The Coming AI Backlash

Council on Foreign Relations | Chris McGuire

Artificial intelligence policy is going to be the central issue in the 2028 U.S. presidential election since rapid technological advances are causing serious disruptions at home. The leading models are expected to be 250 times more powerful by the day of the election, and growing public opposition is being driven by concerns over widespread job losses, the possibility of cyber weaponisation, and national security risks.

The political pressure arises from the fact that federal regulation is almost non-existent, which has caused voters to feel economically at risk and has made them more suspicious of state surveillance. AI is still very unpopular; just 10 per cent of Americans are excited about the technology, and politicians such as Senator Bernie Sanders are calling for immediate moratoriums on data centres. At the same time, the military's use of autonomous systems in Iran has reduced target identification from hours to seconds, showing how important it is to keep ahead of China in terms of technology. In order to avoid a catastrophic situation, future U.S. policy had better both secure global supply chains, put in place safety tests, and redistribute the productivity gains brought about by AI.

Comment

The way that the U.S. Central Command has introduced algorithmic targeting systems into its operations in the Middle East shows a basic change in the speed of the find-fix-track-target-engage-assess process. Because Project Maven has been used to shorten the time taken to identify targets from days to seconds, military commanders are now witnessing an extraordinary speeding up of the decision-making process. The ability to process huge amounts of intelligence data in this way lessens the mental load on the analysts at the Combined Air Operations Centre, but it also causes serious delays when compared with the usual manual procedures for authorising commands.

The main obstacle now lies in target validation. Since the present rules of engagement mandate that human-in-the-loop procedures obtain formal approval from a military commander before ordnance is released from platforms such as the MQ-9 Reaper, the greatest restriction on CENTCOM's operational tempo is no longer the processing speed of Project Maven, but rather the physical capacity of the human operators to check the algorithmic recommendations before weapons are released.

Strategic Question for Discussion

If the operational tempo made possible by Project Maven continues to exceed the limits of human cognitive ability, how is CENTCOM going to keep meaningful human-in-the-loop control without giving up the speed benefits of algorithmic targeting?

The fact is that in order to maintain control it will be necessary to shift human involvement from giving real-time tactical approvals to setting the parameters before the mission and carrying out algorithmic audits. This kind of change means that commanders are likely to take on the role of system supervisors rather than making active decisions during operations. In the end, this development could result in a dependence on automated validation procedures, since the speed of conflicts at the peer level in places such as the Indo-Pacific effectively leads to engagement authority being handed over to the machine.

Share your assessment in the comments below.

To read in details click here