7 Lessons from implementing the RAM in Chile

Authors:

• Julio A. PertuzéSalas, Former Under-Secretary for Economics and Small Enterprises and Lead Expert of RAM exercise, Chile
• Jose Antonio Guridi Bustos, RAM Expert, Chile

Chile was one of the first countries in the world to implement UNESCO's Readiness Assessment Methodology (RAM), a diagnostic tool for assessing how prepared countries areto implement AI ethically and responsibly.

At Foresight Consulting, we collaborated with the Chilean Ministry of Science and Technology to gather information for the RAM and generate insights to update Chile's national AI policy.

The implementation of the RAM consisted of three stages: (1) Diagnosis of Chile's National AI landscape; (2) Developing a National AI Multi-Stakeholder Roadmap; and (3) Main Policy Recommendations for a National AI Strategy. This work was done with the support of the Patrick J. McGovern Foundation.

We distill here seven lessons from applying the RAM.

2. Gather people (and experts) to speak about AI broadly and the discussion will inevitably lead to common places. It is best to have focused discussions.

There are specific challenges in conducting participatory processes. Participants are typically self-selected and may not be representative of the whole of society. Different stakeholders compete for the power to frame assumptions, meanings, and aspirations of socio-technical systems. Policymakers must be transparent about the process's purpose and scope and recognize participants' needs and knowledge levels.

Addressing these needs, Chile's Ministry of Science and Technology decided to have focused discussions around six themes: (1) The Future of Work and AI; (2) The Future of Democracy and AI; (3) The Future of AI in Government; (4) The Future of Human-AI Interaction in Health, Education, and Security; (5) The Future of Regulation and AI; (6) The Future of the Environment and AI. These six thematic roundtables were conducted in different regions to surface local priorities.

The RAM helped arrive at concrete actions, going beyond generalist discussions into potential application areas.

3. AI imaginaries varied across regions.

Even in a culturally homogeneous country such as Chile, the visions about the future development of AI vary across regions. For instance, discussions in the north of Chile gravitated toward the opportunities for the use of AI in the mining industry and in the south about the environmental impacts of AI. The surfacing of local imaginaries about the evolution of AI went beyond the proposed themes, as local communities envisioned unique opportunities and threats related to the evolution of AI in those domains.

Local stakeholders raised concerns about the concentration of technological developments in the Global North and in just a few companies, increasing the likelihood of unintended biases in health and education from AI systems trained on different populations and cultures.

Participants advocated for increased investments in local R&D to avoid excessive dependence on foreign actors and to adapt and personalize AI development to local needs. It was recommended that local governments develop AI strategies, offering opportunities to extend AI's development beyond major cities.

The discussion facilitated by the RAM fostered a critical perspective on the adoption of frameworks and technologies from the Global North. It encouraged an examination of how national and regional identities, particularly in Latin America, can influence the formulation and progression of AI policy and development. Moreover, the implementation of the RAM received valuable feedback and identified areas for improvement when applied in the Southern Hemisphere.

4. Policymakers are eager to share how they use AI (even if they are not!)

As part of the RAM project, a Ministerial Steering Committee (MSC) was formed on August 4th, 2023. This committee was composed of policymakers from different ministries.

Despite the challenges in securing time and insightful responses from busy public officials when approaching them individually, a shift occurs when engaging with them as a collective, inquiring about their work in AI without framing it as an evaluation. This approach catalyzes a fascinating group dynamic where each member is eager to share their contributions and experiences in AI development.

However, substantial knowledge gaps regarding technology and its governance in the public sector were reported. Ministries with experience crossing databases for welfare programs (e.g., education, social development) had more preparation and guidelines in applying AI to these processes. Other ministries were just beginning to explore the uses of AI.

Coordination difficulties and challenges in establishing shared priorities surfaced during MSC meetings since AI is a general-purpose technology requiring multiple ministries' involvement. A key lesson here was to work closely with politicians (in our case, the Ministry of Science and Technology), which was essential for aligning their expectations regarding their role and understanding how their involvement will translate into tangible policy outcomes, such as the revision and update of the AI policy.

5. Resistance when applying evaluation instruments is natural; overcome it by aligning the RAM to local priorities.

Countries may be reluctant to engage in assessments if they do not recognize the value of such evaluations, particularly when they anticipate underperforming in certain indicators.

Central to overcoming this resistance is that governments perceive value. In Chile's case, a strong champion (i.e., the Ministry of Science and Technology) was interested in updating the National AI Policy, and the RAM was a complementary exercise in this endeavor.

The government's collaborative approach in providing thorough and reliable information, driven by their interest in updating the policy, enabled the formulation of more targeted and specific recommendations.

6. Novel recommendations stemmed from multistakeholder dialogue

AI is a general-purpose technology with potential applications in many areas. Stakeholders highlighted new domains, such as the influence of AI, particularly generative AI, on culture and the creativeindustries. A key recommendation from applying the RAM was to form a working group tasked with creating a report on AI's impact in these sectors.

Further, the RAM revealed a lack of ethical analysis of the technologies being used, especially in vulnerable populations such as children, women, and sexual minorities, among others, which further stresses the point of engaging with multiple stakeholders when developing national AI strategies.

7.There is a mismatch between the speed at which institutions adapt regulatory frameworks and the speed at which AI progresses.

In Chile, lagging regulations such as Personal Data Protection, Cybersecurity, Intellectual Property, and Critical Infrastructure, among others, hinder the implementation of AI.

Public officials underscored the necessity for well-defined governance regarding data quality and management as essential for enabling government agencies to share information efficiently and securely. Policymakers also emphasized the importance of educating local government actors on the ethical use of AI and the development of certifications for AI utilization within government.

The roundtables highlighted the need for industry self-regulation and the establishment of innovative Regulatory Experimentation Mechanisms, such as Sandboxes, to adapt to the fast-paced evolution of technologies, particularly for applying AI in critical areas. Additionally, it was noted that governments can advocate for ethical AI principles through the implementation of purchasing regulations and standards.