Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[pisskop]

I'm sorry 28 year old brain cant process this new information

[wierd]

Whiner. My 35 year old brain can learn new things just fine. (but admittedly, *IS* getting more difficult, and I DO notice.)

[Neonivek]

I wonder if gay people existing in a state of not turmoil is "Liberal Agenda"

Then what would the "Conservative Agenda" version be? Like... How would Garnet in Steven Universe act as a "Conservative Agenda" character?

[pisskop]

i didnt read much into it, but it sounds like they want to take preconfigured logic and strap this program to it?

i mean im sure in 30 years it would be 'smarter', but i used to think furbies got smarter too

[Tack]

What would the "Conservative Agenda be?

Nixon in Watergate is 'conservative agenda' to me.

Well you can get into more of the language / predicate calculus style machine learning if you want, that area's far less crowded that the NN / deep learning field these days.

https://www.technologyreview.com/s/600984/an-ai-with-30-years-worth-of-knowledge-finally-goes-to-work/

While they benefit from being easily understandable, they're incapable of directly dealing with uncertainty and probability, being based in formal logic which doesn't work with those things. Also, I don't want to spent 31 years filling up a database. It's not much of a "machine learning" if it takes that much for it to start working.

I suppose it's a 30 year database that other, faster machines can more easily access and parse, rather than having to try and make all of the inferences from scratch.
Good luck to 'em though.

[pisskop]

I wonder if gay people existing in a state of not turmoil is "Liberal Agenda"

Then what would the "Conservative Agenda" version be? Like... How would Garnet in Steven Universe act as a "Conservative Agenda" character?

the 'agenda' of lobbism is to propagate itself.  this involves having an enemy to face and a struggle to overcome.  doesnt matter what 'progress' occurs, what matters is that the lobbism is needed.

[Reelya]

i didnt read much into it, but it sounds like they want to take preconfigured logic and strap this program to it?

i mean im sure in 30 years it would be 'smarter', but i used to think furbies got smarter too

The logic doesn't need to be preconfigured, that's the thing about Cyc. You just feed raw data in and it uses it's layered knowledge of the real world to make sense of it. Basically it can be configured for a new domain by dumping data into it, then it spits out questions about things it doesn't understand, and makes inferences. You know whether it's working by the quality of the inferences it's making, and the types of clarifications it asks for.

e.g. Cyc flagged a possible case of insider trading, given a lot of information about the financial world, by spotting that two people who later were involved with a number of trades used to sit at desks next to each other. There was no formal rule: "sitting next to  another guy means you might be involved with insider trades later on", it infered that on it's own.

And the point isn't that Sergarr should make another Cyc. The 30 years worth of rules about the real world and natural language processing that they put in there is the basis for future applications, no different to 30 years of development towards deep learning neural networks. Saying Sergarr should replicate Cyc is as silly as saying that to utilize deep learning you need to go back and replicate 30 years worth of NN research.

[wierd]

In any society, there are those that stand to profit from a change of the status quo (currently manifested by Progressives and minorities) and those that profit from retaining the status quo. (traditionally wealthy, or landed people.)

Those that profit from retention of the status quo will see conspiracies everywhere, because there really *IS* organized effort to try to change that status quo they profit from-- hence, "agenda!".  In this case, we have a conservative demographic that profits from a status quo (vestigial religiously inspired morality law concerning sexual activities that they profit from, as it justifies their contempt for alt-sexuality, by associating it with illegal, and thus immoral behavior) seeing a "Gay liberal agenda!" in those that seek to overturn that status quo (Gay people who do not want to be put in prison for having gay sex, because of silly sodomy laws that were inspired by the bronze age's brutality) because the latter really *IS* actively trying to table that status quo. (there really *IS* an agenda-- that agenda is the removal of these absurd laws.)

The conservative agenda, conversely, is to strengthen, reinforce, and stonewall any "deleterious" changes to the status quo. (where "deleterious" == "Will disadvantage me, personally", and where "Me" == the conservative person in  question.)  This is because they are all about conserving their power base, be it financial (conservative with money, to accrue additional reserves of capital) or political (conservative with clemency, to accrue additional power via enforcement and social acceptance.)

The issue, is that neither side is capable of seeing that they do in fact, have an agenda.

[Sergarr]

i didnt read much into it, but it sounds like they want to take preconfigured logic and strap this program to it?

i mean im sure in 30 years it would be 'smarter', but i used to think furbies got smarter too

No, it's just an array of facts, like "water is blue", "salt water is sea-blue", and so on. It takes a lot of time to input all these facts in, though, and you have to do it manually, with little automation possible. 31 years before commercial usage is pretty much ultra-terrible by modern machine learning standards.

[Reelya]

i didnt read much into it, but it sounds like they want to take preconfigured logic and strap this program to it?

i mean im sure in 30 years it would be 'smarter', but i used to think furbies got smarter too

No, it's just an array of facts, like "water is blue", "salt water is sea-blue", and so on. It takes a lot of time to input all these facts in, though, and you have to do it manually, with little automation possible. 31 years before commercial usage is pretty much ultra-terrible by modern machine learning standards.

I don't know about that. How many years did it take between the Perceptron and commercial neural networks? That was like 40+ years. You're also talking about modern machine learning. It doesn't make sense to say "this tech's been under development for 30 years, that doesn't compare favorably to how quickly modern Big Data applications are created".

99% of current machine learning applications are based on a small set of algorithms that were developed from 1950-2000+, roughly. They're not designed individually at all, they used off the shelf tools that took half a century to perfect.

