AI and the future of Community and Social Impact work: what changes, what stays human

Key facts

• The ILO puts social work and counselling professionals (0.28) and secondary teachers (0.30) in its lowest band, "not exposed", on a 0 to 1 scale; government benefits officials score 0.45 (gradient 2). Global generative-AI task-exposure estimates; not a Malaysian forecast; not a job-security rating^[1].
• Government and public sector employers in the World Economic Forum (WEF) survey said people mainly do 53% of tasks today and expected 34% by 2030, with 39% done by people and technology together. Global employer expectation, published 2025^[2].
• WEF employers put social work and counselling professionals and secondary and university teachers among the largest-growing jobs to 2030. Global employer expectation, published 2025; not a Malaysian vacancy count^[2].
• Malaysia's National Guidelines on AI Governance and Ethics (MOSTI, September 2024) are voluntary and include fairness, inclusivity and privacy among seven principles^[3].

Social impact work is mostly about people: listening, organising, teaching and making a case for change. AI is changing the reports, proposals and case files around that work. It also raises a question this area cares about: who benefits from an AI tool, and who gets left out?

For routes in, SPM subjects and any registration the work needs, start with Community & Social Impact after SPM.

How to read this page

Observed means someone measured it. Projected means someone estimated the future, so we say who, when and for where. Projections are often wrong.

The scores come from the International Labour Organization (ILO). They estimate how much of an occupation's current tasks generative AI (tools that write, summarise or make images from a prompt) could potentially do, from 0 to 1^[1]. "Not exposed" is a technical label: most of the occupation's tasks scored consistently low for what generative AI can do today^[1]. It does not mean AI will never change the work, and it says nothing about Malaysian employers or funding. Exposure is not job loss either: the ILO says that depends on adoption and on whether workers get the chance to learn the tools^[1].

What is already happening (observed)

• Education is a large use of one AI product. In Anthropic's February 2025 data, education and library tasks were 9.3% of Claude conversations, the third-largest group^[4]. Anthropic usage data (Claude.ai conversations sorted into US job categories), not Malaysian demand.
• Malaysia. We did not find a current Malaysia-specific study that measures how NGOs, social enterprises or public services here use AI. Ask practitioners on Day 2; for now their answers are the closest evidence you will get.

What people project, and who says so

• Public sector employers worldwide. 90% expected AI to transform their work by 2030. They said people mainly do 53% of tasks today, technology 17% and the two together 30%, and expected 34%, 27% and 39% by 2030. Global employer expectation, published 2025^[2].
• Working with AI more than replaced by it, in this sector. The WEF says nearly half of the expected fall in tasks done by people alone in government and healthcare is expected to come from people working with technology, not from automation. Global employer expectation, published 2025^[2].
• Care and teaching. Employers put social work and counselling professionals, secondary teachers and university teachers among the largest-growing jobs to 2030. Global employer expectation, published 2025^[2].
• Skills employers name. Leadership and social influence was a core skill for 61% of employers, and empathy and active listening for 50%. Global employer expectation, published 2025^[2].
• Robots, a US model. Anthropic's September 2026 model rates community and social service among the groups with relatively low robot exposure^[5]. US model data, not Malaysian demand.

Task by task

These cards are our reading of the sources for a classroom exercise, not a forecast. A practitioner may disagree, and their reasons are worth hearing.

Listen to a client and plan their support

• What a person does: builds trust, judges risk, keeps confidences.
• What AI might assist: notes and paperwork, never the conversation itself.
• What this source does not show: caseloads or hiring in Malaysian social services.

Run a residents' meeting or organise volunteers

• What a person does: reads the room and handles conflict.
• What AI might assist: schedules, reminders and posts.
• What this source does not show: how community groups here are funded or staffed.

Teach a class or run a workshop

• What a person does: knows these particular learners and adapts on the spot.
• What AI might assist: lesson plans and worksheets.
• What this source does not show: Malaysian school rules on AI, or teacher demand.

