I was looking at AIMA the other day. You know, Norvig and that other guy?
Well I was reading the book and skimming through and it just hit me. This is all total bullshit. This shit doesn't work at all.
The books a scam written be charlatans - they include a bunch of well known stuff that you have to. E.g. alpha-beta search (invented by john mccarthy), some prolog, lots of talk about bayes law.
But between all that stuff (which is very basic knowledge) they just have surveys. X tried this in 1992. Y tried this in 1994. That's great but I want to write a program that solves a problem - why are they teaching all these methods that fall flat? Because it's the best we currently know?
Person X has some programmatic idea to solve some particular problem that humans are good at but machines haven't done well at yet. X promotes that particular solution as a breakthrough AI moment, and that everybody else is going nowhere.
Person Y has his own view on how the human mind/brain/body works. Y promotes that view as the One True Way to AI, and everybody else is deluded.
Person Z had some success in one particular AI approach, and believes it will scale up to the level of human intelligence. Person Z thinks they've found the silver bullet, because Z has personally not had success with other methods.
They're all half right.
Name:
Anonymous2016-01-10 23:46
I don't think you understand what is the meaning of scam
You make some interesting points but I still hold Norvig to be quite legitimate. The textbook is presumably meant to explain the field and the large amount of existing knowledge to newcomers and not constitute an instance of original research. It is therefore good if it can make references to Prolog and John McCarthy. Norvig's other book PAIP was quite excellent too.
Also, don't both these books, AIMA and PAIP, including many instances of real-world solved problems? For instance PAIP covers parsing natural language, and it also discusses computer algebra systems (which are ripe with useful applications to science and engineering) at length.
Has anyone created something good that they learned how to do from AIMA?
Name:
Anonymous2016-01-11 4:06
By AI do you mean the field of the goal?
The field is all fucked up. Machine learning is profitable, but I wouldn't consider it a valid approach to AI. It might get there, but looking nothing like it currently does.
Back in the good old days, people thought direct symbolic manipulation was a valid approach. Even Turing seemed to think this. I can forgive Turing, but people still seem to believe this today. It's wrong. I think it's possible a symbol-oriented system could work, but it's more likely someone will put together a neural net that does the job first.
neural nets are great at their domain but their functionality is rather primitive (AI is expected to be complex and sophisticated, neutral networks are data models and optimization function) So people see them as "brute-force AI", designed to compensate by massive parallelism of simple units. Like expert systems, they are sometimes used outside of their domain(like e.g. chess) with disappointing results.
Neural nets have some sense to them, I mean real human brains are still a product of Darwinian evolution which can be thought of as a kind statistical optimization
This was very interesting in light of recent stories about probabilistic AI being increasingly used for real-world decisions, for example they have crime scene DNA analysis AIs which arrest people. It would be definitely preferable such AIs were to be "held accountable" as Sussman argues.
Name:
Anonymous2016-01-11 16:45
Aren't the Sussman's thoughts on AI very anthropocentric? Humans represent a very small percentage of brained organisms on the planet, and yet they are the only ones that can explain the reasoning behind their behaviors (and only some of the time, at that). Can we not therefore conclude that this property is not strictly necessary for artificially intelligent systems?
Name:
Anonymous2016-01-11 16:49
>>18 Irrelevant. The entire universe exists solely for humans to use as they see fit. All other minds are mere objects, property to be disposed of ta will. We don't want fucking artificial cat minds running around, and any artificial cat minds should be destroyed. We want an artificial human mind, that acts and behaves human. Anything lesser is worthless.
Name:
Anonymous2016-01-11 16:57
The entire universe exists solely for me, and everything else is my property.
Name:
Anonymous2016-01-11 16:58
What about that artificially intelligent dog in Fahrenheit 451? It was very useful!
Name:
Anonymous2016-01-11 17:21
Am I being trolled? I never had any bad intents you know, just a little bipolar perhaps. I just want to help!
Biology apologists are the reason AI right now is worthless.
Name:
Anonymous2016-01-11 18:52
>>18 As always, it all comes down to the definition of "intelligence". For some problems, the intelligence of a rat would suffice. For others, the intelligence of a human would be a bare minimum.
>>19 Artificial human-equivalent minds would be stupid to build, as they'd also get bored, get restless, forget, and have emotional weirdness. Any approach simulating the mechanics of the human brain will fall into this trap.
Name:
Anonymous2016-01-11 19:18
The propagator model sounds like OOP with getter/setter methods corresponding to propagators and objects as cells.
>>25 Only if by "OOP" you mean the older aspirations of asynchronous message passing.
This is also not new. Rete networks have been a popular form of this independent propagation idea for a very long time, for example. Though it's not cyclic.
The books a scam written be charlatans - they include a bunch of well known stuff that you have to. E.g. alpha-beta search (invented by john mccarthy), some prolog, lots of talk about bayes law.
But between all that stuff (which is very basic knowledge) they just have surveys. X tried this in 1992. Y tried this in 1994. That's great but I want to write a program that solves a problem - why are they teaching all these methods that fall flat? Because it's the best we currently know?
So basically you're saying that they list the stuff that works (and is therefore well known) but also some stuff that was found to not work, and the latter indeed doesn't work?
Ah, humans and their inimitable stupidity!
Name:
Anonymous2016-01-14 12:39
I was looking at AIMA the other day. You know, Norvig and that other guy?
i was expecting you to go full yoshinoya. Go big or go home.
Name:
Anonymous2016-01-14 20:26
>>33 you haven't understood a thing i said and i don't think there's hope you will so just find something else to do that try to enter this discussion
Name:
Anonymous2016-01-14 23:11
>>35 That's because you're not making yourself clear to the audience. How can they be charlatans if they tell you exactly what level of knowledge is known in humanity?