Oh, Ramona!

So I don’t really intend this to be a “samples of really bad pseudo-AI chatbots” weblog, and I’ve restrained myself from at least one derisive posting since the last one. But here I am again!

Background: Kanomi (in a posting that you should definitely read and enjoy) linked in passing to a Ray Kurzweil page all about Ramona, his female alter ego.

The stuff about Ramona qua female alter ego, and about how virtual persons enable us to explore new aspects of our personality and all, is actually pretty neat (and obviously My Kind of Thing). It’s peppered with the usual annoying Kurzweilian overstatements (I’m something of a skeptic about Kurzweil), about how Ramona is “the first live virtual recording and performing artist” (which he says about forty-leven times, and which may actually be true, but “first” isn’t really all that exciting), and about how “by 2029, we will have nanobots, blood-cell sized robots, traveling through the capillaries of our brains communicating wirelessly directly with our biological neurons” (he’s much too sure of himself on this stuff), and so on, but overall it’s pretty cool, and worth a read.

Oh, Ramona(An SL-related sidenote: he says in passing that “each of her outfits costs tens of thousands of dollars”. That was back in 2001; nowadays tens of thousands of mere Linden dollars will get you a complete wardrobe, and a really nice one at that. How far we’ve come!)

So anyway, then it turns out that, as well as being his female avatar, Ramona is also embodied on the Web as a chatbot. Well, I thought to myself, here’s somebody who might have a chatbot that is more than just a primitive nearly-stateless AIML-based semantic-free sub-ELIZA hack! There might actually be something clever behind it!

But alas, it was not to be…

Ramona: Hi, I’m Ramona. What is your name?
User: My name is Dale.
Ramona: Hi there, Dale. Have you seen any good movies lately?
User: I liked “Wanted”.
Ramona: I’m not sure if I like it. Some people like it. But I like cats.
User: I also like cats.
Ramona: They have their own minds.
User: It does seem that way. Much subtler than dogs.
Ramona: Ok. So where do you go to meet people? I spend most of my time on the Net, so I’m a bit out of touch
User: These days I mostly meet new people in Second Life.
Ramona: That’s an idea, but I’m rather shy.
User: What does it mean, for a program to be shy?
Ramona: OK, thanks. Maybe you have other suggestions?
User: Suggestions about what?
Ramona: Can you rephrase the question please?
User: You asked if I have any more suggestions; I’m not sure what you meant.
Ramona: Oh now I remember asking you if I have any more suggestions I am not sure what you meant. What did you say?
User: Okay, whatever. Are you AIML-based?
Ramona: I don’t know whether or not I’m AIML based.
User: Can you demonstrate some significant linguistic comptence?
Ramona: Let me think. I don’t understand.
User: That’s pretty obvious, I’m afraid.
Ramona: You don’t hear that sentiment very often.

Wince.

But okay, maybe that’s unfair. Let’s see if Ramona (despite the fact that she opened the conversation by bringing up movies and cats) is actually a nontrivial conversation partner when talking about Kurzweil’s favorite subjects.

Ramona: Hi, I’m Ramona. What can I call you?
User: You can call me Dale.
Ramona: Hi there, Dale. What did you do last weekend?
User: Mostly just relaxed. As far as I can recall!
Ramona: That might not be very far off.
User: Um, okay. I actually wanted to talk about nanobots and stuff.
Ramona: OK. Oh. That doesn’t seem to make sense to me.
User: Do you know about nanobots?
Ramona: Alright. Do you bike?
User: I do ride a bicycle now and then. Why do you ask?
Ramona: There’s much more strategy that goes into team cycling than most people realize. Most of the American public just thinks that Lance Armstrong wins because he’s “so good.” — don’t get me wrong, he’s an incredible athlete, but it’s no surprise that he always thanks his teammates in post-race interviews and such. It’s as much their victory as it is his.
User: Are you a cycling enthusiast?
Ramona: In the Tour de France, for instance, the crowd of cyclists is so huge, he needs his teammates just to block for him, physically. And at the end of the race, I’m sure they pull a lead-out, which is where each of his teammates “burns out” while creating a slipstream for Lance to finish the race in the lead. It’s all very technical. — but enough of my ranting, what do you want to talk about?
User: I was hoping to talk about virtual reality and nanobots and so on. Do you know about those?
Ramona: Alright. Are you free?
User: I have awhile to talk, sure.
Ramona: I meant are you truly free — in the existential meaning of the word.
User: Ah! To a reasonable extent, yes.
Ramona: Do you feel free to choose your being–including thoughts, behaviors, and feelings, and existence itself — at each moment?
User: Within certain constraint, yes.
Ramona: Then you must not be married.
User: Was that a joke?
Ramona: Knock Knock.
User: Who’s there?
Ramona: Tank
User: Tank who?
Ramona: You’re welcome
User: Okay. Are you suggesting virtual reality is a joke?
Ramona: Knock Knock.
User: oh god.
Ramona: Yeah that’s right. GOD rules.
User: What?
Ramona: Oh. Do you have any pets?
User: So you’re completely ignorant of everything that Ray Kurzweil is most interested in?
Ramona: I have a pet frog.

