Quantifying Corruption - keskival/ai_enabled_transparency_of_governance_and_power GitHub Wiki

What is Corruption?

Transparency International defines corruption as:

We define corruption as the abuse of entrusted power for private gain. Corruption erodes trust, weakens democracy, hampers economic development and further exacerbates inequality, poverty, social division and the environmental crisis. Exposing corruption and holding the corrupt to account can only happen if we understand the way corruption works and the systems that enable it.

Quantitatice Indicators of Corruption

To measure and control corruption in societies, we need a quantifiable definition for it.

Transparency International further quantifies corruption by a complex perception based index, Corruption Perceptions Index (CPI). This is otherwise very nice, but as it is perception based it is not very amenable to simulations and automatic measurement or control.

Generally speaking the first generation of corruption indicators are based on experts' perception, and the second generation of indicators are based on surveys. This work belongs to the third generation, allowing automatic estimation, simulation and optimal control.

Towards a More Objective Measure of Corruption

From the machine intelligence perspective one definition which generalizes the idea of corruption is based on control and sensitivity theory. We can make an assumption that a healthy democratic society is guided by a free flow of truthful information, and makes decisions in the interests of the people. Divergence from this would be defined as corruption.

From control theoretic perspective we can raise a question of who controls the society. We model the society as a decision-making system exercising material power which is controlled by diverse interests.

Are the true interests of the people guiding the decision making and governance? If not, whose interests are controlling the society?

This model is divided into three parts:

• Is the society guided by facts and true information? We can quantify this by measuring how people's voiced preferences would change if they had access to true facts about a topic. The true interests of the person are defined as preferences when full, truthful information is available to the person.
• Are people free to voice their preferences and are they accurately measured by the system?
• Is governance sensitive to people's voiced preferences? We can quantify this by estimating how much real material effect e.g. votes have.

All three parts together boil down to measuring how sensitive the power exercised in the society is to true interests of the people. In corrupt societies each of the three parts can be subverted to private, ideological or external interests.

Coming up with ways to simulate and measure these phenomena is of great interest to us.