American artificial intelligence has matured on warplane ships, raising concerns about its direction

Welcome to forward and share to Moments and WeChat groups! Pay attention, don't get lost, you can find it next time...

Currently, AI is at its best when it comes to working with machines and objects. Experts say attempts to understand human activity will inevitably progress.

Reuters Release time: May 18, 2023 8:52 p.m

The U.S. Navy's first fleet of unmanned surveillance vessels deployed is now operating in the Middle East, with AI programs deciding which targets to investigate and which data to send back for analysis.

U.S. Navy Task Force 59 surface vessels use artificial intelligence for "computer vision" to interpret what ships "see," detect unusual behavior on other vessels, and perform basic command and control functions on smaller vessels.

Their tasks include detecting Iran's arms shipments and other regional activities and sending data back to human operators who decide on operations.

What kind of data is sent back to those analysts is determined by the autonomous ship's own computer system. This is a way to save sending excessive data.

It represents another new use of "edge computing," referring to providing cutting-edge processing where it's needed, with limited or no data links elsewhere, to protect security, limit costs, and address other practical limitations.

Systems like Amazon Web Services' "Snowcone" can securely store large amounts of data and AI programs to analyze that data, and can be taken almost anywhere, including last year's International Space Station and on-site military operations.

"It's really exciting to see these algorithmic programs... Use in our region," Skyler Moore, chief technology officer of U.S. Central Command, said at a recent event at the Center for Strategic and International Studies in Washington.

"We're increasingly recognizing the importance of doing this type of technology adoption work... In a real-time environment with real-time data. ”

Unmanned surface vessels include those made by California-based Saildrone, which makes small boats powered by solar and wind energy, which are also used in marine and fisheries surveys. In the Middle East, the U.S. Navy has taught them to recognize local vessels, such as merchant ships, and mark them when their appearance or pattern of movement changes significantly. Central Command says the result should ultimately be a significant increase in the productivity of their human analysts.

This technique has been talked about for many years and is now seriously supporting the work.

Gen. Bryan Fenton, commander of U.S. Special Operations Command, told attendees this month at SOF Week, a trade fair at SOCOM's home in Tampa, Florida, that his order was to "harness big data like never before."

"Big data... We will need big data to gain a decisive advantage. Its uses range from tracking vehicle maintenance and equipment and personnel readiness to helping commanders make critical decisions.

In September 2021, the U.S. Air Force revealed for the first time that in addition to its use in exercises, the analysis of AI programs is being used for on-site target decision-making in active conflicts, albeit only in combination with human analysts. It does not give details of the work.

As AI becomes more widely used, including the military and others, some believe it will become more controversial. Earlier this month, AI pioneer Geoffrey hinton — who built the world's first neural network in 2012 — left Google so he could freely talk about his concerns about where the technology was headed.

Unmanned fighter

AI-piloted warplanes could be the first automated weapons system authorized to take human lives. In 2020, the U.S. Air Force began pitting real pilots against artificial intelligence programs on simulators.

Last December, those tests entered the real world, with two different AI programs flying a real F-16 on American soil against a human-piloted aircraft.

The Defense Advanced Research Projects Agency (DARPA), which conducted the test, has not revealed whether the robot pilots are superior to their human competitors — though it does say that a human pilot was on the plane during the test in case anything goes wrong.

However, it already seems possible that AI will surpass equivalent drones remotely controlled by human pilots without the second or more time delays required for the transmission of human instructions.

Allowing robot pilots to kill in air combat would violate current U.S. government policy, and the U.S. State Department issued a note this year reaffirming its commitment to "having a responsible human chain of command and control."

The use of this technique for military means has always been controversial. In 2018, Google announced it would not renew its contract with the Pentagon for Project Maven, a Defense Department that uses artificial intelligence to provide better targeting, particularly drones. About 4,000 employees signed a petition asking the company to sign a "clear policy stating that neither Google nor its contractors will use the technology to make war."

Project Maven continues. It was transferred last year to the National Geospatial-Intelligence Agency, which processes U.S. satellite imagery, and is now trying to gather material from a plethora of sources that would outnumber any team of human analysts.

Get the right data

Most short-term development has focused on detection, surveillance, and warspace management, as well as planned maintenance—anything that requires insightful massive amounts of data that human analysts can never match. In 2024, the Pentagon has already allocated $1.8 billion for AI and machine learning, with the bulk of that — $1.4 billion — allocated to the Joint Global Command and Control Program to better connect military land, sea, air, and space sensors and weapons.

Getting AI to work may be easier in some areas than in others. Air defense to identify and shoot down enemy aircraft and missiles has long been seen as a possible focus, a goal further accelerated by Russian strikes against Ukraine and the success of Israel's Iron Dome defense system against rockets launched from Palestinian territories. The same goes for submarine detection.

Currently, AI is at its best when it comes to handling machines and recognizing objects. Experts say attempts to understand human activity will inevitably accelerate.

AI experts say that the quality of datasets is often an issue, as is the correct judgment of design.

"Over-trust in results and misinterpreted algorithms can lead to danger," said a RAND Corporation study last year that drew on experience from other industries, such as health care.

So far, that means the Army may be slightly slower to make the technology work than their Air Force, Navy, space or special operations forces. U.S. Army commanders announced earlier this year that their annual Project Convergence AI exercise, which typically takes place in the fall, would be postponed until February to better integrate new technologies and foreign partners.

Other countries and actors may cut directly into real-world use more quickly. A 2021 U.N. report accused the Turkish-backed Libyan government of allowing Turkey's Kargu "suicide" quadcopter to choose its own targets as troops loyal to Libyan warlord General Khalifa Haftar fled a city.

It is not clear whether there were any fatalities or how autonomous the drone operated.

"The pursuit of AI autonomous weapon systems without binding legal rules that clearly address the dangers is a disaster," Mary Wareham, arms control director at Human Rights Watch, said in February.

"National policies and legislation are urgently needed to address the risks and challenges posed by the removal of human control over the use of force."

Peter Apps is a Reuters columnist who specializes in defense and security issues.

——WeChat public account Bangladeshi Chinese information Original translation, please indicate the source for reprinting