Chapter 2 - Artificial intelligence today

What is Artificial Intelligence (AI)? There are many definitions, among them, the one of the University of Stanford, which identifies it as “a science and a set of computational techniques that are inspired - albeit typically operating in a different way - by the way in which human beings use their nervous system and their body to feel, learn, reason and act” [1].

Novels and science fiction films exploit and give credit to fear of the unknown: Artificial Intelligence is often represented in a negative way, as a danger to humanity, potentially capable of taking over and dominating [2] us but in reality AI is already part of our daily life.

With the digital assistants of mobile phones, driver assistance systems, the chatbots [3] that respond when we call public or private entities, systems for machine translation of texts and speeches, robots in the factory, algorithms that recommend us products and services to our liking and those that assist us in learning, or increasingly realistic and engaging video games.

Artificial Intelligence, for example, can improve our lives by driving in our place, taking care of elderly people in need of assistance, carrying out dangerous and arduous jobs, helping us make informed decisions, rationally managing large amounts of data that would otherwise be difficult to interpret, allowing us to communicate while not knowing the language of our interlocutors, following us in our studies or increasing cultural and entertainment experiences or speeding up bureaucratic procedures.

This is possible in certain specific application areas [4] thanks to recent developments in AI techniques, essentially due to the wide availability of annotated data, to the improvement in learning techniques and to the availability of high computational capacity at increasingly accessible costs.

The main application areas

There are already examples of how PA can benefit from adopting AI solutions; among these we already see effects and applications in the health, education and judiciary system, in public employment, security and, in general, in the management of relationships with citizens.

In this latter field, Artificial Intelligence can be used to answer questions, to look for but also to process documents, fill in forms, to correctly route various types of requests or to perform translations [5].

CHATBOT

for example, a chatbot equipped with AI can respond simultaneously to conversations that come at a given time, eliminating waiting times and, once having understood the needs of its interlocutor, can direct him to the right departments, help him find the documentation he needs, provide instructions to correctly submit his written requests, if necessary also speaking in a foreign language, or translating the bureaucratic terms that the user does not understand.

In this way, public officials can be freed from the task of answering the simplest questions and tasks, which are often the most recurring, saving time to devote to other activities, to provide complex advice and better services.

Some of the functionalities described can also be used in healthcare, for example to manage examinations and tests more rapidly.

In this sector, research into robots that take care of patients is also very interesting, possibly memorising the medicines to be administered to them, the times and doses, or monitoring their state of health by means of biometric measurements to be transmitted to the doctor, even accompanied by a possible hypothesis.

Precisely this type of situation, which could lend itself to significant ethical and scientific controversies or to potential conflicts of interests between patients and doctors, highlights a focal point in which AI is excellently positioned, with its ability to create unprecedented interactions between administrative-technological systems and human beings.

Moreover, it is already possible to imagine programming intelligent systems that read tests results and interpret them, based on the statistics conducted on large amounts of similar data and on their meaning [6]. There are even predictive tools to assess potential risksThere are even predictive tools to assess potential risks of evolution of individual diseases.

Precision medicine, research on personalized treatments, analysis on the possible effects of certain drugs on individual patients are all areas that promise great results, thanks to the ability of Artificial Intelligence to store and process a lot of information on patients; however, at the same time, they require careful monitoring by the medical community as well as strict observance of privacy and ethical and professional standards [7]

Something very similar is happening in schools, where artificial teaching assistants are already available, able to follow students individually, suggesting content and concepts selected to help them develop their skills, deepen their knowledge, or bridge the gap with their fellow students [8].

These tools can be used by Public Administration for staff training, career management and internal organisation.

The security sector can also benefit from the development of Artificial Intelligence technologies. For example, in the surveillance field, with computer vision and natural language processing systems that can process large amounts of images, texts and speeches, to detect possible threats in real time.

Or in the environmental disaster prevention field, where it is possible to perform simulations of the consequences of natural phenomena, both before and after they occur, as well as during their development, helping the authorities to decide how to intervene. Similar technologies can also be used by law enforcement agencies, to patrol cities, based on continuously updated data, linked to the crimes committed in the various areas and to other significant variables.

Finally, also public decision makers can use intelligent algorithms, capable of processing large amounts of data, for example on the resources available in disadvantaged areas, or on the progress of the various sectors - health, judicial, scholastic, agri-food, etc. - in order to obtain a clearer view of the choices they must or want to accomplish. It is possible, as they are experimenting in Japan [9] that personnel in public offices respond in a personalised manner to all citizen requests, using adequately instructed Artificial Intelligence tools, which automatically produce the contents to be sent. It will be up to the administrations to decide how to use the resources freed up by the introduction of these new technologies, in a range of options that include the qualitative improvement of services and the reduction of costs for their operation.

