#2: Some insights on the main concerns’ impacts on AI within the public sector supported by NLP and TM*

INTRODUCTION

Artificial Intelligence (AI) is a subject that has been introduced previously. Indeed, AI started in the 1950s [1]. Nevertheless, only in the last two decades AI could be disseminated through different sectors of society, such as business, industry and public administration, mainly due to three factors: First, the exponential development of tools to collect and store immense volumes and variety of data, also known as BigData [2]. Second, the development of powerful Graphics Processing Units (GPUs) [3]. Third, developing and releasing open source packages such as TensorFlow and PyTorch [4].

Businesses [5], industry [6] and the public sector [7] have applied AI in several different ways and contexts and considering multiple concerns. Some examples of these concerns that have been more intensively observed in the last few years are data privacy, ethics, interpretability, explainability, trustworthiness, and fairness [8]–[10].

Considering this context, I applied fundamental concepts of Text Mining (TM) and Natural Language Processing (NLP), two techniques embedded in the universe of AI, to help retrieve potential papers and extract helpful insights on the main concerns’ impacts within the public sector.?

METHODOLOGY

The methodology adopted to achieve the objectives of this work contains two parts. First, an advanced search (i.e.,?using a search query containing Boolean Operators and Wildcards) was performed, and filters were applied. Second, basic concepts of NLP and TM were applied to help the final selection of the papers to analyse thoroughly.

First, an advanced search was performed in the Database of the Association for Computing Machinery Digital Library (ACM Digital Library). It converges around 170 conferences, symposiums and workshops yearly, with many of these events focused on the public sector. The search query used was (“artificial intelligence” OR “AI”) AND (ethic* OR fair* OR “data privacy” OR interpreta* OR expla* OR trust*) AND (“public sector”). After, three filters available on the website were applied: Last Two Years, Proceedings and Research Articles.

This timeframe of two years is reasonable to avoid obsolescence since advances in this area have occurred at a breakneck pace. In addition, conferences are crucial opportunities for presenting and discussing research in different phases (conception, research, implementation, and updating). The results are then published as proceedings, which can effectively capture cutting-edge innovation. The search query returned 569 results. Filtering the last two years returned 253 results. Filtering by proceedings yielded 206, and filtering by research articles returned 153 results.

Second, some basic concepts of TM (search for words and expressions in the body of the text) and NLP (word count and metric calculation) were implemented to help select the papers. At the end of this process, the result was a set of 50 papers to analyse thoroughly. The results are as follows.

RESULTS

The first result regards the most relevant concern in the 50 papers selected. Code 3 counted the search words in the body of these papers using Code 3. The top five occurrences were: fair (1808), fairness (1349), transparency (806), ethics (696) and ethical (500).

The second result regards the main topic in the selected papers. They were evaluated and categorised according to the most prominent subject in the body of the text as follows: responsible AI (2%), explainability (4%), ethics (20%), transparency (22%), fairness (24%) and trustworthiness (28%). Data privacy was not captured as a specific topic, although embedded in most of the papers.

Finally, analysing the papers more in-depth and retrieving essential insights about state-of-the-art AI development in the public sector was possible based on these topics. The analysis was performed without computational resources (even though possible), and the main insights are summarised in only one paragraph, citing the main references.

Public organisations developing ethical guidelines and legal/regulatory rules for AI systems can play a crucial role in the behaviour of responsible AI public systems, significantly impacting all the stakeholders involved (citizens, companies and public services) [11]. Explainability and transparency are essential concerns and should be implemented and incorporated into AI public systems. They impact decision-makers and citizens that must be aware of and understand these systems and their practical effects [12], [13]. Fairness and accuracy need to be highly considered in AI public systems. One concern impacts each other, and the cost of the trade-off between them is one of the main obstacles to its adoption and regulation [14], [15]. Trustworthiness is vital to preserving fundamental rights such as freedom and impacts the growing confidence in governments, social values and well-being [16]. Finally, the ethical issue is an umbrella. It covers the attention and impacts all the previous concerns regarding AI public systems towards improving their applicability and results to all parts involved [17].

CONCLUSION

The design, development, deployment and use of AI systems require the participation of stakeholders from distinct domains such as technology, public, regulatory and users [18]. The concerns such as those approached in this work are not individuals and should be applied jointly to improve AI systems, ensuring human-centricity and public benefits with progressiveness and sustainability [17].

Finally, with these systems’ increasing sentience and autonomy, regulations and public policies on legal obligations of AI responsibilities should adapt beyond human ethics [11].

REFERENCES

[1]???????S. K. Rogers, M. Kabrisky, K. Bauer, and M. E. Oxley, “Computing Machinery and Intelligence Amplification,” in Computational Intelligence, IEEE, 2009, pp. 25–44.

