Anthropic’s Claude, and other custom workflow automations derived from larger systems. 

What the simulations attempted to see with these simulations are the overlaps of current AI capabilities with human skills. For example, a human performing manual labor such as cooking food has little overlap, aside from maybe asking for a recipe from an AI assistant. 

Meanwhile, a customer service representative guiding a customer on troubleshooting an issue has a more significant overlap or a higher “technical exposure” with current AI systems. 

Overall, it found that AI can already replace 11.7% of the US labor market, with a significant part of that coming from technology occupations, and also “cognitive and administrative tasks… across finance, healthcare, and professional services.”

Coding jobs, it found, had a high level of overlap. “AI systems now write over a billion lines of code each day, exceeding human developer output. We measure skill overlap within computing and technology occupations—the Surface Index—which aligns with real-world adoption patterns from millions of AI users,” it said. 

MIT wrote that the index is designed to help see “where to invest in training” and “which skills to prioritize.” 

“By simulating how capabilities may spread under alternative scenarios, Project Iceberg enables policymakers and business leaders to identify exposure hotspots, prioritize training and infrastructure investments, and test interventions before committing billions to implementation,” it said. 

Could a similar simulation model be conducted in the Philippines to better pinpoint AI-related job vulnerabilities and overlaps?

Microsoft, Anthropic models 

Microsoft and Anthropic both released frameworks for studying AI’s impact on jobs in February and March 2026, respectively. 

First, Microsoft. The tech giant used anonymized conversations with Microsoft Copilot related to how people were using the tool presumably in their line of work, and put it against the skills and jobs in the aforementioned ONET database. 

What it found was that AI tools are most used for “information work” or the creation, processing, and communication of information. 

It tracked how frequent an AI tool is used in their workflow, and how effective the AI tool was in helping achieve the user’s goal, resulting in what it called the “AI Applicability Score.” 

Those in the top 40 highest scores include interpreters, historians, writers, telemarketers, journalists, customer service representatives, mathematicians, data scientists, web developers, and computer user support specialists. Those with low scores include painters, ship engineers, highway maintenance workers, and housekeeping cleaners. 

To some degree, all jobs are affected. Microsoft said: “Because almost all modern jobs have some element of information processing, the study found that AI applicability is widespread across the labor market, affecting almost all occupations to some degree.” 

But it warned that a high applicability score doesn’t necessarily mean that a job will be automated and lost. Instead, it may open up workers to aspects of the job that need a more “human touch,” or looking for ways to use AI to improve their workflows. 

In any case, what the Microsoft study reveals is that because “information work” is part of almost every job, nearly all industries will be affected. What could be done is to identify effectively what parts of the job require human creativity, and to use that creativity to embed AI creatively into their workflows. 

Anthropic’s measure, on the other hand, similarly uses the ONET database as well, but this time puts that data against conversations and prompts done on its Claude tool. 

The results showed that the “10 most exposed occupations” are computer programmers, customer service representatives, data entry keyers, medical record specialists, market research analysts, sales representatives (except technical and scientific products), financial and investment analysts, software QA testers, information security analysts, and computer user support analysts. 

There are overlaps with the Microsoft study in terms of jobs being affected. Likewise, the list is tech job-heavy, similar to the findings of MIT’s Iceberg Index. 

Anthropic said, “This report introduces a new measure for understanding the labor market effects of AI and studies impacts on unemployment and hiring. Jobs are more exposed to AI to the extent that their tasks are theoretically feasible with LLMs and observed on our platforms in automated, work-related use cases.” 

Both the Anthropic and Microsoft studies also use the “task-level exposure estimates” from the Eloundou et al. (2023) model that measures “whether it is theoretically possible for an LLM to make a task at least twice as fast.” 

The metric scores tasks “on a simple scale: 1 if a task can be doubled in speed by an LLM alone; 0.5 if it requires additional tools or software built on top of the LLM, and 0 otherwise,” Anthropic explained. 

In the Philippine government’s efforts to upskill the population, perhaps concerned agencies are able to conduct localized studies in the same vein that distinctly identify these work-AI overlaps, which could possibly lead to employment insecurity. 

Drilling down the exact tasks that can be impacted by AI could lead to a more efficient deployment of an upskilling strategy. – Rappler.com