This post regards the integration of intelligent, or at least relatively intelligent robots, into society focusing on two different views: the “empowerment” view, and the “motive” view. I propose three solutions that both confirm and balance these two views in this post and talk a little about the goal of robotics.

Anyone familiar with robotics is familiar with Asimov’s (somewhat naive but still relevant) rules of robotics:

(1) A robot may not injure a human being or, through inaction, allow a human being to come to harm.

(2) A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.

(3) A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.

The obvious, glaring flaw seen in these rules is that the robots have an incentive to minimize human suffering, and situations in which there are difficult decisions to be made - as those will inevitably be resolved imperfectly - but comparatively little impetus for empowering humans, by, for example, unlocking a car quickly to allow a human to get to work on time.

So the robotic police state from “I, Robot” could potentially be avoided if we focus on the empowerment thing, at least according to these people.

Basically, they seem to be arguing that we should focus on a proliferation of robot behaviors specialized in natural ways to empower humans that doesn’t include some theoretical AI programming specific to each situation or restrictions on behavior; we just do more leg work teaching robots how to model the real world - mostly. They call it “empowerment”.

There seems to me to be two conditions that need to be satisfied for a robot to assist a human: the robot must both know the conditions under which the human is assisted (a directive), and how to actually accomplish this directive.

Accomplishing the directive seems to actually be the easier part and the meatier part: the more efficiently robots can help in immediate circumstances based on simple directives, the more there can be an openness to said directives, which can be narrowed as situations develop and variables present/become clearer/become more easily manipulated. Accomplishing the directive is merely a question of how according to how the robot is programmed and does not address human motive.

However, some argue that robots do indeed need to understand motive like humans do to be effective and safe: the robot might complete a task correctly according to its own understanding of a task, but the way it is completed might be a failure by any other standard. The following example is given in the article:

“Imagine asking a robot to pass you a screwdriver in a workshop. Based on current conventions the best way for a robot to pick up the tool is by the handle. Unfortunately, that could mean that a hugely powerful machine then thrusts a potentially lethal blade towards you, at speed. Instead, the robot needs to know what the end goal is, i.e., to pass the screwdriver safely to its human colleague, in order to rethink its actions.”

Bringing this back to the directive/process of completion of a directive format of looking at it: It seems to me there are a few solutions:

(1) We create an army of specialized robots adept at modeling the real world to do simple tasks without adapting to human motives, and instead train people to not get jabbed in the jugular with screwdrivers by more advanced robots completing more advanced tasks; we focus on training a small number of specialists to work with a small number of more advanced robots for the jobs that cannot be done by “dumb” robots, all the while improving said advanced robots.

(2) We create a new, smaller number of less specialized robots that operate more dynamically with regards to directives and how they change in the presence of human motives to replace the less safe, more advanced robots in (1) with their specialists, but delegate much of the simple stuff to the “dumb” robots described in (1) nonetheless. The flexibility is developed through some relatively efficient AI programming of minimal intensity.

(3) We spend a lot of time looking for some infallible rules to govern the behavior of advanced robots carrying out a multitude of tasks without any glaring inconsistencies/contradictions/weak points or repugnant/harmful implications. If these robots are as intelligent as people, or more intelligent, and are permitted to create their own directives, all the better - so long as they are programmed to behave in the best interests of humanity and there is total transparency.

I posit that (1) and (2) use up a similar amount of resources, and could be steppingstones, in order, to (3), which is ideal imo, even if it poses some problems. This train isn’t stopping.

While I strongly believe you are what you are programmed to be, and there is no reason to believe a robot would be cruel or evil, (3) is a little scary. What principles and behavior would be acceptable to people at large? Surely not the overly simple, pseudo-consequentialist rules Asimov puts forward.

There is a philosophical problem here independent of any specialized knowledge of machine learning/computer science/mathematics/programming, and it is the problem of establishing a clearer epistemology around robotics that empowers the specialists to re-evaluate the problem on a more fundamental level.

I don’t know exactly what that epistemology is, but it begins with establishing this: what is the ideal outcome we are looking for in robotics? Is it efficiency? Relief from hard labor? Profit?

I would say there is one first, best outcome, and it is to gradually delegate as much power to autonomous machines as possible without a net depreciation in quality of life for humans. This makes sense given that that is basically what autonomous machines are made for. This includes going from (1) to (3) gradually. Further discussion of why I think this is expected; I am attempting to be concise here.