Ask YC: AI

https://news.ycombinator.com/item?id=90387

by Novash • 19 years ago

36 87 19 years ago

news.ycombinator.com

The field of artificial intelligence has been always something I wanted to take a peek on, but never really had the chance. Could people here recommend me some good books to start my journey on this field?

nickb• 19 years ago

Start with this book: http://aima.cs.berkeley.edu/ "Artificial Intelligence: A Modern Approach" by Russell/Norvig

It's one of the best books on AI and it's a "standard" AI textbook for most of the intro AI courses.

It's also great as a reference (I have it on my bookshelf beside me) and you can quickly look up algos and implement them in your fav language.

jey11• 19 years ago

Here's the standard AI textbook: http://aima.cs.berkeley.edu/

I'm assuming you mean "AI" in the sense of "fancy mathematical tricks to solve domain specific problems", not in the sense of artificial general intelligence.

wastedbrains• 19 years ago

Jey I agree, I highly recommend Norvig and Russell's book. I had a one semester course back in college that used this. I actually just recently have been programming some of the exercises up in Ruby just to stay sharp with some different kind of coding than the standard webapp stuff. It has some great problems, and has source code available in many languages (not ruby unfortunately, but I guess makes it more of a challenge).

Norvig also has some great essays and example AI code on his site, http://norvig.com/

Retric• 19 years ago

I agree it's a great book but IMO it's important to understand that modern AI is about:

1. Classification and Learning (aka is that apple ripe? or what's the best website for a given search?)

2. Modeling / Machine vision (Where are the rocks around this rover.)

3. Goal Seeking (AKA what's the best path from here to DC.)

With enough resources we can do any of the above fairly well. So using AI is more about understanding how to link the above activities to some useful problem. AKA control a rover when your ping is 15 min or solve a CAPTCHA.

amichail• 19 years ago

I actually don't find this sort of research all that interesting.

I like out of the box solutions more:

http://video.google.ca/videoplay?docid=-8246463980976635143

david927• 19 years ago

That's brilliant, amichail. Thanks.

whacked_new• 19 years ago

Second that thanks. That was expletively amazing. I didn't expect to watch more than the intro but couldn't stop. That's an intersection of CS and HCI cleverness raised to the level of art.

viergroupie• 19 years ago

Fantastic video, thank you!

ivankirigin• 19 years ago

Norvig, Mitchell, Thrun have good books

Also, google "Andrew Moore CMU" his homepage is full of useful tutorials

That said, "AI" is too vague. What do you want to be able to do?

Tichy5• 19 years ago

Maybe "Programming Collective Intelligence" would also be a good start.

Kaizyn3• 19 years ago

In addition to this book, you're probably going to want a good book covering the history and development of the field.

Keep in mind that AI is a broad field. As such you might find it more helpful to identify which computer intelligence problem(s) you want to solve, then start accessing the AI writings that discuss how to solve your problem. This won't give you a general overview, but it at least would reduce what you'd need to learn about to produce something useful.

neilc4• 19 years ago

Yeah, that's definitely a nice, practical book for machine learning stuff (but not really AI proper). For a practical book on AI itself, Norvig's Paradigms of Artificial Intelligence Programming:

http://norvig.com/paip.html

is a great book, and a nice complement to Norvig and Russell, which is somewhat more theory-oriented.

Tichy5• 19 years ago

What is "AI proper", though? I am fairly certain that when the true AI will be built, it won't be an expert system or based on first order logic.

The classical AI stuff is still interesting, of course, and still has practical applications (ie A* search for games). But I tend to count "the other stuff" (ie Data Mining) as AI, too.

bootload• 19 years ago

"... Yeah, that's definitely a nice, practical book for machine learning stuff (but not really AI proper). For a practical book on AI itself, Norvig's Paradigms of Artificial Intelligence Programming ..."

It is ironic that the man who wrote the book uses different methods (maths) to extract meaning from data at google ~ http://news.ycombinator.com/item?id=90298

Something smells about AI.

It has been well funded, attracted the best minds (Minsky, McCarthy, Newell) at places like MIT, Stanford and resulted in lots of useful languages & tools. [0] From Cyc to Powerset companies have tried to use these techniques trying to assign human like intelligence to machines yet have failed.

Why?

Bruce Stirling reckons it has to do a lot with the language. [1] Assigning intelligence to what is essentially a 21'st century equivalent of knitting mills or steam engines of yesteryear. Maybe thats why Google has taken the approach it has. Instead of creating intelligent software to find the meaning, take an insight into users who have pre-assigned intelligence to documents. And use well known sorting, linking techniques and software engineering along with not-so-well applied math techniques. Thus make google appear to make sense of what we search for.