Research a policy question and write it up

• What a person does: checks every source is real and says what the AI claims.
• What AI might assist: finding and summarising studies.
• What this source does not show: how policy teams here use these tools.

Process applications for aid or benefits

• What a person does: handles exceptions and checks the system treats people fairly.
• What AI might assist: sorting and checking forms.
• What this source does not show: whether automated sorting here is fair; fast and unfair can go together.

Write a funding proposal or impact report

• What a person does: makes sure the numbers are true and the community's voice is in it.
• What AI might assist: first drafts.
• What this source does not show: how funders will judge AI-written proposals.

New tasks that are appearing

• Checking automated decisions for bias, so a system does not quietly exclude the people a programme is for.
• Helping people who are offline or less confident with technology to use services that have gone digital (see the People and Society lab brief).
• Verifying AI-written research summaries before they shape a policy or proposal.
• Keeping people's data safe. Privacy and data security, fairness and inclusivity are principles in Malaysia's voluntary AI governance guidelines^[3].

This list is our reading of the sources above, not an official one.

What stays human: hypotheses to test

These are practitioner hypotheses to test, not findings. The sources do not show that any of these tasks will stay unchanged.

• Does being trusted by someone in a hard moment still need a person?
• Who brings together people who disagree?
• Who speaks up for a community and answers to it?
• Who notices who is missing from the room?

Ask a practitioner where a digital system has already changed their work with people.

What this means for you

• Build people skills on purpose. Empathy, listening, project planning and fundraising are the skills listed for this area. Choose the work because it suits you, not because a global score is low.
• Learn enough data and AI to question them. You will be asked to judge reports, dashboards and tools. Knowing how they can be wrong is part of protecting people.
• Treat people's data with care. Never paste real names, IC numbers, addresses or case notes into a public AI tool.
• Get real experience. Volunteering, a student society or a community project shows you the work and gives you something to talk about in applications.
• Watch how Malaysian public services use AI, how the AIGE guidelines are applied, and the National AI Action Plan 2026-2030, led by AI Malaysia since 28 July 2026^[6].

Your next step

1. On Day 2, choose the Social Impact taster (a youth campaign for a more liveable street) or, if your cohort runs four labs, the People and Society lab. Read the People and Society lab brief and try the community service centre task trial.
2. Ask the practitioner: "Where has AI or a digital system helped the people you serve, and where has it left someone out?"

About the evidence

ILO scores (a global generative-AI task-exposure estimate; not a Malaysian forecast; not a job-security rating), on a 0 to 1 scale with the ILO's band^[1]:

• Sociologists, anthropologists and related professionals 0.48, gradient 2
• Conference and event planners 0.48, gradient 2
• Government social benefits officials 0.45, gradient 2
• Policy administration professionals 0.42, minimal exposure
• Social welfare managers 0.37, minimal exposure
• Social work associate professionals 0.33, minimal exposure
• Secondary education teachers 0.30, not exposed
• Social work and counselling professionals 0.28, not exposed

The bands run from "not exposed" and "minimal exposure" up to gradients 1 to 4, the most exposed^[1]. The ILO built its scores from a survey of 1,640 workers in Poland, rounds of discussion with international experts, and an AI assistant built on that input. Every task counts equally, and the ILO says local factors its global study does not cover can speed up or limit the change^[1]. ILO occupation groups do not map one-to-one onto Malaysian job titles, and we do not claim a direct mapping.

WEF figures are expectations from a survey of more than 1,000 employers across 55 economies, published January 2025^[2]. Anthropic figures are its own usage data or a US model, not Malaysian demand. The method behind the task cards is in Tasks, not job titles.

For facilitators

Use the task cards as a starting list for the Day 2 AI task X-ray and ask the practitioner to correct them before learners use them. Each learner supports one card with a source and challenges one assumption. Stop any conclusion that a job is "safe" or "dead": the evidence here does not support either.