Fecha: Miércoles 13 de marzo, 11:00 hs.

Invitado: Dr. Igor Zwir. Assistant Professor, Department of Psychiatry, Washington University School of Medicine. Associate Professor Department of Computer Science and Artificial Intelligence, University of Granada.

Lugar: Sala de Seminarios del IBioBA-CONICET-MPSP. Godoy Cruz 2390, C1425FQD, Buenos Aires, Argentina.

Título: Bringing the whole elephant into view: Can Cognitive Systems bring real solutions to complex problems?

Patricio Yankilevich e Igor Zwir
Patricio Yankilevich e Igor Zwir


Like the story of the six blind men trying to explain the nature of an elephant, current research in cognitive computational systems tries to identify the nature of an illness or complex human behaviors, and/or even anticipate possible sociology-economical phenomenons.

The giant of business machines defines cognitive systems as a category of technology that uses artificial intelligence, machine learning and reasoning, to enable people and machines to interact more naturally and to extend and magnify human expertise and cognition. They claim that these systems can provide assistance to physicians, scientist, businessmen other professionals in a fraction of time it now takes.

Not surprisingly, two of the most famous risk and financial advisory firms handle the term in their websites. One of such companies states: “the cognitive era creates newer, and bigger opportunities for business… imagine helping your workers take an amplified intelligence…”. In contrast, others suggest that cognitive systems are most frequently used as a marketing jargon.

Jargon or not, it turns out that we need help because there are too many hidden/naked/fragmented elephants around. We hypothesize that there are at least three problems that needs to be approached to reveal the nature of the elephant.

§ First, the main parts of the elephants have to be touched.

§ Second, we have to join hands around the elephant to assemble such parts.

In this talk, we will show how agnostic (unsupervised, blinded) machine learning findings can be assembled by multiobjective and multimodal optimization research techniques and utilized to uncover a multifaceted view of the elephant, which in this case is the human being.

The human-being elephant will emerge from assembling local partitions of the genomic variants can be with clinical and/or personality traits, and in turn with brain images like a Lego structure.

These descriptive features are neither simple nor independent. For example, a single genetic variant or it corresponding gene do not consistently cause a mental disorder; rather, it takes many genes operating in concert, possibly interacting with specific environmental factors, in order for a person to develop mental illness or a healthy life.

Indeed, the brain’s connecting cables can be identified by fMRI images that spot functional connections across brain regions (connectome), whereas another type of images reveal the white matter that conduct the electrical signal between different such regions. Here, not only images but also all type of features have to be considered toghether.

The diversity of the assembly in real problems, which are not oversimplified by averaging features or using naive methodological techniques, may produce different types of elephants.

Likewise, for illnesses, we will uncover distinct types of a presumably single disease, or overlapping among diseases. For “healthy personalities” we will identify distinct profiles. We will show how this knowledge will “extend the human capabilities” by achieving an integrative assessment of the whole person in relation to their risk, which will allow to generate accurate person-centered policies: from personalized diagnoses, or business opportunities, to prevention of outbreaks.

§ The third proposed problem is time.

To get not only a real/complete but also a useful description of the elephant, the results have to come up before a seeing person tells that it is an elephant. Here, we will show results of recognizing elephants ahead, even when their parts continuously show up as physiological signals in real-time in an ICU telemedicine environment. Physical alarms based on a medical device provide instantaneous alerts that a physiological signal exceeds a threshold.

However, they do not anticipate future events or risks of adverse events unless the clinicians following the signal incorporate the individual signal into a more complex predictive pattern using their own “intelligence” and experiential intuition. We will show how intelligent alarms associated to the state of risk for a patient and that patient’s future trajectory can “amplified the physician’s memory” to accurate decision making on time. This can be extrapolated to other risky financial, economical or even environmental emergency situations.

The target audience of this talk includes anybody who wants to solve real life problems.