Unlocking the power of AI for better policies

by Emanuele Baldacci, Director of Digital Services at the European Commission's Directorate-General for Informatics

Public policies are increasingly focused on delivering value for money. This is, however, more challenging than ever in today’s ecosystem. 

One the one hand, government budgets are tight and limited resources should be used well to yield expected benefits. On the other hand, demands for public policies to support the well-being of citizens and business growth are increasing. 

Against this background, the policy intervention framework is challenged by several uncertainties: what is the context in which policies operate? How is this affecting their effects? Are behaviours by beneficiaries of public policies affected by both context and policies? Is this supporting or hampering the achievement of public policy goals? 

In addition to these issues, demand for transparency and accountability of policy actions requires adequate instruments to help frame the dialogue with stakeholders on objectives, instruments and results of public actions. 

While these challenges for policy making are important, in recent years advances in the availability of data from different sources and technologies to consume and interpret large amount of information offer a new perspective for policy design and execution.

There are three main priority areas where governments should invest to deliver better policy harnessing the power of data, information and knowledge to design, implement and assess their interventions. In all these areas, AI-powered data analytics solutions offer critical capabilities for the next generation of accountable evidence-based policies.

First priority is data integration. With multiple data sources increasingly available at low production cost (ranging from administrative data, to traces left by interaction of humans with Internet applications to sensor- and IOT-based data) data mash-ups can generate quality information through combination of different sources. 

However, data integration is challenging as multiple sources also come with different formats and data models. Technologies based on machine learning algorithms can help integrate massive data volumes based on clustering, distance minimization and data profiling. 

The development of open data spaces and libraries of data integration algorithms will help governments create the needed policy information capability.

Second, AI-based predictive analytics can be used to assess policy effects at design stage. Using integrated data systems and AI data analytics can help simulate at very low cost expected results of different policy interventions. This allows mapping winners and losers of planned actions and possible second-round effects of public measures. Simulating beneficiaries responses and behaviours with these technologies would make policy design more realistic and credible.

This can assist in optimizing the design of public measures and also help engage open communities in policy design and dialogue at the early stage of the government intervention cycle.

Third, assessing policy outcomes and intervention sustainability can benefit greatly from a data-driven approach. AI-powered solutions can help identify the additionality of these measures, by simulating different scenarios based on large volume integrated data. Advanced regression-based techniques can be used to conduct quasi-experimental analysis on data sets, which integrate administrative data about intervention beneficiaries with context data and information about target and control group characteristics. 

These are three key areas where AI-based solutions coupled with large volume integrated data spaces can be a game changer for evidence-based policy making. Governments in Europe and beyond are now investing on these priorities. The European Commission has also launched a corporate initiative to harness data for better policy making services through the Data4policy initiative, in the context of the implementation of the Communication on Data, Information and Knowledge Management.


https://ec.europa.eu/info/about-european-commission/service-standards-and-principles/transparency/freedom-information/access-documents/information-and-document-management_en

Unlocking the power of data analytics for evidence- based policies is now more possible than ever. Governments face similar challenges and opportunities and cooperation is essential to reuse generic solutions and progress faster to design, execute and assess public policies based on data-driven approaches. 

 

Comments

Submitted by Marius Andreiana on Tue, 17/07/2018 - 16:46

Thanks Emanuele for the comprehensive overview!

Definetly, AI can help spot patterns. When properly directed, it should be able to uncover insights that human analysts missed. For example, what should be the accepted air pollution levels from various sources (car & truck engines, factories in specific industries) in 2020? How can we correlate pollution data with health impact and costs? What financial incentives would make sense to implement those new regulations? What are the projected results with these proposed limits? What cities are already well below the limits and how to they manage that?

All these would require not only good algorithms, but also quality data. Besides focusing on AI software, collecting and managing various data should be a top priority too.

In reply to by Marius Andreiana

Submitted by Emanuele BALDACCI on Tue, 17/07/2018 - 17:00

Good point! Data quality is definitely critical. We should look into the world of official statistics for good examples of data quality frameworks and metrics to measure data quality based on different dimensions. 

Submitted by Richard Krajčoviech on Wed, 25/07/2018 - 15:41

Very helpful initiative. Many governments struggle in digging information and knowledge from the available data. My two cents:

1.  Data privacy and careful anonymization of the data is very important (as already pointed in comment by Mariana POPOVA), so these initiatives become examples of GDPR compliant and ethical usage of AI. We should always think of potential abuse of collected personal information, even with governments.

2. If data are properly anonymized, a cooperation with universities and scientific centers might be beneficial. Because of the wide potential of this initiative, cooperation qwith rather more than less of them might accelerate AI knowledge, experience and competition, and in turn improve European dominance in Artificial Intelligence area.

Submitted by Kresimir Kalafatic on Sat, 27/10/2018 - 18:58

Dear Mr. Baldacci,

I have a few questions regarding "Data4policy initiative, in the context of the implementation of the Communication on Data, Information and Knowledge Management." which you mentioned.

Questions:

1. going through digital documents on EC, EU legal documents, and other documentation most of the documents are available in PDF format. My question is: why PDF documents produced by EU departments don't have digital signature ?

Some state agencies in other countries have directives for digital signing legal documents, public procurement documents and tenders in an effort to protect the citizens and companies from downloading altered and unauthorized documents. Here is example from US department of Agriculture.

https://www.ocio.usda.gov/sites/default/files/docs/2012/DR3640-001%5B1%5D.pdf

https://www.aphis.usda.gov/mrpbs/systems/fmmi/downloads/How_to_Insert_a_Digital_Signature_PDF_Doc.pdf

I understand that the matter is complex, because of different tools used in different departments, but PDF standard is used in most of the departments.

2. DARPA has recently published it is starting project SafeDocs whose purpose is restore trust in the documents because most documents don't have security features embedded and formal proof of the document standards and document tools (readers and viewers) in missing. 

https://www.darpa.mil/news-events/2018-08-09

As a member of AI Alliance I have published several documents on Open Library. One of it is avalable at following link:

https://ec.europa.eu/futurium/en/european-ai-alliance/restoring-balance-between-stakeholders-problems-tools-and-proof-concept

The idea of the paper was to increase the security of current documentary system and documents, and enable easier indexing and finding documents. The procedure uses old technology (available for almost two decades), and is a technical extension of the procedure used by USAD (adds additional security feature). The benefit is that adding the digest in the paper, the reader protects the document, but additionally enables readers to find the document version which was referenced by the document author. Also in the event of  internal or external attack on the document distribution site, the citizen or company can detect the document is not authentic. The simple procedure presented in the paper reduces the attack surface.

Receantly AI HLEG addresed trusted AI question. Having secure and trusty documents is an important component of trusted AI because adversary attacks on documentation (text,video,..) or other data by hackers (using AI or other tools )is a reality in cybersecurity.   

Is there some similar project to DARPA SafeDocs in EU?

To me SafeDocs project looks related to several AI projects in the US.