Imagine the applications and implications of robotics and artificial intelligence in 2045

Follow Me! Collecting data from humans conducting training to develop algorithms that approximate human judgments

Describe your idea

This idea is the feeder for the Genetic Algorithms and Simulation Environments Workshop Idea. In short, we may be able to code algorithms for many human judgments and decisions by looking at training scenarios, focusing on the heuristics used by humans and the environmental cues that drove those heuristics. This approach for encoding judgments and decisions into machine learning relies on the 'fast-and-frugal' model of judgment and decision making proposed by Gerd Gigenrenzer.

Top improvement

1 year ago
jah27 said
0
Open Close
This approach is premised on the idea that humans, [...]
This approach is premised on the idea that humans, who have been fighting as groups for thousands of years, have evolved to the point where we are pretty good at conducting warfare (though not necessarily war). With this in mind, perhaps we should leverage this experience, which is coded in a variety of ways: flight-or-fight instincts, social norms, doctrine, tactics, etc. The key to all of this is identifying the environments for which the heuristics work...and don't.
Add Improvement

What makes this idea new or different?

Gigenrenzer argues that humans largely rely on heuristics to form many judgments and make many decisions. Some heuristics are "hard-wired" into our evolved psychologies (fight-or-flight), and others are learned (social conventions). Gigenrenzer proposes that judging the appropriateness of the judgments and decisions calls for looking at the "ecological fitness" of the heuristics and the environment in which they were formed, then comparing this to the actual environment.

Top improvement

1 year ago
smckay said
0
Open Close
Collecting data on human behavior in training exer [...]
Collecting data on human behavior in training exercises will be a good start, but most likely not sufficient. Other techniques and methods will need to augment the potential limitations. One way to augment this so it is more applicable is to incorporate robustness theory. Though the data collected or as the other committer suggested, heuristics, will be powerful in developing these algorithms, they will still be subject to error that may negate their usefulness, especially in the initial developments. Robustness theory acknowledges the errors and helps identify a 'robust' solution within the error bands.

Second, this data collection effort is only targeting one set of players. We live in a multiplayer domain. Other data sources will need to be identified to create behavioral models of players that we cannot collect data directly from (e.g., our adversaries or competitors).

See: http://onlinelibrary.wiley.com/doi/10.1002/nav.20440/full
Add Improvement

What will be the implications of this idea?

Gigenrenzer argues that this approach to judgment is decision making is the best at handling complex and uncertain environments, where outcomes cannot be optimized.

Top improvement

1 year ago
Echo said
0
Open Close
The value of a good human model might be more than [...]
The value of a good human model might be more than finding a machine that can solve problems as well as a human. If you have a good enough human model, you could use adaptive evolutionary algorithms until you develop a set of heuristics that can consistently achieve your success criteria for the given scenario.
Add Improvement

What other players said

1 year ago - jah27 said
What will be the implications of this idea?
0
Open Close
Using a broad range of human trials gathered durin [...]
Using a broad range of human trials gathered during training events, developers could rapidly evolve the genetic algorithms by focusing on isolating the human 'search' rules, the 'stopping' rules, and the decision rules humans used.
1 year ago - jah27 said
Describe your idea
0
Open Close
This approach to understanding judgment and decisi [...]
This approach to understanding judgment and decision making appears to be the best fit for uncertain and complex environments, which are characteristics of the environment in which we conduct military operations.
1 year ago - jah27 said
What will be the implications of this idea?
0
Open Close
The human trials may also help identify alternativ [...]
The human trials may also help identify alternative environmental cues that drive better judgments.
1 year ago - jah27 said
Describe your idea
0
Open Close
This approach is premised on the idea that humans, [...]
This approach is premised on the idea that humans, who have been fighting as groups for thousands of years, have evolved to the point where we are pretty good at conducting warfare (though not necessarily war). With this in mind, perhaps we should leverage this experience, which is coded in a variety of ways: flight-or-fight instincts, social norms, doctrine, tactics, etc. The key to all of this is identifying the environments for which the heuristics work...and don't.
1 year ago - Echo said
What will be the implications of this idea?
0
Open Close
The value of a good human model might be more than [...]
The value of a good human model might be more than finding a machine that can solve problems as well as a human. If you have a good enough human model, you could use adaptive evolutionary algorithms until you develop a set of heuristics that can consistently achieve your success criteria for the given scenario.
1 year ago - smckay said
What makes this idea new or different?
0
Open Close
Collecting data on human behavior in training exer [...]
Collecting data on human behavior in training exercises will be a good start, but most likely not sufficient. Other techniques and methods will need to augment the potential limitations. One way to augment this so it is more applicable is to incorporate robustness theory. Though the data collected or as the other committer suggested, heuristics, will be powerful in developing these algorithms, they will still be subject to error that may negate their usefulness, especially in the initial developments. Robustness theory acknowledges the errors and helps identify a 'robust' solution within the error bands.

Second, this data collection effort is only targeting one set of players. We live in a multiplayer domain. Other data sources will need to be identified to create behavioral models of players that we cannot collect data directly from (e.g., our adversaries or competitors).

See: http://onlinelibrary.wiley.com/doi/10.1002/nav.20440/full

Advanced to the Marketplace.

View this idea
1

When will this idea make a difference?

Your vote Average vote Total votes
Now
2030
2040
2050
2060+

Inspired by

Genetic Algorithms and Simulation Environments for Optimizing Tactical Behavior and Team Composition
View Idea
© SciTech Futures 2017. All Rights Reserved.

Site Login