From ChatGPT to Government: Making the Elephant Dance

A recent event hosted by Whiteshield delved into the profound impact of AI models like ChatGPT on public policy. The event commenced with a captivating demonstration that showcased the remarkable capabilities of Large Language Models (LLMs).

During this demonstration, we presented a hypothetical scenario where a civil servant attended a 3-day United Nations Conference on Crime Prevention and Criminal Justice, harnessing the power of ChatGPT to tackle a series of increasingly complex tasks. The first task involved summarising three days of conference proceedings into concise ‘key points’ and ‘takeaways.’ The second task entailed constructing a comparative table to benchmark and analyze various countries’ approaches to addressing Crime Prevention and Criminal Justice. The third task required a quantitative analysis of crime prevention data, while the final challenge was to draft a proposal for a new law on Crime Prevention. Remarkably, ChatGPT version 3.5 effortlessly completed all these four tasks within a matter of seconds (refer to Figure 1).

Figure 1: Disruption of the civil servant’s workday

The demonstration made it abundantly clear that the latest advancements in AI are ushering in a new reality for the realms of education and employment as we have traditionally known them. Chat GPT stands out as one of the fastest-growing AI platforms ever developed, achieving an astounding milestone of one million users within just five days. By March 2023, it had garnered an impressive 1.6 billion visits and demonstrated its prowess by performing in the 90th percentile on some of the most challenging examinations known to us, such as the US Bar exam and international math Olympiads.

Beyond the realm of civil service, Large Language Models (LLMs) pose a potential catalyst for accelerating workforce disruptions. A study conducted by Goldman Sachs highlights that LLMs have the potential to accelerate work processes by up to 50%. Furthermore, recent research from OpenAI suggests that over 30% of the US workforce faces a high risk of automation, potentially resulting in the disappearance of more than 25 million jobs (refer to Figure 2).

Figure 2: The impact of LLMs on the US labour market

However, it’s important to note that, in many instances, it’s specific job tasks that will feel the impact, leaving room for workers to pivot towards more high-value tasks.

For the time being, most jobs should remain secure. Sir Christopher Pissarides, a Nobel Prize laureate in Economics and Director at Whiteshield, highlighted an enduring pattern that spans over two centuries: the talk of technology replacing jobs on a massive scale, which has never fully materialised as predicted. One key factor is that it’s often individual tasks, rather than entire jobs, that are disrupted by emerging technologies. Tasks affected by technology can typically be categorised into four groups: complemented tasks, automated tasks, new tasks, and deleted tasks.

As an illustration, consider the job of a teacher (as depicted in Figure 3). Complemented tasks might involve providing more comprehensive feedback, harnessing insights from both teachers and AI.

Figure 3:  The evolution of teacher tasks in the face of digital disruption

Automated tasks encompass the instantaneous sharing and updates of class materials, making them readily available to all students. New tasks may involve crafting customised curricula and presentations tailored to individual students. Furthermore, the traditional task of delivering course materials can be phased out if all students can access personalised presentations on-demand through AI. Collectively, these developments empower teachers to amplify their impact on students through intelligent AI utilisation.

To illustrate further, let’s consider another scenario: the bureaucrat featured in our demonstration could delegate tasks such as conference summaries, data analysis, country benchmarking, and legislative drafting to the Large Language Model (LLM). This would free up their higher-level skills to ensure that the new legislation on crime prevention incorporates relevant feedback and follows the appropriate legislative procedures.

Large Language Models have the potential to drive significant impact within the public sector if the main risks are addressed.

On a broader scale, Large Language Models can play a pivotal role in augmenting the public sector’s effectiveness, both internally and externally (refer to Figure 4). Internally, these models, known for their generative capabilities, not only boost the productivity of civil servants but also enhance creativity. They assist in generating relevant, well-structured content, thereby streamlining and expediting the development and implementation of public policies and legislations, as exemplified in the earlier demonstration.

Externally, Large Language Models can be harnessed to offer personalised customer support services to citizens regarding policies and regulations. This can be achieved through the deployment of ‘virtual assistants’ and more intelligent chatbots, ensuring citizens receive tailored assistance and information.

Figure 4: Internal & external enhancement opportunities for Public Policy and Governments

Another significant impact could be the enhancement of Government communication, tailored precisely to the specific needs of diverse audiences. Furthermore, AI’s ability to predict citizen trends and scenarios can fortify Government decision-making processes.

However, it’s crucial to consider potential pitfalls when applying generative AI tools and Large Language Models (LLMs) to the public sector. These challenges may arise from privacy concerns, data security risks, and the possibility of biases stemming from the vast amounts of data used to train the algorithms. Additional limitations may emerge if the reasoning processes of LLMs, which are not immediately transparent, are automatically and unsupervisedly applied to certain areas of Government decision-making activities.

In summary, the transformative potential of AI, exemplified by Large Language Models (LLMs) like ChatGPT, presents both remarkable opportunities and critical challenges for the public sector. As we navigate this path toward a future where AI augments public policy and government activities, we must remain vigilant in addressing issues of privacy, data security, and algorithmic bias. Yet, if we succeed in mitigating these concerns, the prospect of a more responsive, creative, and citizen-centric government comes into view. By harnessing the power of AI as a strategic ally, we can usher in an era where civil servants are empowered to focus on shaping impactful policies, and citizens experience an even more agile and effective government. The journey ahead holds great promise, and with responsible use, we can realize the ‘win-win’ scenario for the public sector and society at large.