One of the challenges during surgery is to make a proper distinction between vital tissues, structures and organs, both at the surface of the exposed area and beneath it. Currently, identification of different tissue types relies either on the skills and anatomical knowledge of the surgeon (e.g., nerves, blood vessels, different tissue types), or on the results of off-line histopathological analysis on biopsies for assessment of cancer-free margins. These approaches are observer-dependent, or, as in the case of off-line histopathological analysis, also time consuming. A quick, non-invasive, and more observer-independent intraoperative tissue recognition method, applicable both ex-vivo and in-vivo, would be of great assistance during surgical procedures. Hyper-Spectral Imaging (HSI) has a lot of potential for such applications. It combines high spatial resolution with spectral information at numerous wavelengths that cannot be separated by human eye in the visible, and also beyond the visible spectral range. This technique enables enhanced discrimination between different tissue types based on their specific spectral signatures. We will describe the approach to HSI which is being developed within van't Hoff Medical Shared Research Program at TNO in collaboration with multiple clinical, academic and industrial partners. We study and analyze wide-band reflectance spectra of different tissues types collected through a fiber-based probe, both in-vivo during open surgery procedures and ex-vivo on freshly excised tissues. These data are used to identify spectral signatures of different tissue types. Based on this information, we develop data processing techniques to identify and distinguish the target tissues. We also further develop this approach towards a HSI setup for tissue imaging. The first step with such a setup is testing on inorganic phantoms and animal tissues. At a later stage the setup will be employed for ex-vivo human tissue analysis, both at the surface and up to several millimeters beneath the surface, with the final goal of in vivo intraoperative tissue recognition and tissue-specific contrast-enhanced imaging

What are the most important design principles that we should follow to build an Artificial General Intelligence? What should be the key constituents of systems-level models of cognition and behavior? In the target article “Conceptual Commitments of the LIDA Model of Cognition”, Stan Franklin, Steve Strain, Ryan McCall, and Bernard Baars tackle these difficult problems. They propose twelve “conceptual commitments” or tentative hypotheses that form the core of the Learning Intelligent Distribution Agent (LIDA) model that they have been developing over the last ten years or so. Although the article is focused on the LIDA model, these “conceptual commitments” have much broader scope and are offered to the AGI community as specific constraints that should inform the research agenda for the realization of an Artificial General Intelligence (AGI). The twelve specific “conceptual commitments” are of various kinds and have different degrees of importance for LIDA and AGI more generally. Some (Systems-level Modeling, Global Workspace Theory, Learning via Consciousness, Feelings as Motivators and Modulators of Learning, Transient Episodic Memory) are considered to be key for LIDA and also more broadly for AGI. These are general mechanisms of learning, memory and inference that should form the core of realistic, real-world architectures of brain and behavior. Of particular note, the authors highlight the importance (among the other things) of feeling and consciousness, which are regarded as fundamental architectural solutions to the problems of AGI. These themes, which were given minor importance in traditional cognitive (neuro)science and AI, have increasingly gained prominence in the last few years. Putting these themes at center of AGI research is a distinguishing aspect of the proposal of Franklin and collaborators.

We present a system that can learn to represent actions as well as to internally simulate the likely continuation of their initial parts. The method we propose is based on the Associative Self Organizing Map (A-SOM), a variant of the Self Organizing Map. By emulating the way the human brain is thought to perform pattern recognition tasks, the ASOM learns to associate its activity with different inputs over time, where inputs are observations of other’s actions. Once the A-SOM has learnt to recognize actions, it uses this learning to predict the continuation of an observed initial movement of an agent, in this way reading its intentions. We evaluate the system’s ability to simulate actions in an experiment with good results, and we provide a discussion about its generalization ability. The presented research is part of a bigger project aiming at endowing an agent with the ability to internally represent action patterns and to use these to recognize and simulate others behaviour.

What distinguishes the AGI approach from the initial, supposedly equally idealistic and holistic, AI approach? Why do we think that we could make any progress in our recent times? The answer to these questions is not clear