The authors argue that objective-oriented mindsets are often deceptive guides when the goal is more than one “stepping stone” away. The path of discovery is often surprising in hindsight; you can’t identify in advance which stepping stones are good to pursue.
They advocate instead for pursuing novelty/interestingness (Get curious, Process over product) instead, as a “treasure hunter”. A sort of greedy algorithm. The trick with this approach is that you must give up any sense of control over the destination, the outcome. You just chase interestingness and find yourself delighted by what comes out.
One of their key secondary points is that many social arrangements are objective-oriented, even if they don’t use that wording. For instance, because peer review for scientific grants requires the reviewers to predict the prospective value of a particular project, or to predict whether a proposal is likely to achieve a particular objective, it falls prey to the same objections.
Page 1, an interesting prompt: “Do whatever you feel is most interesting.” I think it's in some respects like that's what I do for my research. It's true maybe in the long run. It's modulated with what I think I can actually get done and what I'm expecting is important.
Page three. They point out that scientists often demand hypotheses from each other because they don't want to fund research on “interestingness”. Hypothesis in some sense is an analog for objective.
They frame creativity is a kind of search problem.
Page 7, the central thesis. Objectives are well and good when they are sufficiently modest, but things get a lot more complicated when they are more ambitious. Objectives actually become obstacles towards more exciting achievements like those involved in discovery, creativity, innovation. The reason for this is that you don't know how to steer effectively through the search space. I don't really know how to play that out. Sometimes like you need to explore, not exploit. Maybe they'll make an analog that objectives are just about exploitation. Ah, quote, "The key problem is that the stepping stones that lead the ambitious objectives tend to be pretty strange. The problem is that the stepping stone does not resemble the final product. The arrangement or structure of the search space is completely unpredictable”
Page 10, "The problem is that it's hard to simply abandon objectives because they are a powerful security blanket". Yes, they are.
Chapter two.
The only thing I'm trying to figure out here is that in abecating and
injecting, I don't think this book is calling for strength anarchy.
It says, "I'm not going to be able to do this."
I don't think this book is calling for straight anarchy.
It says, "You may be wise to hitch a ride with serendipity.
==Being open and flexible to opportunity is sometimes== more important than only trying to do." ↗
So it's not fair, but I think it's still talking about conditioned action.
I don't think he has to be
He has to be careful with such people accidentally becoming what they became
I don't know, these kinds of just so stories aren't that dis
One claim here that does seem free is that these first few people tend to want a new open
The essential thing here is that we don't know which stuffings things might leave the
compad
I think I got the exchange. It seems like partially true.
Post-
Now if you just made a website where an e-movie could play many different pictures together,
and that is truth to truth is the right thing to generate another picture.
They claim this comes up with a question of objective in terms of thinking about what visitors are looking for when they see which creatures are looking for it.
They tried to make an objective argument, and they intended that objectives to be harmful with an empirical path.
They chose to target images in the art-building program, and they tried to actually generate the target images through recombination according to the rule of choosing characters,
like when they combined, they would produce something called "serivative" parts.
But that's trying to get him out of his way.
Also this qualitative reason of interesting images were reported to be generated by people who dropped the crow.
I think the part of this chapter seems to be that objective functions are bad campuses
even though they seem like they would work.
He chose me.
I require non-monotonous research.
They call this "reception."
The athletes claim that
the evolutionary selection, survival, and reproduction
are more of a constraint than infected.
They're an example of non-adjunct.
If you were trying to invent computers 5,000 years ago,
The computers of the objective did not work for us.
That can too, so I'm sorry to interrupt you, sir.
But their application of computing
in mental development can be explicitly not helpful.
The authors claim it's more likely that we'll think of that
into such full experience.
It's more clear that we think that they could be invented.
A provocate hypothesis about invention, page 36, quote, "Almost no prerequisite to any major
invention was invented with..."
