AI's Achilles' heel: Pentagon developing tech to avoid rogue machines, slaughterbots in battle

Amid concern that autonomous AI-driven 'slaughterbots' or 'killer robots' can go rogue on the battlefield, the US Pentagon is actively working to address vulnerabilities in its Artificial Intelligence systems.

As per the Defense Advanced Research Projects Agency (DARPA), AI systems could be exploited by attackers using visual tricks or manipulated signals.

Researchers have identified that adversarial AI can be tricked by tiny alterations to its Machine learning (ML) inputs. For example, using tricks and manipulation of data, AI, which is devoid of human discretion, can be fooled into thinking anything is a tank and launch an attack.

Such manipulation can potentially lead to disastrous results on the battlefield.

Guaranteeing AI Robustness Against Deception (GARD)

Pentagon's answer to such a scenario is its Guaranteeing AI Robustness Against Deception or GARD research programme.

"GARD seeks to establish theoretical ML system foundations to identify system vulnerabilities, characterize properties that will enhance system robustness, and encourage the creation of effective defences," states the DARPA website.

"GARD seeks to develop defences capable of defending against broad categories of attacks," it adds.

DrHava Siegelmann, programmanager in DARPA's Information Innovation Office (I2O) explained, "There is a critical need for ML defence as the technology is increasingly incorporated into some of our most critical infrastructure."

"The GARD programme seeks to prevent the chaos that could ensue in the near future when attack methodologies, now in their infancy, have matured to a more destructive level. We must ensure ML is safe and incapable of being deceived," he added.

How will GARDaccomplish this?

As per the DARPA website, GARD's novel response to adversarial AI will focus on three main objectives:

1) The development of theoretical foundations for defensible ML and a lexicon of new defence mechanisms based on them.

2) The creation and testing of defensible systems in a diverse range of settings.

3) The construction of a new test bed for characterising ML defensibility relative to threat scenarios. Through these interdependent programelements, GARD aims to create deception-resistant ML technologies with stringent criteria for evaluating their robustness.

(With inputs from agencies)