Intergalactic Summit

Other than the capsuler themselves.

Technology in New Eden, especially ship technology, continues to rapidly improve. Improved armor and modules, as well as entirely new hull designs roll out every year. Yet, since the beginning of the Empyrean Age, I can think of no appreciable technology to reduce the crew size of capsuler controlled vessels.

I find this disheartening. Every year, millions (billions?) of people serve and die in our ships. Only frigate class hulls are yet designed for autonomous capsuleer control, and even this is only if you avoid modules or the specialized platforms.

Alternatively, there are capital ships that I am lead to believe are wholly controlled by AI. Ships that capsuleers destroy, salvage, and analyze constantly.

How is this possible?

I am not asking for crew-less Titans here. That would be ridiculous. Still, imagine the number of lives not lost if cruiser class hulls could be controlled by only a capsuleer or battlecruisers crews could be halved.

This is one area where I believe the rest of New Eden might look to the Gallente for guidance. In regards to automated systems, their technology is superior to that of any other faction. As a start, we might incorporate their designs into non-Gallente vessels, and perhaps instead of pouring billions of ISK into R&D for shields that have 5% more resistance against EM when overheated while the pilot hums 'God Save the Empress,' we see if there's way to expand and improve upon these systems.

There have been intrepid capsuleers working on this issue for quite sometime, myself included! I've got to bounce some ideas with a lot of folks who were concerned about the average crewmember life expectancy and their solutions to the problem.

To me, one of the biggest hurdles was to get past CONCORDS ban on Strong AI research. Which makes most of my research borderline illegal. But weak AI is too limited for the quick decision making needed for a crew, you see. My solution was to deploy jury rigged medical caretaker drones to act as a surrogate crew, and use a central operating system to coordinate them in the ship.

The research is there but it is highly illegal and worth several times more the price of the crew it is replacing. People have been using crew alternatives despite the ridicule and I'm sure with enough research we could make this practice common place.

Because of the rapid development in various fields, both software and hardware applications responsible for managing and maintaining ship subsystems have become increasingly complex. Capsuleer or not, a single person's brain simply can not comprehend the amounts of data being processed by those systems each second.

Of course we could consider various AI or sub-AI options to automate parts of those tasks, but, as Hydreiga-san suggested, reaching a solution would probably require either surpassing or violating appropriate CONCORD directives.

Ah! Makkal-haani, you are filled to brimming with wonderful questions today. I'm glad you're back :) Big questions, though. There's a lot to unpack here. These questions are right in my field, so I will answer to the best of my abilities!

First, CONCORD interference. There are injunctions against the development of a lot of AI systems, especially on the general intelligence problem - the so-called "Strong AI" problem. Which, well - we really don't need Strong AI to do the work of a ship. It's a fairly narrow field of intelligence, so a specialized intelligence system is more than sufficient!

A lot of that is due to how most companies develop and research their AIs, really! Most follow the Creodron model of network development. This involves developing a multi-layer decision network with appropriate inputs and outputs, placing the network into a simulation of the environment in which it is expected to work, and iterating with evolutionary selection until a network emerges which can sufficiently solve the problem. The company can then copy the network and use it as the decision engine for the system in question. It's fast, relatively cheap to do, and can be surprisingly flexible. Most drones we see today are developed in this model, as an example.

The problem with this is model is that it's only as good as the simulation in which the network is trained. No simulation can perfectly mirror the real world, or even more than a small slice of it - at least not within a realistic budget or timeframe. So these networks are trained on approximations of the scenarios they face when in service. As such, their behaviour fails at those points where the approximation and reality don't meet. When you couple this with self-learning systems you can get very strange results, well outside of expectations. An example of these are rogue drone hives.

Another way of developing decision networks, and the one which we follow at Lai Dai Research Biotechnology and Cybernetics, is much more precise. Instead of relying on black-box-magic to build our networks we do the hard work of deconstructing problems to their logical parts and building specialized architectures to solve those problems. It's much more time consuming, much more expensive, and much slower to market, but tends to minimize the unexpected behaviours. It's not deeply competitive on the general market, though - we mostly put our work into implants and certain high-importance modules within larger systems.

To your last question - a brain isn't truly "hijackable", it isn't a generalized processing substrate. A Capsuleers' implants generally are, though, as are many other implants, such as the TCMC.

The central question, of why we don't see more automated crew, and whether it's possible. It's certainly possible to automate a larger ship entirely. Like I said, the role of crewmen on a ship are fairly narrow problems. They require intelligence within that field, but they don't need an enormous degree of flexibility. It's mostly laws that prevent their deployment, to be honest! Laws and a good dose of plain old human fear. If more research and development broke away from the Creodron model we might see more crew automation and reduce the staggering casualties. Given the Gallentine grip on that particular field, though, and the very real capitalist concerns for any company making the switch, I don't think that will be likely for some time.

As an aside, LDRBC is pleased to offer a number of crew and worker automation system solutions for tasks such as those noted; please feel free to contact me with inquiries and a quote.

- S

I must say I am truly intrigued by Your methodology dr. Scherezad. I consider my level of expertise in the field as rather high yet I was not aware that Lai Dai's A.I. research is almost entirely based on developing application-specific systems rather then general purpose neural networks.

I would be interested in discussing more details if Your corporate doctrines do not forbid sharing such knowledge.

That will never happen, hence my approach...

In theory each A.I. is able to reach a point of transcendence beyond which it's learning abilities would become limitless (or, using mathematical terms, grow exponentially) thus surpassing Your crew chief in a matter of (consider this a very careful estimate) days.

In practice we are constrained by CONCORD directives brought up earlier which strictly prohibit designing self-aware A.I. constructs.

If such an AI ever came into existence it would determine the only weak point in the ship left - us. We would become obsolete, which we will never allow. CONCORD are absolutely right to restrict this. I'll stick with my crew thanks. Did you know that women with a healthy body fat ratio are best suited to space travel? It almost makes me believe there is a god...

Which is why I do employ expert systems, Schere, I just don't think they can replace humans - especially in combat.

Nevertheless there will always be the (in)famous "what if?" question..

It is true that the merit of creating intelligent, autonomic systems is not granting them the gift of self-awareness but supplying them the ability to learn, adapt and evolve with the problems they face no matter how small and straightforward. The possibility of those systems ascending and gaining one grows in proportion to the level of task complexity and independence we imply on them.

We can always limit that possibility by sandboxing the A.I. with a closed set of variables and/or logic resources but we are still creating an automata, a system that is designed to work stand-alone. There will always be the probability that, as the process of learning progresses with time, instead of following principles, the system will start creating it's own and prioritizing them.

I understand You can be tired of the argument popping up with every occasion (believe me I am to) and please do not feel offended by me contradicting but I do not think we have the comfort of allowing ourselves to underestimate this kind of outcome dr. Scherezad.