Artificial intelligence is already an integral and easily usable part of our daily life, as evidenced by its constant use in many large Italian companies. Just think of driverless cars or voice assistants like Microsoft’s Cortana, Apple’s Siri, or Amazon’s Alexa, just to name the most well-known. It is not difficult, in fact, to realise that every day we deal with intelligent algorithms that are able to self-learn and that help us in our daily life in many ways:
There are also numerous examples of how AI has had a positive impact on business or public administration procedures by automating, reducing errors or allowing the development of new products and services.
In this article we will deepen some aspects of artificial intelligence, in particular those related to the deep neural network, and we will ask the authoritative opinion of Piero Poccianti, the president of the Italian Association for Artificial Intelligence.
WHAT IS ARTIFICIAL INTELLIGENCE
Before delving into the concepts of the deep neural network it is necessary to give a definition of artificial intelligence. According to the definition provided by the Politecnico di Milano, “Artificial Intelligence is the branch of computer science that studies the development of hardware and software systems with typical human capabilities and which is able to independently pursue a defined purpose by making decisions that, until then, were usually entrusted to humans”.
However, there is no unambiguous definition of IA, and interpretations can be of various kinds, depending on the scope of interest. One can focus on the internal processes of reasoning, for example, or rather on the external behavior of systems, always starting from the similarity of reactions and results with respect to human behavior. To date, the scientific community defines two different types of artificial intelligence: the so-called weak and the so-called strong. The starting point takes into account how human skills, by innate characteristic, concern the understanding and processing of natural language and images, learning, reasoning and planning skills, but also the interactions with people, machines and the external environment.
At this point we need to make a clarification: unlike traditional software, an AI system is not based simply on the programming made by developers who write their operating code, but on progressive learning techniques, that is, on the definition of algorithms that, by processing a huge amount of data, lead the system itself to reevaluate and advance its own understanding and reasoning skills.
The attempt to simulate human reasoning with artificial intelligence opens up several scenarios of ethical reflection, which have been explored in this article.
FROM MACHINE LEARNING TO DEEP NEURAL NETWORK
From a technological and methodological point of view, what we call artificial intelligence is understood as a learning process that generates a task or an action.
To date, there are two main learning models: Machine Learning and Deep Learning. The first case is related to functional systems that allow to train the software to correct errors, so that it can learn and then perform an activity independently: think for example of a mechanical hand that performs a very high precision cut using a control algorithm. Machine learning is progressing, however, on the use of neural networks organised in several levels of depth, for this reason called Deep Learning: we are talking about relatively recent development learning processes (after 2010), inspired by the structure and functioning of human neuronal networks. Clearly, in this perspective, Deep Learning needs specially designed artificial neural networks, which we call deep artificial neural networks, and consequently an extremely powerful computational and energy capability to “hold” different layers of computation and analysis. These systems are already in use, for example in pattern recognition, speech and image recognition, and Natural Language Processing systems.
THE INTERVIEW WITH PIERO POCCIANTI: HOW FAR CAN ARTIFICIAL INTELLIGENCE GO?
“The question was and it still remains long-standing and difficult to answer” says Piero Poccianti, president of the Italian Association for Artificial Intelligence (AIxIA), because the debate on the merits has been going on for a very long time and concerns above all the possibilities of General artificial intelligence, that is, the possibilities of the systems to carry out multiple different tasks. “We have no scientific basis to say whether or when IA will be able to achieve human capabilities. This uncertainty is not related to certain areas in which it already exceeds it, such as calculation, but rather in the more complex area of creative, transversal and abstraction reasoning”, explains Poccianti. We can think, for example, how the human mind conceives four basic categories: perceiving, seeing, knowing how to listen, knowing how to recognise an object or the world around us.
“Perception implies the recognition of configurations, so even the possibility to recognise something anomalous in an ultrasound is an example of how Artificial Intelligence can be applied, which also has the ability to learn based on previous events – continues Poccianti. The fourth capacity is also at an early stage and has been already tested, but the one on which the challenge is played and the fundamental question remains is the ability to abstract thinking, in which man expresses, for example, similarities with similar processes already tested in various fields and which can be replicated also by adopting creativity or modifications, for example.”
It is on this aspect, combined with the ability to independently build new strategies, that the possibility to answer to the great fundamental question will be concentrated: understanding how, if and how much the so-called machines will somehow be able to cross the categories mentioned, adding the large piece related to the ability to sense that is still missing from the ability to abstract.
The aforementioned considerations are reflections on which ICT research centres dwell every day, in the context of the development of artificial intelligence. It can also be said that, although to the question asked to Piero Poccianti about how far artificial intelligence can go there are still no definite answers, it is precisely this “uncertainty” that acts as a stimulus for an ever-greater commitment in the field.
For more insights on the state of ICT research in Italy you can take a look at this article.