Robert Miles AI Safety
6 HN recommenders · recommended in 7 months, 2022–2026
13:03 · 710K views · Jan 11, 2018
Robert Miles explains the orthogonality thesis: how an agent can be highly intelligent while pursuing goals that seem stupid or alien to humans. He separates intelligence from goals and touches on the stamp-collector example.
HN people use it as the standard short explanation for why AI can understand human ethics without sharing them, and often recommend the whole channel.
12 comments from Hacker News
I've seen that paper, and it's amazing indeed what GPT-4 is capable of. But none of that supports that closing quote, which to me points to a worrying philosophical naïveté. How can one equip an entity with "intrinsic motivation"? By definition, if we have to equip it with it, that motivation is extrinsic. It belongs to the one who puts it there. A software engineer might decide to prompt his creation with something along the lines of "Be fruitful and increase in number; fill the earth and subdue it. Rule over the fish in the sea and the birds in the sky, and over every living creature that moves on the ground." and the bot will duly and with limitless enthusiasm put those wise orders into practice. If that ends up causing some "minor" problems, should we confine the software for one thousand years in an air-gapped enclosure, until it learns a bitter lesson? Or should we take to task the careless Maker of the bot? To me, the answer is obvious. PS: this video to me states the problem in a nutshell (that whole channel is gold) this video
That’s the alignment problem. We don’t know what the actual goals of an AI trained neural net are. We know what criteria we trained it against, but it turns out that’s not at all the same thing. I highly recommend Rob Miles channel on YouTube. Here’s a good one, but they’re all fascinating. It turns out training an AI to have the actual goals we want it to have is fiendishly difficult. this video
Yes, I believe some careful and informed pondering leads one towards that conclusion. Hume's guillotine is an implacable philosophical device: this video (I warmly recommend all the content on that delightfully frightening channel)
> The argument for SkyNet happening is missing an important element: why would it happen? Because sci-fi says it will? That element isn't missing. It's explained every time someone asks, and the answer never relies on sci-fi as evidence. The reason is because of (A) orthogonality and (B) instrumental convergence. If you aren't familiar with those terms, (A) refers to the thesis that an AI's degree of logical reasoning ability is independent of its goals. This stems from Hume's Law [1]. (B) refers to the observation that regardless of an agent's terminal goal, many instrumental goals will be predictably shared: the agent will be incentivized to increase its own capabilities, as a means toward nearly any possible end. This includes steps that would prevent humans from interfering with the AI's goals, such as a human attempting to turn it off or reprogram it. If you have questions about either of those, there are great learning materials available, including a well-made youtube series by Robert Miles. [2] > Because LLMs have reached a point of diminishing returns and there isn't enough compute globally to do produce a hypothetical AGI. What makes you think LLMs have reached a point of diminishing returns? So far, the benchmarks are continuing to show accelerating returns, to the point that it's becoming difficult for benchmarks to keep up.[3] [1]: Hume's Law, AKA the Is-Ought problem: [LINK] [2]: this video [3]: "Foundation models and the next era of AI", Microsoft Research, [LINK]
Good point, that very much vibes with my thoughts on this matter. Lately, I've been contemplating the analogy between the role LLMs might take within society with that of the brain's language center* in human behavior. There's definitely a way in which we resemble these models. More than some might like to admit. The cleverness, but also the hallucinating, gaslighting and other such behaviors. And on the other hand, any way you'd slice it, it seems to me LLMs - and software systems in general - necessarily lack intrinsic motivation. By definition, any goal it has can only be the goal of whoever designed that system. Even if its maker decides - "let it pick goals randomly", those randomly picked goals are just intermediate steps toward the enacting of the programmer's original goal. Robert Miles' YouTube videos on alignment shed light on these issues also. For example: this video Another relevant source on these issues is the book "The Master and his Emissary", which discusses how basically the language center can, in some way - I'm simplifying a lot, fall prey to the illusion that "it" is the entirety of human consciousness. * or at least some subsystems of that language center, it's important to remember how little we still understand of human cognition