Exploring safe AI use in Indonesia.

How much work can be delegated to AI?

• One view was that delegating work to AI calls for a change in mindset.
• Planning was seen as something that should not be delegated to AI; what can be delegated is the execution. Work such as coding was seen as delegable, provided the planning is detailed.
• For technical work, agentic flows were said to make things very easy today (so much that users tend to be “spoiled”).
• The way people work with AI was also seen as having changed: from manual chat prompting at first to multi-agent systems with very fast workflows today.
• The biggest question raised: when AI makes a mistake, who is responsible? The middle ground proposed:
  • Decisions are not handled case by case.
  • The foundation is set at the start: a system with guardrails and clear escalation points.
  • AI can decide within those limits and escalates to a human when it reaches a dead end.
• How tasks are split between AI and the human in the loop was proposed to be decided through team discussion.
• On the EU AI Act:
  • It was noted that the EU AI Act already has guidance on this.
  • Another view was that, although the rules exist, the EU AI Act is still not comprehensive.
• On cognitive impact:
  • One view was that the more work is offloaded to AI, the weaker people’s ability to decide becomes.
  • The counterargument: offloading frees room for other cognitive functions and keeps people from becoming the bottleneck that slows progress.
  • Another view was that, over time, it takes more and more effort not to delegate to AI.

How do we balance the roles of government, practitioners, academics, and frontier AI developers?

• Balancing the roles of these parties was seen as very slow. Even before AI governance, internet governance was seen as still having a lot of room for improvement (not yet balanced), so it was proposed to take internet regulation as a reference.
• The WSIS principle of multistakeholderism came up: every stakeholder has to be brought into the conversation.
• Examples from other countries:
  • China: AI use is restricted in some protected professions, meaning jobs that are considered not yet ready for AI.
  • South Korea: AI is offered free of charge in some sectors as an incentive to learn and to raise AI literacy.
• Regulatory focus:
  • Micro level: governance and adoption frameworks.
  • Macro level: where AI meets other laws already in place.
• It was seen as necessary to first understand who the users are and what they need, rather than only discussing frameworks.
• What the government can do is identify the sectors AI will affect, then protect those sectors.

What concerns do governments and major global institutions have?

The impact of AI on the workforce

• The impact was seen as happening not only at the micro level but also at the macro level, including layoffs. An example from the Philippines: large layoffs are expected as customer service roles are replaced by AI.

The macroeconomic impact

• More layoffs → less tax from formal employment → less state revenue.
• States need new approaches to raise revenue.
• An example: a tax on foreign AI. Even if the data is not stored in the country, usage can be detected through the IP addresses accessing from that country.
• The challenges differ by country:
  • Japan: no layoffs. The challenge is the culture of seniority, that is, whether senior staff are willing to take part in reverse mentorship (seniors taught by juniors).
  • Thailand: the impact on education is pushing a human-first approach, such as assessing students through face-to-face interviews. The challenge is that there are few educators compared with the number of students.

The impact on organizations and individuals

• Examples were given of companies restricting AI because of regulation:
  • In one EU member state, some companies ban AI entirely for certain jobs for fear of breaching the AI Act.
  • Some companies do not allow AI-generated code because of regulation.
  • One view was that Indonesia might consider something similar to protect workers.
• The wider scope of work brought by AI was seen as widening the career gap between junior and senior staff.

Concerns from the private sector

• Demand for junior roles is falling while demand for senior roles is rising. The senior positions needed go unfilled, because junior workers cannot quickly match the skills of senior workers. As a result, senior workers have to overwork to meet productivity expectations.
• Productivity is rising, but at the cost of wellbeing, as reviewing AI output was seen as exhausting work.
• Companies were seen as demanding ever higher productivity because AI is now available.
• One view was that this issue needs regulation, and that regulation should account for upskilling junior workers, so that job openings do not only ask for senior levels.

Can we stay creative as humans under today’s conditions?

• It was proposed to build in the fields where we are subject matter experts, combining domain expertise on top of AI platforms.
• Values, taste, and certain personas were seen as things AI cannot replace.
• The role of critical thinking, problem solving, communication, and other skills that AI cannot replace was highlighted.
• An example use case: market research with AI as the respondents. For those “respondents” to be valid, though, data from humans has to exist first.
• Australia: rather than only making students AI users, its curriculum teaches them how to be problem solvers.

Could Indonesia standardize data protection, such as masking data before it leaves the country?

• One view was that one way for Indonesia to compete in the AI race is to have sovereignty over its own data.
• Another view was that data is no longer a commodity, that big companies already hold our data, and that data no longer matters as much as it did.
• Old data was seen as already taken, but new data needs protecting.
• Developing fine-tuning for specific local cases was proposed.
• Running Edge AI locally was proposed. With limited internet access in Indonesia, Edge AI was seen as worth exploring.
• A global initiative, Digital Public Infrastructure, could be put to the government so that there are common assets, such as clean datasets.

What should we hold on to as humanity’s boundary with AI?

• Ethics was seen as what keeps people the subjects of AI. Ethical and responsible use of AI positions it as innovation rather than something met with resistance.
• Protecting work, for example through AI workforce regulations.
• Protecting education, for example by banning AI for primary school pupils, as was said to happen in NYC, to preserve critical thinking.