Multi-agent systems embedded in the social cloud offer more value then ‘Apps’.
Key words: agents, multi-agent, dynamic networks, social networks, Recommender system, cloud, cloud agent
Mobile apps have not been very gratifying. Testing an app last year gave some clarity to what I felt to be a problem with the current App ecosystem. And, this is not just a mobile issue, but also for traditional computing platforms. I have been thinking of this subject for years. This is just, finally, a very simple and pragmatic example.
Last year I downloaded an app that locates the cheapest gas based on my current location. Whether cheap gas should be used in one’s car is not the point here. The app could have been one for finding the best licensed massage therapist or bookstore. The point, is this using mobile computing to its full potential?
What if the cheap gas station is located in an area where crime is very high? Should I risk a carjacking just to save 3 cents? What if I’m about to run out of gas now, is the cheapest gas too far away? We can get even more complicated of course. What if I have to be at an appointment, shouldn’t the cheap criteria be augmented with route info; the cheapest gas is the one easiest to get to on my way to or from my appointment.
In short, the current app is one-dimensional. Real life is multidimensional and the human brain easily makes decisions within this mostly analog fuzzy chaos. If an app cannot make decisions or recommendations in that same world, it collapses the dimensions, it is a dumbing down.
How can the app be made more dimensional? AGENTS. The app should really be an Agent that cooperates with other agents to fulfill a need, in this case finding cheaper gas. Thus, it should talk to other autonomous agents, such as:
- law enforcement to grade destination
- vehicle network for fuel requirements
- retail for
- hours of operation
- Social Net
- map routing, and so forth.
It should also be informed by human agents in a trusted relationship with the user. What we then have is An Ad Hoc Dynamic Network of Social Agent Recommenders (AhDyNoSAR).
The Mind Map Diagram below gives a contextual view of this idea.
SynergisticMobileAgentSocialCloud Mind Map (click for larger view)
Let’s look at another example. Someone is walking in neighborhood that has a few restaurants. The embedded Agent notes that the last time the person ate was a few hours ago (based on shopping venue, Calendar, etc.). The shop’s agents are contacted and a decision processing workspace is created. Is the person currently viable, do they have cash or credit available? Each store will check inventory and accounting ratios, does it need to offer a discount or promotion to this person? More agents mobilize to assert their criteria. What are the person’s tastes, dietary restrictions and allergies, past intake (who wants pizza twice in one day?), and other multidimensional agents in a problem space hierarchy are evoked.
After all agents complete their reckonings and the spontaneous net reaches a stable resonance, the person’s intimate personal soft computing agents make a decision. It turns out that the person is currently following their spiritual observance and is fasting today. This result is sent into the local agent milieu and starts a new search for resonance, so no food, how about some clothing or reading material? Again a new recommendation graph is created, religious and political leanings are queried, clothing and accessory rules are fired, ah, that is a very old turban, here are some suggestions.
Unfortunately, the person has now walked into a new map space, a neighborhood park. Now new agents awake: social engagement, entertainment, sexual, defensive.
It would be so gross if the information that this new cloud offers is shown as ads. A better approach is that this information space is entered as a virtual world, using technologies like that of Massively multiplayer online role-playing game (MMORPG). The consumer becomes an Avatar moving through Recommendation Space, a superimposed view on current locality based environments. Instead of or in addition to other consumers, the other characters are the various agents most visible recommendation goal.
Unlike Apps an Agent should always be considered adversarial. That is, even when an agent provides a benefit, it also can allow intentionally or via weaknesses a loss of security and privacy since it must negotiate information with other agents. Thus, though current or future standards may be used, they must be in virtual application spaces that use encrypted anonymous data. This will be just as virus and other malware, an ongoing battle.
It would not be optimal to require a download of an agent to each user’s location or device. Instead, agents will exist in the cloud as a multi-agent system. A user will have a private cloud virtual machine and address space for agent storage and recommendation space. To handle disconnected use, an agent will have a mobile agent shadow. It will provide simple assistance and will punt decisions and actions it cannot handle until connection to the cloud is established.
With Apps, the app provider may require purchase or try to enforce lock-in or an advertising monopoly. This can also be accomplished by centralizing the App marketplace. This may not work directly with Agents. Agents may not even provide an obvious visible function. For example, an agent may just contribute parking meter locations and status to other agents that use a map agent.