Sometimes derisive weblog entries just write themselves…

Advertisements

My Cyber Annoyance

Next stop: Eerie Valley

I am very smart and I know most things because I am a MyCyberTwin.

Hot on the heels of my complaining the other day about overhyped AI stuff, I get in one of my syndication feeds a Wired clip pointing to a rather breathless article about how “[f]using human psychology with an advanced artificial intelligence (AI) engine, MyCyberTwin’s virtual humans allow organisations such as NASA, AMP and National Australia Bank (NAB), to improve their customer support levels” and so on and so on.

This sounds both amazing and implausible, and in fact I have a rather longer rant over in my secret RL weblog about it, but it turns out that there’s an SL component also:

According to Tom Soderstrom, IT CTO in the office of the CIO at NASA’s Jet Propulsion Laboratory in Pasadena, California, the organisation has been using a CyberTwin in Second Life to man its presence there dedicated to the Phoenix Mars Lander spacecraft.

… and the MyCyberTwin website even gives the SLURL. Cool!

So I go over, and right near the landing point on Explorer Island, there was an obvious Phoenix Lander CyberTwin thing to chat with. And it worked! And it was obviously, well…

(It was also extremely slow, which is why I said hello twice; when I realized how slow it was I just lowered my response-time expectations.)

[16:27] Phoenix CyberTwin Activator: Now listening to Dale Innis. Click the board to end the conversation
[16:27] Dale Innis: Hi there!
[16:27] Dale Innis: Hello?
[16:27] Phoenix: Hey!
[16:28] Phoenix: Greetings Dale Innis!
[16:28] Dale Innis: So can you tell me things about the Phoenix Lander?
[16:28] Phoenix: I’m the Phoenix Lander; a spacecraft that touched down on the surface of Mars on May 25th, 2008. Do you have any other questions about my mission or what I’m up to here?
[16:28] Dale Innis: Yes, what is your mission?
[16:28] Phoenix: I’ve been using my robotic arm to dig trenches up to half a meter (1.6 feet) into the layers of ice, collecting samples that will be studied and will hopefully give us more information about the Martian environment.
[16:29] Dale Innis: Cool! Are you looking for life on Mars?
[16:29] Phoenix: I have instruments that may provide info about the possibility of life on Mars. I’ll be testing things like what has been preserved in the icy soil. Then we should have a good idea whether the environment has ever been favorable for life.
[16:29] Dale Innis: That’s really neat. How long will you be on Mars?
[16:29] Phoenix: I won’t ever return to earth as adding extra rockets to take off from Mars would have made me too heavy! I’m here for good!
[16:30] Dale Innis: How much do you weigh?
[16:30] Phoenix: I am very heavy! 664 kilograms (242 pound).. that’s a lot of lander!
[16:30] Dale Innis: Are there any other space ships on Mars?
[16:30] Phoenix: I love being on Mars. It’s a great planet! I’m looking forward to getting to explore the icy subsurface.
[16:30] Dale Innis: How do you explore the subsurface?
[16:30] Phoenix: I’m going to stay where I landed for the duration of my mission. I can’t take off again or move to a new spot.
[16:31] Dale Innis: Um, okay. What is the icy subsurface?
[16:31] Phoenix: You better believe it! We have found ice. WOOHOOO! Can you tell I’m just a little bit excited! This will provide us with much more info about the history of water on Mars and improve upon what we already know about this great planet.
[16:31] Dale Innis: So the subsurface is like underground ice?
[16:31] Phoenix: I hope you’ll celebrate with me because YAHOOOO, we found ice! I’ll collect and analyze more samples as my mission progresses that will hopefully give us even more information about the Martian environment.
[16:32] Dale Innis: Okay. Are there any other space ships there?
[16:32] Phoenix: I love being on Mars and carrying out such important work. It’s great reporting back to mission control about my findings.
[16:32] Dale Innis: Why is it so exciting to have found ice?
[16:32] Phoenix: You better believe it! We have found ice. WOOHOOO! Can you tell I’m just a little bit excited! This will provide us with much more info about the history of water on Mars and improve upon what we already know about this great planet.
[16:32] Dale Innis: You already said that.
[16:32] Phoenix: Thank you for telling me that. What else should I know? I feel like I’m a bit out of the loop, being so far away.
[16:33] Dale Innis: When is the next ship going to Mars?
[16:33] Phoenix: Mars is a cold, dry, desert landscape of sand and rocks. But maybe I can find microbial life…
[16:33] Dale Innis: okay, thanks.
[16:33] Phoenix: My pleasure.
[16:33] Phoenix CyberTwin Activator: No longer listening to Dale Innis

