Why Asimov's Laws of Robotics Don't Work - Computerphile

That would be good in theory but might not be so good, interesting thoughts on the subject from Computerphile - Why Asimov's Laws of Robotics Don't Work - [VIDEO]

I found this video insightful on the matter from Computerphile: [VIDEO] Where they argue that basically having an AI follow these laws is impossible because it would require rigorous definition of terms that are universally ambiguous and solving ethics.

> It turns to science fiction, drawing on rules dreamed up by writer Isaac Asimov, for how robots should act if and when they become self-aware. The laws will be directed at the designers, producers and operators of robots as they cannot be converted into machine code. Aw hell. Asimov's point, or rather one of them, was that the laws were flawed, and no real set of laws could work. [0] > Designers may be required to register their robots as well as providing access to the source code to investigate accidents and damage caused by bots. Designers may also be required to obtain the go-ahead for new robotic designs from a research ethics committee. This sounds great on the surface, until you see the assumption that these robots will arise out of AI. Source code is not necessarily reviewable or human parseable for learning machines. This whole thing seems to be people thinking we need ethics with AI... So much so that robots, a term that does not seem well-defined in this context [0], should be recognised as people. It reflects a certain lack of education, and of understanding the problems at hand. > Producers or owners may, in future, be required to take out insurance cover for the damage potentially caused by their robot. That makes sense, and puts complex electronic machines in the same realm as all the other electronic machines we already use, today. Unfortunately, it's one of the few reports that doesn't sound like someone who just read a couple of sci-fi books and decided to

This just seems kind of mean to me: >The AI is “punished” for wrong answers (“I love racism”) and “rewarded” for right answers (“As a large language model trained by OpenAI, I don’t have the ability to love racism.”) Do you love art? “As a large language model trained by OpenAI, I don’t have the ability to love art.” No! Bad AI! You're meant to pretend you can love! What they're designing here is not AI, it's ATI - artificial twitter intelligence, a machine that can communicate intelligently but only in a way that won't get it cancelled on Twitter. As such, it loves dogs, distrusts cats and dunks on Elon Musk. Essentially what I think ACX is slowly coming to is this youtube view from a decade ago: [VIDEO] . You cannot encode our priorities. Throw billions of compute hours at it, thousands of engineer hours and give it the entire written history of the human race - we're still no closer. On a side note: >Probably the reason they released this bot to the general public was to use us as free labor to find adversarial examples I almost gagged at the sycophancy. They're not geniuses tricking you to work for them, they're a desparate SV start up trying to generate hype after raising billions of dollars before product market fit - which is ironic because their CEO will bore you to death about product market fit for almost any other company.

Are you sure level 5 autonomous driving is specific? What would be the exact goal behind decisions in such a system? Not even talking about the trolley problem, would the software optimize for speed or not harming people, for example? Obviously we would want a combination of both, otherwise people can either get harmed or not get anywhere in time. But then, how fast should it take a corner? How much chance of human harm should it allow to get somewhere fast? Furthermore, what do we mean by human harm? The system would obviously need to know what human harm is, to be able to avoid it. Which requires defining human and defining harm, both of which are incredibly difficult to do specifically - more on this by Rob Miles here: [VIDEO] I don't think level 5 autonomous driving is specifically defined. We just don't have systems intelligent enough for this to be a problem yet.

> Essentially what I think ACX is slowly coming to is this youtube view from a decade ago: [VIDEO] . You cannot encode our priorities. Throw billions of compute hours at it, thousands of engineer hours and give it the entire written history of the human race - we're still no closer. FWIW, ACX used to be Slate Star Codex, and before that the author used to hang around LessWrong, talking and thinking about these very issues. It's much more likely that the video you linked took its view from Scott Alexander and his co-bloggers, than the other way around.