Thread by @RikeFranke

The German Army (Amt für Heeresentwicklung, to be exact) has surprised many and published a position paper on #AI for land-based forces. Will translate bits and comment.

Alright, let's do this!
The paper is available here 👇
It was published only some ten days ago, but was finished in August/September.
#AI #KI #Bundeswehr (1/)
augengeradeaus.net/wp-content/upl…

The first thing to note is that with regard to process, this paper is a bit weird: Germany has a national #AI strategy (which mentions military AI in *one sentence*) but no military #AI strategy for the whole of the Bundeswehr. However, now there is this paper for the Army.
(2/)

(I wrote about the Germany's national #AI strategy and how it compares to other European strategies here: ecfr.eu/publications/s… )
(2a/)

The paper is a "position paper" (not a strategy), and was written by the "Amt fuer Heeresentwicklung", whose job it is to develop new concepts and ideas for the Army. "This position paper is the basis for the goal-oriented use of #AI in the ground forces". (3/)

Interestingly, it is *one of three* papers - there will be one on #robotics and autonomous systems (yey!) and one on "digitisation of ground forces". So more to follow! (4/)

The paper starts with a scenario in which the army is securing a critical infrastructure with the help of four platoons of 5000 small #drones each, of different sizes, some of which armed. These drones form #swarms (with a extension of "up to two km").
(5/)

In the scenario, the #drone swarms are released from aircraft and attack enemy drones electronically and kinetically. To me, the way it's described sounds like a flying minefield (not directed against people though). (6/)

The scenario is reasonable enough (though the #drone-against-drone scenario seems very specific). It's interesting though that it's so focused on #droneswarms, as there are many other #AI applications. (Some mentioning of data collection and analysis though.) (7/)

After the scenario, now we're in the paper itself. It starts with a section on the framework which describes the future of combat. Most important, according to the authors: increased speed, and although more data and more surveillance not necessarily more information. (8)

This means: "In such intense combat phases, a human can no longer effectively dispose of the means of action available to him and assign individual goals. The control of the defense must therefore be done by #AI. The troop leader still releases the effect." (9/)

And: "No matter how well trained, staffs will not be in a position to master and evaluate the accumulated flood of information without #AI support. Big data-based methods [...] are a mandatory prerequisite for a well-founded and sufficiently fast staff work." (10)

Authors argue for #AI-supported human decisionmaking. Fine, but I am unconvinced by👇which suggests that there are only some things that #AI can do.
As this is future-focused, may have been worth acknowledging that AI may be better in most, if not all things, and what that means.

The paper is rather clear on the element of human control: "Man must retain the power of decision over life and death. The principle of effective human control applies."

Very interesting is this: "..new technologies can help to improve the enforcement of international humanitarian law, in particular the protection of the civilian population in armed conflicts, for example, by increasing the precision and scalability of the weapon's effect..."

Not the first time we've seen this '#AI can improve compliance with IHL' argument - see the French military strategy (which is more explicit)

https://twitter.com/RikeFranke/status/1174377152260792322

Unsurprisingly, the paper has a section on LAWS ("#Killerrobots"). It differentiates between automated, if-then systems, and autonomous machine-learning #ML based systems. I actually don't find this too bad, but know that others, such as @drfranksauer aren't happy with this.
(15)

In any case, most important on this is the following: "In addition to political and legal reasons, the use of future conceivable LAWS is also not wanted from a military point of view."
However, as opponents may develop them, defensive systems will need to be developed.
(16)

Next is a section on drivers of this development, which is quite good, but mostly stuff people are likely to know already, so skip. (17)

More important: the areas of action for army development.
The paper describes four:
1) Further development of existing systems, such as improving ISR with image recognition.

2) New weapon systems development. This I found surprisingly weak as it solely mentions small #drones. 🤷‍♀️ (19/)

3) Using of #AI in HR and logistics, such as predictive maintenance,
4) AI in training and education
(20)

Finally, the paper proposes a whole range of new units and working groups. The problem is that as this is only a position paper of one branch, it's only suggestions, not actual plans. (21)

Interestingly, the most interesting part of the paper, in my view, is the annex! Because that's where all of this becomes a bit more concrete.
Example: #AI-enabled image analysis - the paper recommends the establishment of a database.
(22)

On small #drones, the paper is walking a *really thin line* with regard to LAWS/#killerrobots: It speaks of AI-enabled drones which can take on military missions *autonomously* - such as reconnaissance, but also "targeted effect", which I understand as kinetic action.

On swarms of small #drones, the paper makes a few good proposals, such as the development of such systems and how to use them best (important!). (24)

And here they finally also mention other #AI-enabled applications, such as battle management. (25)

So, what to make of all this?
As with the French military #AI strategy, first of all, a 👍 for publishing this at all! The German MoD is notorious for just not publishing such papers, so I was positively surprised.

That being said, it's not the German military AI strategy but one position paper by one branch, which is
a) not as impactful, and
b) slightly weird in terms of process.
I hope there will be more.

Re the content: the focus on #drone swarms surprised me a bit, as there are loads of other #AI-enabled applications (some of which are mentioned in the annex).

The analysis of the future of combat, which focused on "Hyperwar", a very fast-changing battlefield and different means being used is convincing and shows that the right questions are being asked.

That's all from me on this for now.
If the topic interests you, have a listen to our latest @Sicherheitspod episode in which we discuss this paper and the topic of #AI in warfare more broadly (from min 40:09)
soundcloud.com/sicherheitshal…

@Sicherheitspod Maybe of interest to @mchorowitz @EBKania @jdomerchet @jeangene_vilmer @jchnoel @JackieGSchneid