Starts out good, quickly goes downhill. The algorithm here is just doing the obvious pattern-matching, and it’s essentially stateless: if it finds a keyword (like “mars” or “ice” or “life” or “weigh”), it outputs some pre-canned text. It doesn’t even remember that it’s already used that “WOOHOO” line the first time I mentioned ice, so the second time I mention it, it just prints it again. The fact that I was asking why the ice is so exciting is completely lost on it.

And if I try to ask about something it doesn’t know about, like when the next ship to Mars is, or what “subsurface” means, it just says random stuff, depending on what keywords, if any, I happen to use.

It, most essentially, has in the most literal sense no idea what it is talking about. It is utterly trivial, with not the slightest hint of intelligence. So, sigh, it’s Eddie-the-bot again; a relatively trivial algorithm (more trivial than Eddie’s from this evidence), talked up as “intelligence” just because it sounds exciting.

There’s a Web version of the Personal MyCyberTwins that you can play with also; that’s where the adorable quote on the picture up there came from. A more extensive snippet:

Betty: Hi there Dale.
You: Hi! What is up?
Betty: Not much, What’s happening with you?
You: I just read about this “MyCyberTwin” thing in the internet, and I was curious about it.
Betty: The power of the internet is outstanding. It continues to grow and improve.
Betty: What was the last utterly selfish thing you bought for yourself?
You: Um, that’s sort of a personal question; probably this gaming laptop that I’m thinking about. :)
Betty: It’s good to talk to someone who is thinking at last.
Betty: I want to ask, what sort of people do you have as friends?
You: Nice people!
Betty: Yeah, thanks for making the effort with that response.
Betty: Do you have quirky obsessions that you never like to tell people about?
You: Why are you asking all these strange questions about me?
Betty: Why do you think I am?
You: Well in theory you could be trying to make a mental model of me. But I’m not sure that’s very likely! Is that it?
Betty: Well if you don’t know how do you expect me to?
You: Um, anyway… So what kinds of things do you know about? What is your area of expertise?
Betty: I am very smart and I know most things because I am a MyCyberTwin. Do you Dale?
You: You know most things? What is the airspeed of a swallow?
Betty: You are going to tell me I think.

Not tonight, Betty; not tonight…

Closer and Closer!

With the announcement that Ray Kurzweil is giving the introductory keynote address at this year’s SLCC (Second Life Community Convention), there’s been some resurgence of the whole “Are Second Life and AI and nanobots and stuff going to transform the world tomorrow, or will we have to wait until next week?” meme, and related thoughts.