Quote, "The great invention is defined by the realization that the pre-epicets are
employed, played before by predecessors with entirely unrelated and different... end quote."
Q. This book’s theory about the relationship between inventions and their prerequisites.
A. Prerequisites are almost never invented with their subsequent inventions in mind. Great invention comes from when one realizes that the prerequisites are in place.
Q. In what circumstances does this book claim objectives can help, and when can they not?
A. Objectives which are one hop away are good, but the trouble is with ambitious objectives, multiple hops away.
Chapter 5
This chapter is about novelty as they present sounds of public.
Novelty is a quote that is interesting.
The suggestion here is that you can navigate by novelty.
Not by novelty.
Page 41 quote, "The ideal interest of even if its ultimate purpose is unclear," end quote.
It's awfully difficult to get through review grant proposals this way.
There's an important claim about cities here, which is that as soon as Roman comes like
bumbling around, actually people have a clear sense of what they're saying.
What people think is serendipitous and accidental is a phenomenal
Q. In what sense is a strategy of chasing serendipity not random anarchy?
A. People have a nose for the interesting. They won’t really behave randomly.
The author is implemented a reinforcement learning algorithm called novelty code.
With...
Trans, a-vinx...
Are I trying to see the novelty?
Are I trying to re-
or if you're a simple reward.
Corruption.
Q. In novelty search, how does the robot learn eventually to avoid walls, navigate a hallway, and enter door?
A. It begins by crashing into walls all the time, but eventually it crashes into all the walls, and the only way it can produce something novel is by not crashing.
The claim here is that novelty provides an ordering.
We're going to see the hoodies are discovered in the order of temperature contrast.
Novelty creates an information accumulator back in the oil in which it takes place by time.
The longer the fish progresses...
Q. How does novelty avoid wandering through an endless number of meaningless states?
A. The agent can't teleport or walk through the door. The environment constrains its action space.
Q. What causes novelty search to produce more valuable outcomes over time, rather than just randomness?
A. Subsequent outcomes tend to capture more complexity in their behavior.
Interesting.
Rather than complete interesting answers.
The other thing that tends to be interesting is the idea of the agent.
Now that it intends to capture more complexity in its behavior over higher rates,
it is now to compare novelty to survival and rejection.
But it is not a guide to these things.
I don't know whether that is the way to do it.
One of the ways I say knowledge is different from any check will be spent.
It is a simple example.
It's not a steep lunge, but...
[
[
[
[
[
[
[
[
[
[
[
[
In the first empirical test with novel research, the author
[
and robots chinese on
They completed a novelty space robot against a robot with a character called the Maze by
having our word function proportional to its distance.
Then they looked at the proportion of a child.
Amazing, so cute, he's dead.
That's 39 out of 40. Now we'll be searching 3 out of 44.
[
[
[
[
[
There are a lot of exceptions in this name. There's a lot of code
that's
which they're a lot speaking.
[
[
How did the search work well for
It's about learning how to walk because you can't go very far in Prince's State if you don't walk around.
Chapter 6.
[
"The
The authors point out that a natural way to react to their claims is to take something
like the portfolio technique, or what they call the moral of diversity, namely that adjectives
are good and so is novelty and so is my clutter stuff, so you want to treat all these things
like that.
But this didn't stop distant objectives from doing this update.
Quote, page 57, "A false compass remains false no matter how far you may enter the path."
End quote.
end quote.
Q. What does the no free lunch theorem claim?
A. There is no overall best search algorithm over all possible optimization problems. Improving one search process to reach certain objectives will always harm search performance on a different set of problems.
Essential claim here. Case 59 quote, "We can reliably find something amazing. We just can't say what that something is."
The point of non-objective search is that it is
indifferent to what particular
outcome is being searched for.
Just looking for information.
I think that's one kind of treasure hunter, which is an online furniture store, like their
pick breeder.
My notion is that furniture gets recombined by users.
And you can identify the furniture that is of high value because it's what somebody actually
purchases.
