7: People wanted it hushed-up

AI in the Battlespace: Chapter 7

7: People wanted it hushed-up

Colonel Liu takes the final session himself and starts with an anecdote.

“Let me tell you about a wargame called ‘Millennium Challenge 2002’. With 13,500 participants and live plus simulated training sites, it featured a simulated Iran (the Red Team) against ourselves, the Blue Team. We were using advanced weapons not planned for real deployment until at least five years later.

“In the stories afterwards the Red Team, Iran, was said to have used an advanced AI which thought outside the box. This wasn’t actually true. The strategist and leader was Lt. Gen. Paul Van Riper, a Marine Corps veteran with 41 years experience. Once battle had commenced, the Blue Team issued a surrender ultimatum. Van Riper turned them down. Since the Bush Doctrine of that period included preemptive strikes, Van Riper knew that Blue Forces would now be coming to get him.”

Liu looks out from the lectern. Most of his audience will not have heard of this war-game and will be asking themselves: how could we not win against Iran?

“As soon as our carrier group had advanced beyond the point of no return, Van Riper hit them hard. Missiles from land-based units, civilian watercraft and low-flying planes tore through the fleet. Explosive-rigged speedboats decimated the Navy using suicide tactics. It was a saturation-attack which overwhelmed our defenses. Within ten minutes Van Riper was victorious. Nineteen US ships and 20,000 U.S. troops ended up at the bottom of the Persian Gulf.”

The audience is stunned.

“The exercise was rerun,” says Liu, “and the Blue Forces respawned. Van Riper was forbidden to  counter the subsequent Blue landing. He was denied radar and not allowed to shoot down incoming aircraft. The exercise was scripted to let Blue land and win. Van Riper walked out when he realized his commands were being ignored.”

He flips a new slide on to the screen. It shows a board position - mid-play - for the Chinese game of Go. The legend says: AlphaGo vs world champion Ke Jie.

“After the AlphaGo matches and the later AlphaZero successes in chess, Demis Hassabis, the founder and CEO of DeepMind, said they had peeked into the hidden layers of AlphaZero. They wanted to analyse the high-level game-concepts it had synthesised. He said DeepMind's program had developed game-plays which no human had ever conceptualised during thousands of years of study.”

He pauses, looks at the audience with a very straight face.

“You can imagine how interesting we found that. But the next steps were very far from plain sailing.”

Liu explains that some tech companies had scruples about working for the military - but were far from having a monopoly on AI talent. Cyber Command's own version of AlphaZero was tasked with defeating a hostile carrier battle group. It was told to be less aggressive than the traditional doctrine of area-denial using ICBMs with thermonuclear warheads.

“We gave the program new tools: opportunities with subsea-oil containers and submerged air-vents. It’s interesting what happens to a carrier when the seawater it’s floating in halves in density, or when the ocean around it turns to fire. Things do not end well.”

People are working through scenarios: giant steel warships falling into the abyss; steel and flesh vaporising in a two-thousand-degree inferno.

“So here's the take-home from our symposium today. The next-generation AI application is not the autonomous land-warrior. That’s way too hard. Look instead to creative AI battle-programs planning and directing theater operations better than any human being could. Thanks for attending and be reassured: we’ve got the best tech in the world and we’re on the case.”

---

The audience left feeling that this had been one of the better Pentagon one-day symposia. It had been interesting and had made a lot of sense. Now they knew where the serious military R&D was going they felt comfortable. There would be useful tech coming down the tracks: serious, world-beating force-amplification in troubled times.

---

Next     Previous.