Episode 2165 - June 27 - Tiếng Anh - Bản Tin Công nghệ Thông tin – Ngày 23 tháng 6, 2024 - Vina Technology at AI time - Lê Quang Văn | Podcast - Nhac.vn

I.T. News – June 24, 2024

1 - Two startups join forces to make self-flying tankers, dogfighting AI, and more

Merlin Labs, which aims to test an AI-powered KC-135 within a year, is to buy F-16 AI driver EpiSci.

By Audrey Decker. Defense One. June 21, 2024.

One startup has Air Force contracts to rig a KC-135 tanker and C-130 airlifter for autonomous flight; the other created the AI pilot moving on in a DARPA dogfighting program. Now they’re joining forces.

Last week, the Boston-based Merlin Labs announced that it would acquire EpiSci, a California-based firm whose AI helped take Air Force Secretary Frank Kendall on a high-profile spin in an F-16 last month.

EpiSci, and its sub, PhysicsAI, are the only teams that have been a part of all three chapters of DARPA’s effort to develop an AI fighter pilot: the 2020 AlphaDogfight Trials, the 2020-24 Air Combat Evolution Program, and—as of last week—the Artificial Intelligence Reinforcements program. Of the four companies whose AI agents operated on Kendall’s flight, EpiSci CEO Bo Ryu said, his company’s was the “most trusted and best performing.”

Merlin, which makes an AI pilot of its own, expects to test it aboard an Air Force KC-135 tanker within the next year, said CEO Matt George. This “core” pilot could allow the Air Force to reduce the number of pilots on its mobility aircraft from two to one, freeing up a bunch of aircrew—and enabling the Air Force to watch the AI operate during real-world missions with a human next to it, he said.

If the EpiSci deal closes as expected in the next two to three months, Merlin will be able to add advanced skills, such as F-16 qualifications, to its core pilot. That would make the company the first to offer an AI pilot that can fly various types of missions, George said.

In February, Merlin signed a deal with the Air Force to demo its AI pilot on a KC-135; just last week, the company won a $105 million contract to bring that pilot onto C-130 transport aircraft. The company will start its KC-135 work over the next couple of months and fly the aircraft with the autonomous system in the next year, George said. After that, it will move on to C-130 tests.

2 - Artificial Intelligence at Merlin

Alex Naiman, CTO. Merlin Labs. June 6, 2024.

AI and Aviation Systems

Artificial intelligence (AI) innovations dominate technology headlines today, as breakthroughs in software leveraging large language models (LLM) surprise the world with apparent reasoning skills. Still, challenges remain before claiming that artificial general intelligence has the ability to rival human thinking. What is clear is that while AI can and will do amazing things, the technology is still probably at its initial surge on the Gartner Hype Cycle.

As an industry, aviation tends to be conservative in adopting new technology. The adoption of new technology requires meeting the high bars that make air transport safe, such as providing strong evidence that the technology will work as expected and when needed. Certifying systems to aviation standards thus requires substantial up-front investment, so the new technology must provide a clear value to operators before they will adopt it. Showing clear value and providing a way to prove that it is safe are what we must do when taking a technology as novel and dynamic as AI and introducing it into an aviation system.

How Merlin Uses AI

Merlin is building autonomy for aviation, and it’s important to state as a first principle: autonomy and AI are not synonymous. Many of the functions needed to fly an aircraft autonomously can and should be built using “traditional” software. We see AI as a tool in the toolbox of software design components and we use it only where it’s the best tool for the job.

There are two main areas where we currently believe the set of technology loosely called “AI” can benefit the Merlin Pilot that we are building

Perception

Software that uses convolutional neural networks has proven to exceed any other methods for certain kinds