My inclination for mathematics and physics, but concurrently the passion for concreteness of mechanical engineering, as well as the interest for social phenomena, politics, and economics, have guided my studies towards the wide field of the complexity science. This relatively new and multidisciplinary area of research investigates the emergence of complex collective behaviors of systems generally composed of simple interacting elements.

During my doctorate, I focused my activity on the modeling of human groups behavior, both trying to investigate the determinants leading to the emergence of the collective intelligence, with the aim of simulate and describe social phenomena and scenarios (teams, squads, parliaments, etc.), but also trying to exploit the emergence of the human groups’ collective intelligence and criticality for developing a new swarm-based optimization tool, the Human Group Optimization Algorithm, for solving complex mathematical and engineering combinatorial optimization problems, with the final goal to design a swarm of direct interacting robots.

The ability of animal and human groups in solving complex problems is widely recognized in nature. Flocks of birds, schools of fish, colonies of ants and bees, are just a few of the best-known examples in the animal world, where interaction and collaboration between agents with limited ability and capabilities, allow them to solve problems exceeding individual skills (foraging, defending from predators, building structures, etc.). This phenomenon is known as Swarm Intelligence.

Similarly, human groups, exploiting the potentialities of social interactions and sharing their limited knowledge, are collectively able to achieve much better performance than single individuals can do.

We daily experience and use results of the human group collective intelligence. Think about networks of people that through internet and sharing platforms join and collaborate each other to create something cool, intelligent and extremely useful: Wikipedia, where millions of people spread across the world share, almost voluntary and free, their knowledge, creating the biggest and most accurate world encyclopedia; Linux, a free and openSUSE operating system, with a source code can be edited, improved, checked and shared by everyone, an OS covering 99% of world supercomputers; Amazon and eBay, emblems of the well-known power paradox: the best way to gain power is to give it away.The superior ability of groups in solving tasks originates from collective decision making: agents (animals, robots, humans) make choices, pursuing their individual goals (forage, survive, etc.) based on their own knowledge and amount of information (position, sight, etc.), and adapting their behavior to the actions of the other agents. Even though the single agents possess a limited knowledge, and the actions they perform are usually very simple, the collective behavior, enabled by group-leaving and social interactions, allows the knowledge and information sharing, leading to the development of a superior intelligence of the group.

Different studies have been proposed in the literature, trying to model the human behavior. Most of them recognize the imitation as the fundamental mechanism by which the phenomena of crowds, fads, and crime could be understood. The imitation is, in fact, the basis of learning: children imitate their parents and peers, entrants into an industry imitate more established firms, governments in less developed countries imitate governments in more developed countries. However, if everyone strictly imitates, improvements in choice cannot occur, while it is widely recognized that humans’ cooperation led to achieving astonishing performances in the history of humanity.

One of the main results of my research activity has been to model, consensus-seeking apart, the rational evaluation that characterizes human beings, to compare alternative strategies in terms of costs and benefits and efficiently solve a problem. During my doctoral program, I deeply studied a decision-making model (DMM) inspired by human groups behavior [1], conceptually proposed by my doctoral supervisors G. Carbone and I. Giannoccaro. The DMM attempts to capture the previously mentioned two drivers of the humans’ behaviors in groups, i.e., self-interest and consensus-seeking.

Agent’s choices are made by optimizing the perceived fitness value, which is an estimation of the real one, based on the level of agent’s knowledge. However, any decision made by an individual is influenced by the relationships it has with the other group members. This social influence pushes the individual to modify the choice it made, for the natural tendency of humans to seek consensus and avoid conflict with people they interact with. Consequently, effective group decisions spontaneously emerge as the result of the choices of multiple interacting individuals.

In [2] we identify the criticality as the condition triggering to the emergence of the collective intelligence of the group, showing that just when the dynamics of the system changes qualitatively from a disordered (low consensus) to an ordered state (high consensus), a significant amount of information leaks from the complex fitness landscape to the group of agents, improving their abilities to explore the search space and leading to a superior performance of the group decision making in terms of fitness values.

The DMM has also been used to analyze management problems [3], simulating how a team of individuals with different expertise, in charge to solve a task, like design a new product, converge towards a shared solution of the design process by interacting with each other.

We investigated the influence of the team hierarchical structure and social network topology on the team performances, providing indications about how to effectively design a team. A similar formulation can be easily adapted to the simulation of any kind of social decision-making problems, i.e. modeling of parliaments and squads’ dynamics, coordination of human crowds, etc.We also introduced a new promising optimization algorithm belonging to the class of swarm intelligence optimization methods [45-6].

The proposed algorithm, the Human Group Optimization (HGO) algorithm, is developed within the previously mentioned DMM [12] and emulates the collective decision-making process of human groups. By tuning some control parameters (HGO) [45] or through a Self-Organized process (SO-HGO) [6], it is possible to make the system undergo the critical transition towards the state of collective intelligence. While this state being active, a cooling schedule is applied to make agents closer and closer to the optimal solution, while performing their random walk on the fitness landscape.

To test the ability of the algorithm, its performance has been compared with those of Simulated Annealing (SA) and Genetic Algorithm (GA) in solving NP-complete problems, consisting in finding the optimum on a fitness landscape. In all cases, the proposed algorithms have been shown to significantly outperform the others, especially under limited knowledge conditions.

