Seminar cercetare DFT „A new view of the Dark room problem under the Free Energy Principle framework”
„In this presentation I will start by introducing the Dark room problem and the solutions offered to it by Karl Friston, Andy Clark and others, under the Free Energy Principle framework. I will then formalize the Dark room problem, discuss its versions and understand how the premises and its conclusion were treated by Friston and others, essentially rejecting the intermediate conclusion that agents tend to inhabit Dark rooms. In the end, I will propose my own solution to the Dark room problem, showing that agents do tend to inhabit dark rooms in many occasions and this is an effective solution to the problem of reducing free energy. This goes against the usual opinion of Free Energy Principle proponents. What is in accordance with Friston and others is the idea that agents tend not to stay too long in these dark rooms due to the fact that this does not lead to long term reduction of free energy. Long term reduction of free energy will not happen because of average short term free energy, as Friston suggests, but rather due to other processes that are simply linked to models that have long term reduction of free energy as their objective. An essential part here lies with the amended Free Energy Principle that I propose, according to which agents can receive direct input parameters, instead of using their own statistical models for reducing free energy. This helps with explaining why human agents do not fall over long periods of time into the trap of Dark room problem. It is because engaging in reach social and cultural environments reduces the free energy over long term due to the use of direct input parameters, as explained by the amended Free Energy Principle.”
