Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Siri meets Eliza (jordanmechner.com)
254 points by user24 on Oct 25, 2011 | hide | past | favorite | 29 comments



Although Eliza was coded for a less powerful machine, the most powerful mainframe in existence in 1964 was the CDC6600, which was able to put out about 1 MFLOPS. The iPhone 4S is apparently capable of 138 MFLOPS. ( http://www.phonearena.com/news/Apple-iPhone-4S-benchmark-tes... )

We've managed to put the equivalent of hundreds of room-sized mainframes from Eliza's time into something that can fit in your pocket but, in the last 47 years, A.I. has advanced so little that Siri is using the same bag of tricks as Eliza. All that distinguishes Siri from Eliza is some licensed speech recognition software, a nice collection of pre-programmed geeky easter-eggs, and the fact that Siri is plugged into a collection of apps and search engines that allow it to perform a few basic, but useful tasks beyond simply fooling around. However, those tasks do not really require, or make use of, A.I. at all. Google's voice interface for Android makes no pretensions towards sentience, but is capable of everything Siri is. Google just hasn't wrapped their voice interface software up in a cute anthropomorphized Eliza wrapper or taken the app integration as far yet. That, and perhaps a little interface work, is pretty much it.

It is simply astounding that, in spite of Siri having access to such vastly superior resources, Eliza still comes across as the smart one!


Well, that's not really fair, because Siri isn't designed to try to have a conversation -- it's designed to just get a few things done easily via voice commands.

The state of the art in chatting is probably something more like Cleverbot, which is way, way better at having a conversation than Eliza.

(It occurs to me that hooking Siri up to cleverbot would be pretty awesome...)


i think it says more about how simple real humans are when they interact with each other. as limited as Eliza is, it has worked many times. how many times a day have you had face-to-face interactions with people that are simply canned responses. i remember keeping a phone call going with high-school girlfriend by randomly responding with one of the following : "uh-huh", "yeah" or "really".


The best part of that page is comment #4 (quoting)...

Reminds me of conferencing calling two Dunkin’ Donuts together and listening silently as hilarity ensued:

"Hello, Dunkin Donuts."

"Yes, this is Dunkin Donuts."

"Yes, this is Dunkin Donuts."

"Yes."

"Hello?"

"How can I help you?"

"This is Dunkin Donuts."

"Yes, it is!"

"Hello?"

It was the last time my friends’ parents let us use their fancy new (1983) Merlin phone unsupervised.



There's an amusing prank where you call two Chinese restaurants: http://www.youtube.com/watch?v=Iz9ra9FqNKQ

(I think they pulled this prank on a late night show as well, I can't find the video though.)


Related call where 2 sex line workers are let loose on each other: http://www.youtube.com/watch?v=OoyDQ327mm8


Not that I have any experiences in this regard, but I hear that SkypeOut and the conferencing feature is ... convenient.


Back in the days of phreaking, this was called a 'Cheese Box' and was used to great effect by calling two locations of any chain. From what I recall, a report in Phrack called two 711 employees who started yelling at each other 'NO, THIS IS 711, WHO IS THIS‽'


Not exactly surprising results, but pretty entertaining.

Eliza had fewer non-sequiter's, generally kept up the conversation better. Eliza pretty clearly wins the discussion. But Siri demonstrated beyond a doubt that she is a woman of action by searching the web, finding businesses, and calling people. By virtue of actions speaking louder than words, I declare Siri the winner of the shouting match.


Siri: We must do something. This is something. Therefore, we must do it.


This conversation is like listening to some executive's over-caffeinated assistant trying to help her boss's overly Xanax'd therapist. Not sure what the take-away here is other than Eliza is such a lazy program. It just falls back onto open-ended questions when confused and doesn't really do anything productive at all.


Eliza is actually an AI-building platform. The "doctor" program was like a Hello World, it didn't really show what the platform was capable of. https://secure.wikimedia.org/wikipedia/en/wiki/ELIZA


That’s the point, though, isn’t it? I think Weizenbaum was pretty devastated when people actually took it serious.


Using open-ended questions and parroting mirrors what good human listeners do. We treat it as lazy only because we know that in this instance it happens because it's the easy way out of a situation where the program doesn't have another response.


It's entertaining, but the thing that struck me most after reading that "conversation" is how far we still have to go in AI.

That can also be taken as good news, ie there is still a lot to be done/discovered in this interesting and very difficult field.


I did a report during my undergrad (2005) on AI chatterbots and compared various bots; they all had huge distances to go before approaching any kind of real conversation. During my masters (2009) I studied computational linguistics and found that the gap was even wider than I'd previously thought. Even generating a natural sounding 2 sentence exchange is very difficult, let alone a lengthy conversation, and as far as anything approaching genuine understanding and insight? We don't even know where to begin yet.


Yeah, but then again, it takes the human brain about 15-20 years before you can have a somewhat reasonable conversation with its owner, and that's dedicated hardware custom-built for that purpose.


I guess you don't interact much with kids. 3 year olds will give great conversation and sometimes very interesting insights.


Based on the following:

SIRI >> Sorry, I can’t provide maps and directions in Canada.

ELIZA >> Have you tried?

I would say that Eliza is smarter than Siri.


A great hack and easily the funniest thing I've ever read on HN [damning with faint praise unintentional].

Parry v Siri would might be Youtube worthy.


I was just going to suggest a match with Parry myself! Parry was seriously better than Eliza and had a more natural tone. I'd love to see someone try this. I'd try it, but I don't have an iPhone 4s and don't have a running Parry engine.


The Singularity is not near. :)


This was very similar to an exercise we had in a class I took a decade or so ago "Theory and Practice of Non-Linear and Interactive Narrative" (http://web.mit.edu/21w765j/www/Home.html)

Also, I think they took 'yow' out of emacs at some point (or at least the actual Zippy the Pinhead quotes - the command is still there), but the "psychoanalyze-pinhead" command (which pipes random quotes from that comic into Eliza) is really trippy.


"I lunge for the phone, stopping the experiment."

That's been my exercise (gym membership expired 3 years ago) as I try different Android voice recognition apps. Add random people you no longer remember on your Facebook account which is automatically put in my phone contacts. It should be a sport.


I'm just bummed that he stopped the experiment before Mr Jose Fuentes got involved.


My Hobby: Proxying conversations with my cow-orkers to Eliza.

Try it, great stress relief. I've only done this once, and they did realize what was going on. Will try to dig up the chatlog, but I doubt I can find it.

Appy polly loggies to the xkcd guy. ;)


Anyone tried a Turing test on Siri yet? She'd fail the first time she tried to call someone or do a web search, but if you could head her off before she can try to do anything but answer it might be quite convincing.


http://www.youtube.com/watch?v=WnzlbyTZsQY

AI vs. AI. Two chatbots talking to each other




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: