Arti cial intelligence has been often perceived as a promising eld that can provide methods and tools that can be extended to other research areas and exploited in industrial applications. Digital libraries and archives can thus have a twofold bene t. Since research and evaluation can hardly be carried out without the development of a working system, based on a real collection accessed by real users, it could be expected that arti cial intelligence plays a central role in the design and development of the many components of a digital library or of a digital archive. This short communication starts from a review of what was expected from arti cial intelligence in the research areas that converge in the interdisciplinary eld of digital libraries, concluding with a number of considerations on what could be today's expectations from the variety of persons { researchers, stakeholders, librarians, users { involved in the design, development and access to organized digital content.
Methods and tools developed in the eld of Arti cial Intelligence (AI) are pervasive in many other research areas. Advancements in very di erent elds, from robotics to multimedia indexing were possible thanks to AI research results. AI is an interdisciplinary eld both in the scienti c background of researchers who contribute to the eld and in the applications of its results.
Among other elds, it is probably particularly relevant to investigate the impact of AI in the development of Digital Libraries and Archives (DLAs). This because DLAs are themselves multidisciplinary both when considered as existing institutions and when considered as an applicative research area. The issue can be faced by comparing the early expectations, about how AI could contribute to the advancement of disciplines related to DLAs, with nowadays expectations, about the directions in which user experience can be improved using AI.
One of the rst concepts taught to university students in computer science and computer engineering is the, apparently simple, equation data + context = information. The context in which this important piece of information is delivered is usually a course on database systems, where its direct implications in software design are more evident. In this case the goal is mainly to provide the user with all the necessary background to contextualize and give meaning to the data, but since the early days of automatic information processing the idea of exploiting AI techniques to automatize this process has been considered as an important issue.
For instance, already in 1980, Schank published a short communication [
The integration of database and AI research has been discussed in the
literature since a long time, as witnessed by the introduction of the term intelligent
database [
Another research eld that is usually perceived as tightly connected with
AI is Human-Computer Interaction (HCI). Also in this case, the relationship is
witnessed by the introduction of the term intelligent user interfaces, although
the main focus was probably on the development of natural language processing
techniques. Yet, it has been argued [
As reagards DLAs, the idea of exploring how AI can contribute to the eld
motivated the organization, already in 1997, of the international workshop
\Articial Intelligence and Digital Libraries", which was part of the International Joint
Conference on Arti cial Intelligence (IJCAI). It is interesting to note that in the
same year it was presented a paper on the integration of machine learning
techniques directly targeted to the development of an intelligent digital library [
As it can be seen, these four main points are all related to the research
areas previously described, from HCI to IR and database research. It can be
noted that most of these expectations have been, at least partially, ful lled. For
instance, machine learning plays a central role in the management of multimedia
content, which is increasingly present in DLAs. Recent approaches to automatic
identi cation and classi cation of images, music and video largely relies on AI
techniques [
Given the centrality of users in a DL system, the integration of AI research results in working digital libraries has to take into account nowadays expectation of all the types of users, from stakeholders and librarians to the general public. These expectations are not supposed to be the main driver of individual research projects, but can highlight the general direction to which projects might converge. Even if there is a time span of 45 years, nowadays expectations can probably be expressed by a famous quote by Marvin Minsky, in a 1970 interview to Life magazine:
In from three to eight years we will have a machine with the general intelligence of an average human being. I mean a machine that will be able to read Shakespeare, grease a car, play o ce politics, tell a joke, have a ght. At that point the machine will begin to educate itself with fantastic speed. In a few months it will be at genius level and a few months after that its powers will be incalculable
This sentence actually summarizes the abilities of the main character of the movie Her (Spike Jonze, 2013), a hand-free operative system voiced by Scarlett Johansson. Apart from being able to compose music and create jokes, as envisaged by Minsky, the operative system in Her seems to understand the content of documents and actively use this ability to lter out irrelevant mails, assign priorities to messages, and even spontaneously select texts based on their artistic quality in order to publish a book. Of course, science ction always looks ahead of the actual time, but at the same it is an important litmus test on which computer abilities are considered crucial for an e ective improvement of the user experience.
Thus, today's expectations are probably not very di erent from the ones envisaged in the eighties and early nineties: AI should provide results that help computer programs to automatically infer a context in order to transform structured and unstructured data into understandable information.
To this general idea, it could probably be added another facet on the concept
of understandability. Users, in particular scholars, are willing to understand why
a system recommends, suggests, lters, a particular set of documents [
The need for integrating intelligent techniques in all the many components
of DLAs, from search tools to data management and user interfaces, emerged
decades ago and resulted in the development of many intelligent tools that are
used to manage large digital multimedia collections. Yet, very few digital library
systems dare to call them selves intelligent [