Kybernetik Intelligent Agents- Decision Making Mohamed Oubbati Institut für Neuroinformatik Tel.: (+49) 731 / 50 24153 mohamed.oubbati@uni-ulm.de 03. 07. 2012
Intelligent Agents Environment Agent Intelligent Agents continuously perform 4 functions: 1. Perception (sensory data + interpretation). 2. Reasoning. 3. Decision making. 4. Action selection to affect the environment in order to meet their design objective.
Intelligent Agents Environment Agent Intelligent Agents continuously perform 4 functions: 1. Perception (sensory data + interpretation). 2. Reasoning. 3. Decision making. 4. Action selection to affect the environment in order to meet their design objective. How to make decisions?
How to design an agent? Reactive Pro-active Social
How to design an agent? Reactive Pro-active Social
How to design an agent? Reactive Agent Environment Perception (World state) Conditions rules stimulus response (e.g. if-then-else) Action Execution this is a simple reflex agent...
How to design an agent? Model-based Reactive Agent Environment Forward Model (how the world evolves) Past Experiences Internal State Perception (World state) Conditions rules Action Execution this agent is reactive but uses its internal state...
How to design an agent? Pro-active An agent is generally asked to do things, i.e. his existence should have a sense (for us). Pro-activeness means taking the initiative in order to achieve goals; not simply reacting to preceived events.
How to design an agent? Agent Pro-active Forward Model (how the world evolves) Environment Past Experiences Internal State Perception (World state) Goal What action should be selected? Action Execution this agent has a goal directed behavior...
How to design an agent? Balancing Reactive and Pro-active It is desirable that agents systematically work towards long-term goals, but also react appropriately to changing conditions. This remains an open research problem
back to the important question: how to make decisions?
How to make decisions? 1. Let s consider an agent in a complex, dynamic, and sometimes novel environment. 2. This agent has to follow a long-term goal. How to proceed?
How to make decisions? 1. Let s consider an agent in a complex, dynamic, and sometimes novel environment. 2. This agent has to follow a long-term goal. How to proceed? Maybe research in Artificial Intelligence (AI) can help.
How to make decisions? How to proceed? - by pre-programming the agent for every possible situation?
How to make decisions? How to proceed? - by pre-programming the agent for every possible situation? NO
How to make decisions? How to proceed? - by pre-programming the agent for every possible situation? NO - by training the agent to learn the consequences of every possible action in every situation?
How to make decisions? How to proceed? - by pre-programming the agent for every possible situation? NO - by training the agent to learn the consequences of every possible action in every situation? NO
How to make decisions? How to proceed? - by pre-programming the agent for every possible situation? NO - by training the agent to learn the consequences of every possible action in every situation? NO What would the AI guys say?
How to make decisions? How to proceed? - by pre-programming the agent for every possible situation? NO - by training the agent to learn the consequences of every possible action in every situation? NO What would the AI guys say? 1. If situations have been seen before, the agent has simply to remember what has tried in the past. 2. If situations are slightly different, the agent would be able generalise from what it has learned.
How to make decisions? How to proceed? What about completely new situations? actual AI:?! That might explain, for example, why until this moment their are no cars that can navigate autonomously (without driver) in urban streets.
How to make decisions? How to proceed? What about completely new situations? actual AI:?! That might explain, for example, why until this moment their are no cars that can navigate autonomously (without driver) in urban streets. It is time to thing about a new AI that considers intelligence as a whole, and not only for pre-specified applications
New AI An agent should be intelligent, in an abstract sense, without being specialized in pre-defined tasks.
New AI An agent should be intelligent, in an abstract sense, without being specialized in pre-defined tasks. but how to program an agent without even knowing the task? This is the challenge of the new (since 2001) field of research: Artificial General Intelligence (AGI)
Danke für die Zusammenarbeit und viel Erfolg bei der Klausur