I think AI has lots of good practical applications weapon-tech-wise or not here are some scenarios I can see it being used.
Surgery - 9 out of 10 chances are that a human can make a mistake during invasive surgery and it does happen which why medicine is probably the best covered business by insurances. Take the human out of the equation and couple AI with robotics to perform the surgery.
Prosthesis - One of the biggest problems with modern-day prosthetic surgery and replacement is their inability to adapt to situations fast enough and more importantly their inability to adapt to the wearer and his/her needs. This is where AI, once again coupled with bionics/robotics, can start to make a huge difference. Take the simple example of leg replacements and the bionics involved in keeping pressure, stability and precision for things like foot movement. Today, this is achieved by sheer mechanics that doesn't allow room for too much beyond what physics has in store for the wearer. But imagine if the AI was able to work with the prosthetic to offer things that our regular limbs could do like tension management of limbs to cushion falls or automatic prevention of discomfort by administering the drugs based on synapse reading and user monitoring. The applications here are endless and while most of what we read is really geared towards AI being by itself this coupling is where we'll start to see the biggest breakthroughs in the technology.
Medical Care - People who require round the clock medical care generally end up getting a nurse/caregiver to be with them to do things like monitor their well-being, administer drugs, ensure that in the case of an emergency the patient gets the best care possible. But if you had an AI that was capable of doing all this and couple it with the fact that it never goes off or takes a break then you could get these machines to start taking care of people with a better chances of success.
Now for the outlandish stuff
Accident Prevention Systems - Imagine you're in a high speed car accident and need medical care. The chances of an ambulance reaching you via the road is directly dependent on the road being free to allow that response unit to get to you. Furthermore, a response unit is a waiting system, it gets activated when the accident is reported, not when it occurs. If you had a system that monitored highways and acted on the scene to remove victims of the accidents from the scene and even administer medical-aid then you give the response unit a chance to get there and the victims of the crash to survive as well.
Troop Deployment and Enforcement Systems - The biggest problem with deployments in any war is that the core system for deployment isn't making the decisions for safety. Targets aren't really recognizable even if tech is provided because lets face it we're only human and even if a targeting system could find all the targets its not smart enough to make a decision on the level of threat posed or provide instant counter-measures to ensure the deployment block is safe from harm. If the system was able to make those decisions and do any of the following - call for support, neutralize threats, provide auto-shielding then countries and military personnel would be much safer and loss of troops and lives would be reduced. Furthermore, the ability to deploy troops anywhere on a field with the minimum loss of life could be achieved with more combat experience thereby increasing the chances to win.
Education and Learning - This one probably has the highest benefit out of the lot for me. AI could improve learning by catering to students based on evaluation and circumspection of their learning patterns. For e.g. students who excel at Math but tend to do poorly at Literature could be given specialized learning course patterns that enforce the weaker of the two learning curves using historical information from first-contact to goal fruition. Imagine a teacher that is always around to help whenever necessary and has the ability to change its teaching pattern based on your learning skill tree.
Most if not all the stuff I have mentioned isn't really science-fiction. Robotics is progressing at an alarming rate and AI in its infancy is capable of making smart decisions where necessary to re-enforce the physical demands of the robotics. The question isn't really when anymore, the question is who is going to do this. When I look at things like what MIT is doing with prosthesis, B.Dynamics is doing with robotics and now what China has done with an AI that is able to mimic human expression through speech I don't think its too far fetched to think that most of this will come in the next decade ahead or earlier.
There are far more creative things like space exploration, setting up colonies based on habitat analysis and reconstruction, safety management in apartments, crowd control during riots, etc. but those will all come in due course once the systems developed start to learn and improve based on their first-contact to them being able to make decisions on their own.