Natural language processing base AI, which is what Cyc is, is an all or nothing thing. Your thing either reads and comprehends the text, or it doesn't. NNs are more of an incremental thing, they can match the patterns sort of, then you throw more power at them and they match it slightly better. Taking 30 years to design and build the tech that can read English isn't a bad thing, it's needed since the machine needs to interface with humans to learn the rules, there's little way to automate it, since any machine which could grade Cyc would need to know English already ...

Trying to do what Cyc does with NNs would actually be a doomed project. e.g. pump a pile of dog and not-dog photos into an NN and it'll learn simply to say yes/no as to whether a picture shows a dog. That doesn't help it answer "what is a dog". Language-processing AI's such as Cyc might well be the "glue" that builds workable human-interfacing AIs that include NNs as subcomponents. e.g. Cyc could understand that a set of pictures show dogs and not-dogs then it could infer (from things we tell it) that it a neural network would be a good way to distinguish photo sets, and it would be able to set up the training system to add that capability to itself. Trying to do that in reverse: NNs to language processing, would be much more difficult.

[wierd]

That is only true if a good, consistent, and robust formal language is developed that the linguistic AI can work with.

Human languages are full of ambiguity.  The AI would need a quality, highly precise definition of what "dog" is, how "dog" relates to "cat", etc.

[Sergarr]

i didnt read much into it, but it sounds like they want to take preconfigured logic and strap this program to it?

i mean im sure in 30 years it would be 'smarter', but i used to think furbies got smarter too

No, it's just an array of facts, like "water is blue", "salt water is sea-blue", and so on. It takes a lot of time to input all these facts in, though, and you have to do it manually, with little automation possible. 31 years before commercial usage is pretty much ultra-terrible by modern machine learning standards.

I don't know about that. How many years did it take between the Perceptron and commercial neural networks? That was like 40+ years. You're also talking about modern machine learning. It doesn't make sense to say "this tech's been under development for 30 years, that doesn't compare favorably to how quickly modern Big Data applications are created".

99% of current machine learning applications are based on a small set of algorithms that were developed from 1950-2000+, roughly. They're not designed individually at all, they used off the shelf tools that took half a century to perfect.

The difference is, if you lose a built-up database for a neural network, you can re-acquire it relatively quickly, less than a year in most cases, because data acquisition process is quite easy to automate in most cases. Whereas, with this 31-year database, if it's lost, you'll have to start from scratch, and it'll again take 31 years to do it because it's manually constructed. It seems much more fragile, much easier to lose.

Also, we don't even know if that 31-year database actually works. They say that it's ready for commercial applications, but is it, really?

[Reelya]

Why would it be lost? The database exists in every copy of the software.

And why do you think it's not already deployed? It's made by a company, of which Cyc is their core product. How d'ya think they stayed in business for 30 years?
https://en.wikipedia.org/wiki/Cyc

We know it's ready because it's been used in various projects for most of the three decades, that's how they stay in business. When the creators say it's "ready" they mean ready to expand into many more niches than the ones which they currently get paid to deploy it to.

[Neonivek]

Well my Hypnosis business isn't going anywhere.

It seems like the best business practice you can make right now as a hypnotist without a patreon is to offer free hypnosis sessions... But charge for hypnosis files (or have the files be part of the patreon).

I mean I have an editor for my hypnosis files, but goodness are they not exactly my area of comfort. I am great at picking out individual hypnosis subject's weaknesses and desires and playing off of those.

Though oddly enough people compliment my voice all the time. O_o

Soo going back to offering hypnosis sessions for free.

[Sergarr]

Why would it be lost? The database exists in every copy of the software.

Maybe it wouldn't, but I still don't like it when stuff is vulnerable to extinction like that.

And why do you think it's not already deployed? It's made by a company, of which Cyc is their core product. How d'ya think they stayed in business for 30 years?
https://en.wikipedia.org/wiki/Cyc

We know it's ready because it's been used in various projects for most of the three decades, that's how they stay in business. When the creators say it's "ready" they mean ready to expand into many more niches than the ones which they currently get paid to deploy it to.

Given that I've never heard of it being used, chances are its applications are very obscure places that are not very demanding in terms of commercial applicability. That wikipedia page seems to confirm that.

And that's not surprising, I mean, it's a formal logic engine, it's fundamentally incapable of dealing with uncertainty, which renders it inapplicable for the majority of competitively useful applications. It certainly isn't capable of solving Go the way AlphaGo did it, hell it's probably not even capable of playing chess or checkers. It doesn't actually do anything other than answering semantic questions in a fixed manner, it's not capable of dynamically learning and exploring new things on its own.

The sort of thing I'm looking for is similar to a war strategy and tactics analysis, where you have state space, actions which your agents could take in it, and a certain problem you need to solve using those agents, and a result would be a description of a plan of actions, possibly conditional in case of uncertainty, that would have a near-optimal chances of solving the problem. This "Cyc" thing is not it, the reinforcement learning is not it, the deep learning stuff is definitely not it, too. I want something that probably doesn't exist, but I still want it, not something else but it.

Well my Hypnosis business isn't going anywhere.

It seems like the best business practice you can make right now as a hypnotist without a patreon is to offer free hypnosis sessions... But charge for hypnosis files (or have the files be part of the patreon).

I mean I have an editor for my hypnosis files, but goodness are they not exactly my area of comfort. I am great at picking out individual hypnosis subject's weaknesses and desires and playing off of those.

Though oddly enough people compliment my voice all the time. O_o

Soo going back to offering hypnosis sessions for free.

...hypnotism is a business?