In the real world eventually someone has to pay the piper. So too will the development and use of agents must be rewarded. Some options are:
An agent can contribute to an advertising stream that ultimately reaches the consumer facing user interface device.
- Agent micropayments:
Agents will negotiate among their collaborators to maintain a balance of payments, an agent of agents, and this payment is satisfied by the user or the user’s fee structure that the network provider maintains.
The consumer will purchase agents. If the fidelity and number of agents is adequate the quality of service is greater.
Of course, the internet is currently wide open and thus this opens up predation to another level if Agent “sandboxes” are porous, if personal data is not secure.
The present cavalier attitudes regarding personal privacy exhibited by the large Internet service providers is a big warning sign that giving agents access to even more information would be just another data mining delicacy ripe for exploitation.
And now for an even more far out scenario. In a classic Science Fiction novel, before a character dies, a copy of their knowledge is captured. This intelligence is then available for implantation into someone as an “Aspect”, an agent that can add its unique expertise and judgment to the human host. That is a more radical direct means for accomplishing something that the social networking may evolve into, a means to collect knowledge and translate that into a ubiquitous intelligence.
Presented was a critique of conventional app centric mobile computing and a suggestion that Agent technology can provide a more realistic computing environment. The term Agent was not defined here. Perhaps the difference with an App is just intent or where the output is ingested. The experts are still debating Agent technology and its applications.
- The Future Of The Virtual Personal Assistant
- The iPhone 4S
- Cloud Computing: Software Agents: http://francisshanahan.com/index.php/2010/cloud-computing-software-agents/
- The Kynetx Rule Language
- Kynetx and Personal Data Services: Project Neck Pain
- Eric Schmidt’s Commerce Fantasy
- Soft Computing
- Eric Schmidt’s Commerce Fantasy
- Multi-agent system: http://en.wikipedia.org/wiki/Multi-agent_system
- Mobile Agent: http://en.wikipedia.org/wiki/Mobile_agent
- Subsumption architecture
- PERCEPTION-BASED INTELLIGENT DECISION SYSTEMS
- “The Rise of Cloud Agents“, http://www.readwriteweb.com/archives/the_rise_of_cloud_agents.php
- “Toward a Consumer Product Information Resource“,
- David Reilly. “Mobile Agents – Process migration and its implications“, http://www.davidreilly.com/topics/software_agents/mobile_agents/
- “Fundamentals of Multiagent Systems“, http://jmvidal.cse.sc.edu/papers/mas.pdf
- Software Agent: http://en.wikipedia.org/wiki/Software_agent
- “Recommender System”, http://en.wikipedia.org/wiki/Recommender_system
- “Methodology Tree for Forcasting”, http://www.forecastingprinciples.com/index.php?option=com_content&task=view&id=16&Itemid=16
- FIPA: The Foundation for Intelligent Physical Agents. http://www.fipa.org/
- The Intelligent Software Agents Lab: http://www.cs.cmu.edu/~softagents/
- The Berkeley Initiative in Soft Computing
- “Networks, Crowds, and Markets: Reasoning About a Highly Connected World“, http://www.cs.cornell.edu/home/kleinber/networks-book/
- “Recommender Systems”, Resnick, P., Varian, H. R.; Communications of the ACM, March 1997/Vol. 40, No. 3. http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.151.6578
- Case-Based Reasoning: Experiences, Lessons, and Future Directions. Menlo Park: AAAI Press/MIT Press, 1996. Leake, D., ed..
- “Wireless Ad Hoc Networks”, Zygmunt J. Haas and Jing Deng and Ben Liang and Panagiotis Papadimitratos and S. Sajama, 2002. http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.13.8133
- Stan Franklin , Art Graesser. “Is it an Agent, or just a Program?: A Taxonomy for Autonomous Agents“, 1996, http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.52.1255
- Semantic Web, http://en.wikipedia.org/wiki/Semantic_Web
- ConceptNet: http://csc.media.mit.edu/conceptnet
- “Siri: A Virtual Personal Assistant”, http://tomgruber.org/writing/semtech09.htm
- Software Agents Group: http://agents.media.mit.edu/projects.html