.
FRDCSA | internal codebases | Agent

Agent Architecture Diagram: GIF

Jump to: Project Description | Capabilities

Project Description

Capabilities

• Some ideas for unilang. 'screen -r' and 'daemon' might assist with the type of agent interface we want. Right now, the problem with unilang is many fold. But I won't deal with this problem, I will side step it and start cso manually.
• unilang should also reply that the agent is not active.
• Figure out what's wrong with elog, not recording anything, are we missing a necessary agent?
• The agent that handles reading list stuff should mix and mince reading to break the flow of it and play on particular interests.
• Complete more agents and refactor the agent start stop code to be more intelligent.
• An agent that does its own planning and reasoning. Right now all I do is write code.
• I have to develop an intelligent agent sooner or later.
• Build a better agent system that allows start and stop.
• We need a system that handles its own contingencies, therefore avoiding exception programming. Call it agent based programming.
• Natural language is something that operates on top of an intelligent agent, so without that agent, you can't understand natural language.
• It has to be an intelligent agent fully capable of making certain decisions. Note that complexity will be difficult. It has to have a lot of knowledge about itself.
• Maybe use BDI agent based approach for setanta. Could use Jason, Spark or something.
• 'Intentions are both (i) a means of reducing the computational complexity faced by an agent by at least temporarily committing to one procedure and not constantly reconsidering possible ways of performing each task and (ii) a commitment to a certain behavior that helps coordination when dealing with other agents or human users.',
• Some nice systems that I saw today were: Heart Rule or something. Proteus conversation agent. Bunch of stuff from a website indexing natural language generation systems and stuff from the natural language index.
• Should have a facilitator agent, which is responsible for starting and stopping other agents. And, it maintains execution logs, which you can view, multiple simultaneous ones.
• General the contact features of manager to any agent... and then use the audience tell told stuff. In fact that is the basis for the AA.
• Should use WSD on corpus entries and also run ExtractAbbrev.java on them to get abbreviations like Emacs unilang agent, run this through termios, and use that to boot strap translation. Also do anaphora resolution by running it on chunks of aligned texts.
• If an emacs unilang agent command finds that unilang is not running, start it.
• Use rememberance-agent code for manager context classification
• Let's get manager going by turning him into a unilang agent and allowing him to send messages, or execute programs. For instance, if the daily items haven't been done by the end of the day, manager reports that.
• In other words, an entire agent that reasons about these things, including its own belief structures would be useful. There is software for this, but it isn't very straightforward and probably not free.
• If a singular item is entered, look it up and if it is some kind of purchaseable item, or a software system etc, route to appropriate agent.
• Should we have an unilang SQL agent?
• Sorcerer, -a agent-based
• Sorcerer, -a agent based
• Sorcerer, -a multi-agent
• Sorcerer, -a agent
• test-agent, shutdown
• test-agent, die now!
• test-agent, this is another test
• test-agent, test
• agent diamond deregistered
• test-agent, done
• deregister test-agent
• The busroute agent does not encourage analytical plans, however, but does allow for quick action.
• Got to get busroute converted to be an agent. Depends on that dialog functionality.
• Now fix up broker as an agent, so I can tell it I need a glass casserole dish
• Should be able to start an agent from this window, have a unilang, start agent audience, type feature.
• Might be worth it to print which agent it is in front of every line of any of the agents.
• Need to make sure that you can change the speed of that one system. You could rebuild and make it a unilang agent I'm sure.
• Come up with agent debugger framework for unilang.el, that is, you load it up and it helps you debug the unilang agent you are developing.
• Dreams. I dreamt that I bought a mini electric car for $400. It looked like a very small jeep. It rode pretty slow. But I was able to drive it places. I was at CMU. I accidentally stopped in the undergraduate lounge, because I was hiding from people. Earlier professors had made fun of my theory, so I let, and then made a loud noise in the hallway a little earlier. Even earlier in the dream was dreams about sitting in in various CASOS meetings that I wasn't supposed to. Almost by accident, and then having a run in with an agent of theirs. There was an open source convention of sorts and I got to meet an important guy, and he discussed his position, which made sense.
• One way to make myself more effective is this - when I type an instruction, like "get latest knoppix dvd" into unilang, it is going to route this to the appropriate agent using the agent management system.
• Fix bug with unilang-emacs-agent, where it aays args out of range, when receiving certain messages
• By not giving away too much information which can't be true or is specific to agent interaction history, I could then meet in person and have very little to study in the briefing.
• Some features audience needs, before I forget: should model who knows what, and more specifically, who knows which agents are programs. Also, if there is no danger in admitting to being an agent, the agent should then work with the user to determine what are the weaknesses in the agents dialog. This feedback should be sent to developers, or architect.
• Making Emacs into a unilang agent greatly increases our abilities here.
• Clearly make an agent for bugzilla.
• job-search should really just become savant, and profile should be changed to agent. Thus, capabilities management for agents is completed. That is useful, more generally. job-search then uses capabilities management for the agents. Pretty cool.
• We can use the AIM agent to quiz people on topics in real time to verify their knowledge.
• Convert busroute to be a unilang agent, by tomorrow
• Anticipating User Tasks Using Commonsense Reasoning, Alex Faaborg, Chris Tsai, Henry Lieberman, The goal of this project is to create an agent that predicts tasks users may be interested in adding to their to-do list, based on the context of appointments they are adding to their calendar. The agent uses Commonsense Reasoning to predict tasks that users are likely to add. For instance, if the user makes the calendar appointment \u201cgo fishing\u201d, the agent may recommend the task \u201cpack fishing equipment.\u201d By automatically generating potential tasks, the agent will save user\u2019s time, and will help them remember actions they may commonly forget.
• Improve mobile agent knowledge suit.
• Use Cyc, eventually logicmoo to model world state. Make sure that we have to ask manager "permission" to do everything, and also all requests. manager, I am hungry. This initiates manager to plan for the world state. This calls the Cyc agent for world state, which it sends to manager, who then sends the modified domain file to the verber agent, which then generates the plan, which is then returned to the manager agent.
• For all agents. I.e., unilang::agent::agent should parse that automatically.
• fix bug with not checking agent names.