SECURITY

One of the main problems in all these areas is that of guaranteeing the security of the AI systems: entry of an attacker in systems that, for example, will be able to drive a bus, could jeopardize the lives of those on board, as too could tampering with the surveillance tools of a public place make it vulnerable to attacks, or knowledge, by a hacker, of the model used by the Artificial Intelligence for the management of sensitive data, could make such data insecure.To effectively address these issues, it is necessary to put in place various solutions, some of a technical nature, others of a more general nature, raising awareness among citizens and administrations. From a technical point of view, it is important to work to ensure that, by law, certain standards are guaranteed, both for the security of the data used by the algorithms, as well as for the security of the algorithms themselves, imposing that, for their implementation, the security by design paradigm is always followed [10]

The potential of AI in Public Administration

The potential of Artificial Intelligence for Public Administration is manifold. Nevertheless, the scientific community and public opinion highlight some critical issues that must be taken into account in order to prevent distorted effects in the application of these tools and technologies. With reference to machine learning systems, it already happens today that both the data that an AI feeds on and the algorithms it consists of produce bias [11] distorted interpretations of the information in its possession - affecting its “reasoning” [12] and leading it into error. Making predictions with inadequately designed tools can only lead to wrong and, in many cases, even ethically incorrect decisions [13].

Furthermore, it is worth pointing out that, currently, Artificial Intelligence is able to carry out, with a certain precision, a small number of cognitive activities26 referring to specific sectorial domains, generally lacking background knowledge.

If correctly designed and implemented, AI technologies can in fact guarantee concrete prospects for improving the quality of life. In the relationship between citizens and public administration, they will be able to allow greater accessibility to public services, facilitating a substantial reduction in their costs, with benefits in terms of reduction of social spending, which can thus be reallocated. In perspective, it will be possible to enhance many procedures with adequate automation, offering citizens the opportunity to interact with the State in a more agile, effective and personalised way.

We will all benefit from this, including the elderly, the disabled and citizens belonging to the disadvantaged categories. This will allow Public Administration to recover and strengthen the relationship of trust with the community. In order to encourage this recovery, introducing technologies such as AI in PA requires technical and administrative, but also systemic, narrative and aesthetic interventions, capable of generating meaning and involvement.

PA may also use AI methodologies and technologies to support the rationalisation and integration of its databases, in a perspective of semantic interoperability that makes it easier to circulate information between administrations to the benefit of end users [14]. The enormous wealth of knowledge generated and collected over time in such databases is often “invisible”, because unstructured, dispersed in multiple archives and largely in the memory of people rather than of the organization. Making “invisible knowledge” visible is one of the potential and most promising areas of application of AI in the PA with solutions able to read and understand and classify the contents of documents and reports produced over decades to obtain the most relevant information. AI will also affect the redefinition of the relationship between public and private with a view to greater transparency through the implementation of e-procurement practices. Among the areas that in the next decade will benefit from the AI revolution, in fact, will be precisely that of public procurement.

Here the technical terms data sheets extracted from the glossary will be inserted for a better understanding of the rest of the document.

Footnotes

[1]Artificial Intelligence and life in 2030, One hundred year study on Artificial Intelligence, Stanford University, 2016, p. 5.
[2]Think of the computer Hal9000 in 2001 Space Odyssey (Kubrick, 1968) or of the exterminator robot in Terminator (Cameron, 1985).
[3]Ref. detailed box in “The main areas of application”.
[4]Ref. “Artificial Intelligence and life in 2030, One hundred year study on Artificial Intelligence”, Stanford University, 2016, p. 9.
[5]Ref. Hila Mehr, Artificial Intelligence for Citizen Services and Government, Harvard Ash Center for Democratic Governance and Innovation, 2017.
[6]Ref. IBM’s “Watson for Oncology”, imagined as a true oncologist’s assistant in the fight against cancer.
[7]For example, the paradox of automation: doctors could rely too much on the machine (that in psychological experience “never fails”) and not take responsibility for a diagnosis different from that of the machine
[8]Ref. http://www.lastampa.it/2017/06/15/multimedia/tecnologia/woogie-lassistente-robot-che-aiuta-i-bambini-a-studiare-e-non-solo-ZlrNV7rKtt3MqoKZ54U3uM/pagina.html.
[9]Hila Mehr, Artificial Intelligence for Citizen Services and Government, Harvard Ash Center for Democratic Governance and Innovation, 2017, p. 8.
[10]In essence, this paradigm requires that security is pursued already when designing Artificial Intelligence databases and algorithms.
[11]Ref. “Bias and inclusion”, AI NOW 2017 Report, p. 14.
[12]Ref. box di approfondimento in “Prevenire le diseguaglianze”. for Democratic Governance and Innovation, 2017, p. 8.
[13]Ref. “Sfida Etica”
[14]Ref. The interoperability model of the Three-Year Plan: https://pianotriennale-ict.italia.it/interoperabilita.