DARPA Seeks an Algorithmic ‘Theory of Mind’ to Predict, Manipulate Future Behaviors

by Tim Hinchcliffe | The Sociable

DARPA is looking to predict, incentivize, and deter the future behaviors of the Pentagon’s adversaries by developing an algorithmic “theory of mind.”

“The goal of an upcoming program will be to develop an algorithmic theory of mind to model adversaries’ situational awareness and predict future behavior”

DARPA, “Theory of Mind” Special Notice, December 2024

The US Defense Advanced Research Projects Agency (DARPA) is putting together a research program called “Theory of Mind” with the goal of developing “new capabilities to enable national security decisionmakers to optimize strategies for deterring or incentivizing actions by adversaries,” according to a very brief special announcement.

DARPA says, “The program will seek to combine algorithms with human expertise to explore, in a modeling and simulation environment, potential courses of action in national security scenarios with far greater breadth and efficiency than is currently possible.

“This would provide decisionmakers with more options for incentive frameworks while preventing unwanted escalation.”

“The program will seek not only to understand an actor’s current strategy but also to find a decomposed version of the strategy into relevant basis vectors to track strategy changes under non-stationary assumptions”

DARPA, “Theory of Mind” Special Notice, December 2024

The author of the Theory of Mind special notice is Eric Davis, who joined DARPA in February, 2024.

Previously, Davis was the principal scientist of artificial intelligence, machine learning, and human-machine teaming at Galois, a tech R&D company whose clients include DARPA, the US Intelligence Community, and NASA, and whose partners include the Bill and Melinda Gates Foundation.

The Theory of Mind special notice does not mention who the adversaries are, but once such an algorithm is developed, there’d be no putting this genie back in the bottle.

The 2017 edition of the “DoD Dictionary of Military and Associated Terms” defines “adversary” as “a party acknowledged as potentially hostile to a friendly party and against which the use of force may be envisaged.”

The dictionary does not define the word “party,” so in this vague concept, an adversary could be anyone perceived to be “potentially hostile” and therefore worthy of the use of force.

Going back decades, DARPA has sought to monitor, predict, and modify human behavior by collecting and analyzing as much information on people as possible.

“The goal of the Total Information Awareness (TIA) program is to revolutionize the ability of the United States to detect, classify and identify foreign terrorists – and decipher their plans – and thereby enable the US to take timely action to successfully preempt and defeat terrorist acts”

DARPA, Total Information Awareness (TIA), July 2002

Source: DARPA TIA program (archived)

Following the attacks on September 11, 2001 DARPA announced its now-defunct (or potentially splintered) “Total Information Awareness (TIA) program in July, 2002 “to revolutionize the ability of the United States to detect, classify and identify foreign terrorists – and decipher their plans – and thereby enable the US to take timely action to successfully preempt and defeat terrorist acts.” Read Full Article >