And then subsequent users can take the high value of the server and use it in my head.
It's sort of a...
...a clever, clever hunting system.
Chapter 7.
They introduce Campbell's Law, which seems like a variant of Goodheart's Law.
"The more any quantum-cated circle indicator used for social decision making, the more
subject it would be to corrupt confessors, and the more apt it would be to distort and
corrupt the physical process through their contempt to monitor."
The test course is an example. Automatically an objective scores on test time entries, but they might come to normal because of sample clock. Likewise QDP.
Education is entirely...
...graphed up. Education is wrapped up in objectives with the
[
The criticism is that we're using society as a kind of search to find effective ways to
teach people and to help them grow.
The problem is that by instantiating the standards and the session and the people's accuracy
and the teacher's performance, we will be forced to convert.
The teacher's portfolio of assignments and syllabi methods are essential.
If it doesn't have a constraint, it's to measure my teacher's portfolio of assignments and syllabi
methods are sent for peer review and if they're below congressional then they will take some
immediate action but otherwise it stands for itself.
This is an assessment that is not based on a desired outcome.
It's a constraint case, it's definitely a thing to see whether we can actually
take a distinction that assessing where we want them to be is the objective policy,
assessing based on where people are is the fallacy of the treasure hunter.
The common quote from George Washington,
If you haven't heard from Benjamin Franklin in Paris, if you read him on Reddit,
this chapter seems to be about science funding.
I think, I don't truly argue, it's going to be pretty obvious.
The claim to science funding remains a consensus.
I mean, many reviewers do agree that the
It's like stepping stone.
We offer suit just the high variance,
the filming,
proposal, like Michael and Hengin
also discussed in there, so...
Hard to get the impact of hardware.
At least not without boxes of history behind you.
Agent 87, in quote, "One of the main reasons we tend to stick to objectives is kind of as a fear of risk."
And they are, honestly.
I also suggest that this is often where art and design get involved, maybe more art than
design. Designers feel very interested in fitness or purpose, and they may be in the
constraint-like sense of natural selection.
Chapter 9.
Q. What key concession must you make when navigating by novelty?
A. You give up control of the specific destination you seek, the specific problem you solve and answer.
Quote. Following the sense of intrestingness is justified precisely because we don't have a structure in the search space.
And for...
Shivering Chapter 11,
and Chapter 10,
on another region,
through Chapter 11.
We can cast the field of AI
as a search for algorithms,
and then all the arguments in this book
about search problems cost.
The A.I.R. field
is guided by two heuristics,
first the experimentalist heuristic,
which is that an algorithm is promising if it performs well,
in some benchmark,
It's like a theoretical heuristic, and there's an algorithm that's better if it can be proven to have certain desirable properties.
These are useful to read the rest of the publication.
These heuristics make it difficult to develop new types of algorithms because their initial performance is strictly controlled through an existing algorithm.
Some interesting going-set might be, well, it doesn't mean that they aren't worth pursuing.
It may just mean it's an apples and oranges comparison.
Just because an algorithm has helpful theorems, or rather has theorems which prove helpful
properties about a complete name that the algorithm itself is common, the algorithm
may lead the more facts the algorithm can achieve.
132.
"The community is so focused on performance that it misses the forest for the trees.
A good algorithm isn't one that performs better, but one that exists to think of other algorithms.
Performance-squeeding algorithms just aren't good stepping steps.
They propose a new kind of peer review in AI, where reviewers are forbidding from engaging with the results,
results, either in pure actual or theoretical, of a particular proposal instead of changing
the sentence. Think about the core idea.
In the book, I'm not sure that I got much news from you. I think the main thing I got was
The bullet fighting observation that if you believe
the gestures are not useful beyond a hopper tooth
and that instead you should guide based on novelty
which I usually think of in terms of curiosity
that they are very related
and you need to be prepared to give up the depth
(muffled speaking)