IQ (moved subtopic)

I feel like I should nitpick regarding musicians. There is more than a little overlap between many of them and people good at math. At least that is what I am told.

Look, we can debate semantics all day, but I think the bottom line is that you disagree with the relative value you perceive society places on the sort of ability measured by IQ tests. We can call it intelligence or we can call it gobbledymumbojumbery, it's not going to change the underlying issue here.

if we're nitpicking then it may not be wrong to disagree with the statistician if it will impress a girl you fancy.

Except that in almost all games the statistician will be recommending the option with both the higher EV AND the (more exciting) higher volatility.

I've always wondered why the statisticians get all the girls while the reckless show offs miss out.

There's a more important general point which is that disagreeing with the naked conclusion (absent of assumptions/etc) of an expert is often not disagreeing with their expertise.

I’m not gonna say it. But I’m thinking it.

God, it’s so soft… right over the plate…

Fight the urge, Xnerd

I'm still waiting for you to explain how robots measure distance without trigonometry btw.

How do they play ping pong without being able to measure distance?

You should see my dog catch a frisbee. It would blow your mind

So will robots process sensory inputs like humans and animals, without having things like Pythagoras' theorem explicitly programmed?

I'd be surprised if this was the case.

They will use trig as much as we do.

If there's a difference between machines and humans then it will to do with conciousness. Not using trig to play table tennis

The uncertainty principle of Quantum Mechanics suggests that if a person puts all of their mental effort into a single aspect of the world, then there is another aspect of the world, the aspect conjugate to the first, that they can never be the best at. For example, if they trained as hard as they possibly could to identify the location of objects, then they could not be the best at identifying the momentum of those objects. They could still be better at both than a dumb or even an average person, but they can't reach the peak that someone who specializes in the conjugate variable can.

Also note that in Quantum Mechanics, the states which minimize the total overall uncertainty of both conjugate variables are referred to as the Intelligent States. These states do not necessarily have to be equally balanced between both conjugate variables. They could be squeezed far to one side and still achieve the lowest overall uncertainty.

So is that what is referred to as "Artificial General Intelligence"? As in, basically a lab-grown human brain?

You must be jbouton's spiritual successor.

What's consciousness?

the standard trend these days is to just let the ai try everything and then discard the bad results and continue with the good ones - that way you just worry about developing a blank slate which can later be applied to many different applications

right now the biggest bottleneck in manufacturing with robots is that it can take months to program something like this

the hope is that in the future, when you want to make changes to the design or even develop a new kind of product, that the robotic arm can figure out what changes it needs to make to its routine on its own without requiring a team of engineers using micrometers to carefully give it exact instructions and then test to see if it followed it perfectly

a very good question.

I invite you to consider the much simpler chezlaw question 'when will the A in AI start to be deemed offensive'

https://arxiv.org/pdf/2408.03906

That's not guaranteed to get to the optimum answer. It could be that a walking stride length of 2.5 feet is better than 2.4 or 2.6 but that 2.7 is better yet. This actually came up in a toy holdem game I proposed where the single opponent moved in every time and your minimum calling hand in a freezout diminished as your stack did. There were situations where you folded hands like jj until your stack fell to a certain size but once it fell further you started folding it again until your stack fell further still.

this is a general problem of local peaks in search algorithms

Dinasours may be another example. Solving mazes a more obvious one

You have to attempt to explore the whole space bur far away disconected solutions are profoundly difficult.

yeah there's definitely some pathways it could go down where A>B so it goes A1>A2 and the A1a>A1b when for unforseen reasons B4d was the optimal approach but it'd never get there because we long discarded the B decision tree a long time ago

they do try to combat this by ensuring it doesn't follow too strict of a pathway to allow for some innovation to take place

this is really interesting though where google's deepmind played vs a pro

the end result was that the ai discovered all these strategies which absolutely crush humans - but they are also strategies which no human could ever do because it requires a level of micromanaging that even the top pros who can do up to 600 actions per minute are unable to keep pace with

this ai played an equivalent of 200 years worth of games of just protoss vs protoss in preparation for this match where they mixed and matched it both observing top pro games vs playing vs itself

it's a really interesting video though

You understand that there's a tradeoff between knowing the correct strategy response to a game state, and knowing how the strategy response changes with respect to changes in the game state. Do you not understand that people specialize in one meta strategy or the other? Any algorithm we construct could specialize in one or the other or equally balance both. A smarter algorithm will be better overall at both than a dumber algorithm. We could have two smart algorithms, each specialized to conjugate meta strategies, and each will almost always lose to the other at tasks where their specialization has an insurmountable advantage (e.g. if the goal was to find the maximum of a concave function, then the rate-of-change meta-strategy will easily beat out the check-and-compare values meta-strategy; and your own example shows a possible reverse scenario). Both algorithms could still be considered equally intelligent if compared against a wide enough and balanced enough variety of tasks.