Thanks to a series of expedients, the improved version of HGO, the Self-Organized Human Group Optimization algorithm (SO-HGO) [6], performs considerably better than the former, both in terms of averaged final performance over a certain number of replications, that in terms of the shape the distribution of the final values around the average. Here the system is capable of bringing itself autonomously at the criticality, where the collective intelligence of the group emerges and performs at best, presumably realizing the so famous Self-organized criticality (SOC), the mechanism by which the complexity arises in nature. I can’t say more about SO-HGO because the relative paper is still under review; new details will be hopefully published soon.

I am an engineer and basically, I like to solve real problems; for this reason, I want to end this session just spending some words about the possible implication of the work done in our daily life. We are living in the era of the artificial intelligence, a century that will bring humanity to the development of machines able to learn, extremely fast and accurate, capable to perceive their environment and take actions maximizing their chance of success at some goal, devices that will access to huge data of information and knowledge, much more of those which human beings will be able to access, showing better problem-solving abilities of them.

We are already witnessing to the first prototypes of self-driving cars, devices able to recognize images and able to understand human speeches, as well as of interpreting complex data, able to compete and defeat humans in high-level strategic games (Chess and Go).

It could be quite ambitious to include in the algorithm developed the possibility for the agents to learn and acquire knowledge as the exploration process goes on. This goal could be achieved introducing an artificial neural network allowing the agents to build up their perceived landscape. This approach could be the initial step for designing swarm of direct-interacting robots, able to learn and behave like humans do, through the implementation of the Self-Organizing HGO algorithm. The ability of the optimization procedure to reach a good solution even in a noisy environment, or rather when each agent of the group possesses only a limited knowledge of the problem to be solved, could have an astonishing impact on the efficiency of these swarming robots for solving complex problems. Moreover, the field of application could be extremely wide, from the developing of swarms of drones for military actions to the realization of groups of robots replacing human beings in dangerous and uncomfortable jobs, like mining detections, cleaning and maintenance acts in hostile environments, but also in monitoring activities, detection of hazardous events, leakages of chemical substances, etc. This approach could reveal itself more powerful than the classical one with a unique specialized robot, mainly because of the robustness of the swarm (if one robot fails, the rest of the swarm continues working), but also because the decomposition of tasks could allow improved and faster performance accomplished by simpler robots, which in total could cost less than a unique more complex single machine carrying out the same tasks.

Finally, I really advise watching next video to have an idea of where we already are in the artificial intelligence revolution, that, faster than anyone could have predicted, is completing changing our life.

 

SCIENTIFIC PUBLICATIONS
International Archive Journals peer-review
1. Carbone, G., Giannoccaro, I. (2015). Model of human collective decision-making in complex environments. The European Physical Journal B, 88(12), 339. https://doi.org/10.1140/epjb/e2015-60609-0
2. De Vincenzo, I., Giannoccaro, I., Carbone, G., & Grigolini, P. (2017). Criticality triggers the emergence of collective intelligence in groups. Physical Review E, 96(2), 022309. https://doi.org/10.1103/PhysRevE.96.022309
3. De Vincenzo, I., Giannoccaro, I., Carbone, G. (2017). How social network features and organizational structure impact team performance in uncertain environments. Emergence: Complexity and Organization, 19(2). 10.emerg/10.17357.03a0541223537bc7607aa7035475ba39
4. De Vincenzo, I., Massari, G.F., Giannoccaro, I., Carbone, G., Grigolini, P. (2018). Mimicking the collective intelligence of human groups as an optimization tool for combinatorial problems. Chaos Solitons & Fractals, 110, 259–266.  doi: 10.1016/j.chaos.2018.03.030.
5. De Vincenzo, I., Giannoccaro, I., Carbone, G. (2016). The human group optimizer (HGO): mimicking the collective intelligence of human groups as an optimization tool for combinatorial problems. arXiv preprint:1608.01495.
6. De Vincenzo, I., Giannoccaro, I., Carbone, G., Grigolini, P. (2018). A self-organized optimization algorithm mimicking the collective intelligence of human groups. Submitted to IEEE Transactions on Evolutionary Computation.
International congresses with peer review
1. Giannoccaro, I., De Vincenzo, I., Carbone, G., Ising model of the dynamics of inter-organizational teams, European Conference on Complex Systems, September 22nd-26th, 2014, Lucca (Italy).
2. Giannoccaro, I., De Vincenzo, I., Carbone, G., An Ising-based approach to the study of inter-organizational team dynamics. The IEEE International Conference on Industrial Engineering and Engineering Management, pp. 1209-1213, December 9th-12th, 2014, Selangor (Malaysia). doi: 10.1109/IEEM.2014.7058830.
3. De Vincenzo, I., Giannoccaro, I, Carbone, G., Factors affecting the performance of autonomous teams with project leader: a social network approach. IFKAD 2015, June 10th-12th, 2015, Bari (Italy). ISBN 978-88-96687-07-9, ISSN 2280-787X.
4. De Vincenzo I., Giannoccaro I., Carbone G., Mimicking Human Groups in Solving Optimization Problems, 19° International Working Seminar on Production Economics, February 22nd-26th, 2016, Innsbruck (Austria).
5. De Vincenzo, I., Giannoccaro, I., Carbone, G., The Human Group Optimization (HGO) algorithm: Exploiting the collective intelligence of human groups as an optimization tool for complex landscapes, CSSSA 2016, September 19th-23rd, 2016, Roma (Italy).
Other publications
6. De Vincenzo, I., Emergence and exploitation of collective intelligence of groups, PhD Thesis in Mechanical and Management Engineering, Politecnico di Bari, January 2017.