20710610 - Storia dell'intelligenza artificiale - LM

The course provides students with the general understanding of the development of Artificial Intelligence (AI) projects.
The aim of the course is the introduction to machine intelligence, with special regards to the recent evolution of data science and machine learning.
At the end of the course students will acquire the basic knowledge about the intelligence concept and its transformation in order to emerge via the algorithmic strategies of digital machines

NUMERICO TERESA

teacher profile | teaching materials

Programme

The origins of Artificial intelligence and its open ethics and politics questions

The course is dedicated to the birth of Artificial intelligence which happened more or less together witht the birth of the first electronic computers. The term was invented in 1956, but the discussion around the possibilities and consequences of what was called mechanical or machine intelligence started during the early Fifties and even earlier.
The Famous Turing Test created a huge interest since it was invented and proposed in 1950 by Alan Turing. Many years and many different methods were considered since AI origins, now many perspectives and technologies changed since then, in order to solve problems that required intelligence and problem solving capabilities to be solved.
At present there are many soft bots, many robot and many artificial devices that with the help of algorithms seem to espress intelligence and oblige us to change the concept of intelligence itself, both if applied to humans or to machines.
the most interesting phenomenon is the tendence to attribute to machine intelligent capabilities, mainly when we don't know precisely how the machine works.
This social dimention in attributing intelligence to machines was underlined by Turing too at the very beginning of machine intelligence reflection.
However this characteristic risks to produce unintendend and undesired consequences for human beings. we know that not many ara capable of understanding how to program a deep learning algorithm, and that more and more even programmers ignore how the machine produced the output results, because the layers of calculations are too difficult to follow in details also by those who programmed them. We know that only a small group of programmers and experts know the details. they tend to be trained by the same universities and to be hired by the same few multinational companies.
the technics that are more successful in AI are the machine learning programs, that build theri previsions on the manipulation of past series of data. Between the machine learning solutions the most relevant at present are the deep learning techniques. Deep learning solutions are based on multilevel layers of learning networks, in which the nobody knows what is happening inside the hidden layers. the programmers don't know precisely the reasons behind the output of the neural network to a specific problem. In some context particularly sensitive such as predictive policing, face recognition, image recognition, clusterization, risk assessment in recidivism, personnel selection, insurance premiums, welfare services ect. it is difficult to make previsions, because the structural incertitude of the future events. In these situations the strength of rhetoric about the trustworthy of algorithmic devices risk to create self-fulfilling expectations that transform the world according to previsions that are not right or wrong. the future in created by anticipating it.
this picture shows that there are political and social problems that need solutions when we introduce a new AI tool. Innovation is not enough we need to build tools that are fair and useful to promote the development and wellbeing of all humans and not only of a small minority.
the course proposes some methods and histories to understand the political and social dimensions of the technological choices around us, with special regard to AI.

Core Documentation

Margaret Boden (2018) Artificial Intelligence, a very short introduction, Oxford Univ. press, Oxford
Copeland J. (2004) The essential turing, Clarendon Press, London (a selection of pages)
Guglielmo Tamburrini (2020) Etica delle macchine, Carocci, Roma.

Type of delivery of the course

The course is held in attendance, except in the case of a new health emergency

Type of evaluation

For students attending the course it is possible to write an essay on a planned theme and to give a presentation during the course For students not attending the course, there is a written exam to evaluate the preparation on students based on the texts in the course reading list.