[0] http://en.wikipedia.org/wiki/Artificial_intelligence

[1] ITConversations, 'Bruce Stirling, The Internet of Things, 58m, 26.4 mb, 2006MAR06' ~ http://www.itconversations.com/shows/detail717.html

mooneater• 19 years ago

Having looked through the introductory texts, I am left wanting to better understand current state of various parts of field.

Any suggestions on where to get more of that higher-level commentary?

pchristensen2• 19 years ago

I recommend "Beyond AI". It gives a high level history, present, and future of AI. It looks through the lens of a search for artificial general intelligence.

http://www.amazon.com/o/asin/1591025117

Novash• 19 years ago

I mean AI as Learning Algorythms and Decision Making programs.

jamiequint• 19 years ago

In that case get "Programming Collective Intelligence" like Tichy said, probably the best place to start.

Kaizyn3• 19 years ago

You might want to look at 'cognitive science' as well because that field tries to explain how humans think and make decisions for use as AI construction models.

jey11• 19 years ago

There currently are no good general solutions, so the best tool(s) to use strongly depends on the specific problem you're trying to solve, resources you have available, and how good the answer needs to be.

Reading through the Russell & Norvig book would give you a good overview of the sorts of techniques available.

robg2• 19 years ago

Unfortunately, too often, that perspective is lost. I know of no general solution to what defines intelligence. So I'm very dubious as to whether an encompassing artificial version is possible.

I see lots of specific solutions to specific problems. That, to me, is the best approximation of intelligence. To the extent that many specific solutions can be approximated and optimized, we'll be very well off. It's easy to denigrate a calculator and spell-checker, but I seem much more intelligent than I would be without them.

richcollins• 19 years ago

Also see Mitchell's machine learning: http://www.cs.cmu.edu/~tom/mlbook.html

rkabir• 19 years ago

my TA secretly recommended Norvig's book over Winston's, but we used Winston's "Artificial Intelligence"

For more philosophy, read Minsky's books / papers, and the readings outlined here: http://ocw.mit.edu/OcwWeb/Electrical-Engineering-and-Compute...

Minsky's paper "Steps toward Artificial Intelligence" reads like a summary of 6.034 at MIT.

Speaking of which: http://ocw.mit.edu/OcwWeb/Electrical-Engineering-and-Compute...

downer24• 19 years ago

I recommend starting from scratch and ignoring the field. All the field will teach you is very complex ways to NOT have AI.

Kaizyn3• 19 years ago

Why would you want someone to do this? Wouldn't that simply lead one to remake all the mistakes that have been made in the field?

downer24• 19 years ago

Doing what has been done in the field would repeat the mistakes.

Doing something original might actually be, you know, original.

pg• 19 years ago

Those lessons can be useful, though. You might make the same mistakes.

downer24• 19 years ago

Or you might come up with something that works, by avoiding having your mind crammed and distorted by the classical mess.

anaphoric2• 19 years ago

I have taught AI many times now and I must agree.

You don't know how many times I have heard a student claim that they have made some type of major advance. But on inspection it usually comes down to something that requires exhaustive search.

I think that there is definitely a place for new ideas in the field, but I don't think they will come from complete novices. I think they will come from those who have CRITICALLY read the basic literature. And probably from those who have read more than just the basics too.

My own pet peeve on the field now is the over emphasis on asymptotic complexity. Sure its important, but its not exactly an AI result. The scruffies need to make a come back and start coding! See my own meager efforts (http://www.anaphoric.com)

jey11• 19 years ago

The problem isn't that the field called "AI" is useless, it's just that they accidentally got away with being called "AI". Without the vast body of research done under the banner of "AI" you wouldn't have OCR, speech recognition, fraud detection, search engines, ...

That said, I also don't think you need to be an expert neural network wrangler and genetic algorithms shaman if you want to study "true AI". You should however have an understanding of the fundamentals behind the techniques used in classical AI so that you can rationally dismiss them, or hell, maybe you'll even find some amazing lead or inspiration in the existing work. After all, the field of AI did start out with ambitions of creating "true AI" (whatever that is).

downer24• 19 years ago

When someone says "AI", I think they mean AI, not e.g. path-finding for writing a Quake bot.

Of course a real AI would be much better at all those things you mentioned anyway -- like deciding which of my e-mail messages are spam.

Or carrying on a conversation.

P.S. I'm an AI.

lkozma• 19 years ago

Ignoring existing work in AI is as if you would try building a flying machine, ignoring existing knowledge in aerodynamics. You'd be right back where Leonardo was. People are doing that, nevertheless, but if you do that in AI, chances are you'll be just fabricating crackpot theories.

downer24• 19 years ago

The difference between the field of AI and the field of aerodynamics is the state-of-the-art in the latter has produced working results -- to say the least!

Meanwhile humans farm gold and solve CAPTCHAs.

mojuba5• 19 years ago

I'd recommend starting with a definition of intelligence. Once you clearly and unambiguously define it, consider you have 51% of the task done already.

Ask YC: AI

Related Stories