[2]???????S. Demigha, “The impact of Big Data on AI,” Proc. - 2020 Int. Conf. Comput. Sci. Comput. Intell. CSCI 2020, pp. 1395–1400, 2020, doi: 10.1109/CSCI51800.2020.00259.

[3]???????Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, May 2015, doi: 10.1038/nature14539.

[4]???????M. Langenkamp and D. N. Yue, “How Open Source Machine Learning Software Shapes AI,” AIES 2022 - Proc. 2022 AAAI/ACM Conf. AI, Ethics, Soc., pp. 385–395, 2022, doi: 10.1145/3514094.3534167.

[5]???????A. I. Canhoto and F. Clear, “Artificial intelligence and machine learning as business tools: A framework for diagnosing value destruction potential,” Bus. Horiz., vol. 63, no. 2, pp. 183–193, Mar. 2020, doi: 10.1016/j.bushor.2019.11.003.

[6]???????C. Ferreira and G. Gon?alves, “Remaining Useful Life prediction and challenges: A literature review on the use of Machine Learning Methods,” J. Manuf. Syst., vol. 63, pp. 550–562, Apr. 2022, doi: 10.1016/j.jmsy.2022.05.010.

[7]???????G. Misuraca and C. van Noordt, Overview of the use and impact of AI in public services in the EU. 2020.

[8]???????European Commission and Directorate-General for Communications Networks Content and Technology, “Ethics Guidelines for Trustworthy AI,” Luxembourg, 2019. doi: https://data.europa.eu/doi/10.2759/346720.

[9]???????European Commission et al., “AI Watch - European landscape on the use of artificial intelligence by the public sector,” Luxembourg, 2022. doi: https://doi.org/10.2760/39336.

[10]?????J. Whittlestone, R. Nyrup, A. Alexandrova, K. Dihal, and S. Cave, “Ethical and societal implications of algorithms, data, and artificial intelligence: a roadmap for research,” 2019. [Online]. Available: https://www.nuffieldfoundation.org/sites/default/files/files/Ethical-and-Societal-Implications-of-Data-and-AI-report-Nuffield-Foundat.pdf.

[11]?????A. Deshpande and H. Sharp, “Responsible AI Systems: Who are the Stakeholders?,” in Proceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society, Jul. 2022, pp. 227–236, doi: 10.1145/3514094.3534187.

[12]?????N. Maltbie, N. Niu, M. Van Doren, and R. Johnson, “XAI tools in the public sector: a case study on predicting combined sewer overflows,” in Proceedings of the 29th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering, Aug. 2021, pp. 1032–1044, doi: 10.1145/3468264.3468547.

[13]?????M. Loi and M. Spielkamp, “Towards Accountability in the Use of Artificial Intelligence for Public Administrations,” in Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, Jul. 2021, pp. 757–766, doi: 10.1145/3461702.3462631.

[14]?????R. Islam, S. Pan, and J. R. Foulds, “Can We Obtain Fairness For Free?,” in Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, Jul. 2021, pp. 586–596, doi: 10.1145/3461702.3462614.

[15]?????M. Jakesch, Z. Bu?inca, S. Amershi, and A. Olteanu, “How Different Groups Prioritise Ethical Values for Responsible AI,” in 2022 ACM Conference on Fairness, Accountability, and Transparency, Jun. 2022, pp. 310–323, doi: 10.1145/3531146.3533097.

[16]?????G. Maragno, L. Gastaldi, and M. Corso, “e-Government for Public Values creation: a systematic literature review,” in DG.O2021: The 22nd Annual International Conference on Digital Government Research, Jun. 2021, pp. 386–397, doi: 10.1145/3463677.3463692.

[17]?????J. H. Lim and H. Y. Kwon, “A Study on the Modeling of Major Factors for the Principles of AI Ethics,” ACM Int. Conf. Proceeding Ser., pp. 208–218, 2021, doi: 10.1145/3463677.3463733.

[18]?????E. Papadopoulou and T. Exarchos, “An Ethics Impact Assessment (EIA) for AI uses in Health & Care,” ACM Int. Conf. Proceeding Ser., 2022, doi: 10.1145/3549737.3549755.

* Author Notes

(1) This work was accepted as a Poster Presentation at the 2023 IEEE Conference on Artificial Intelligence (IEEE CAI), Santa Clara, California, USA. Nevertheless, it was withdrawn and not presented due to personal reasons. Hence, it was also not published in the proceeding of the conference.

(2) An extension of this work containing more robust insights, the codes used in the NLP and TM steps and a more complete evaluation of the methodology was published on the arXiv and can be viewed at the following link:

https://doi.org/10.48550/arXiv.2308.02042