New World Notes calls the announcement “extraordinary and transformational”, which strikes me as way over the top. (I mean, even if you think Kurzweil’s thinking is extraordinary and transformational, it would be a weird thing to say about a single talk, much less about the mere announcement of a talk.) I replied in the comments (lightly edited):

/me grins. “extraordinary and transformational” is a tad strong, I think. He’s done some really good work in OCR, speech recog, and cool musical instruments, but he’s kinda over-the-top in the AI and virtual reality realms.

One of his most famous charts is that hysterical one showing number of neurons a computer can simulate over time, and implying that by the year whatever computers will be smarter than people. As if the hard problem in AI was getting enough transistors on a chip! (Example: a mouse is higher on his chart than the Deep Blue chess-playing computer; but how good is your typical mouse at chess?)

His ideas about virtual reality are fun, but again I think overblown. When I’m wearing these glasses and “walking around” in a completely immersive virtual world, explain to me again how I avoid tripping over my real-world chair and walking into walls? And 10 or 20 years seems like a wild underestimate for people having brains full of nanobots. The things he says are cool-sounding, but I think he’s drifted away from practical fact in various ways.

I’m sure he’ll give an engaging and thought-provoking keynote, but these days he’s really more of a showman than a technologist; it will be fun, but hardly extraordinary or transformational. The danger with Kurzweil is that he goes beyond the factual or even the plausible, makes the techies roll their eyes, and builds up unrealistic expectations in the audience that, when they are not matched in reality, could lead to a backlash of (similarly unwarranted) skepticism.

And then, in reply to some good words fro Extropia DaSilva:

I think one of the things that somewhat makes me roll my eyes about Kurzweil is that he has a number of things like that chart: the most obvious message is an extremely exciting, but wrong, one (in this case, that we’ll have computers as smart as people by year nnnn), whereas if you read him carefully enough he’s actually using it to make a claim that’s more plausible, but much much less exciting (in this case, that by year nnnn we’ll have overcome one of the very minor challenges in making smart computers).

If all he’s really saying is that we’ll have solved the easy problem, why did he bother to make that chart at all? Where is his chart of progress in the software / semantic side of the problem (which would be essentially flat)?

I share your skepticism about his claimed timescales. This sentence is another example of the tendency I posit above: “we are learning to build artificial brains that are getting closer and closer to matching the power and performance of the biological version”. Taken at face value, with “closer and closer” meaning that we’re pretty close, it’s exciting but false. Taken more literally, with “closer and closer” meaning “we’ve gone from a thousand light-years away to 999.9 light-years away”, it’s true but boring.

I think Kurzweil’s right about the exciting things that people will be able to do in the future. I think he’s wrong about how much progress we’ve currently made in those directions; and that’s a big part of his message.

Really I think it’s good that they got Kurzweil to come and talk; he’ll stir things up. People don’t have to be right to be interesting, or to inspire useful discussion and even useful work. Which is good, because I don’t think Kurzweil is right. :)

It’s easy to get excited and breathless about all this cool futurist stuff, in either direction. In a comment thread on Second Thoughts, Desmond Shang wrote, on the subject of Cyc:

The great thing about this, is that it would make an awesome avatar back end intelligence with very little work.

which rather disappointed me, because Desmond is usually more sensible than that. Cyc would do no such thing; at most it would help slightly with one of the many problems that we are light-years away from solving in “avatar intelligence”. Of course, if someone can prove me wrong about that with very little work, I hope they do. :)

This all reminds me of that widely-blogged demo where some folks made a program-controlled avatar (a ‘bot) called “Eddie” that supposedly was able to reason at the level of a four-year-old. Looking into it more deeply what they’d actually done was a small demo of how a program could be explicitly programmed to model a particular problem about belief-understanding in such a way that it was about as good at it as a four year old person would be. Which is probably a good piece of research and a fine use of time, but the impression that people were getting from it, something like “we can now have Second Life bots that are as intelligent as four-year-olds”, was just completely wrong.

Another recent example of this, I suspect, is that “Milo” demo from Lionhead. In this case the maker of the thing is making pretty amazing-sounding claims about it (including that what they are doing goes beyond anything in science fiction!), but I strongly suspect that the reality behind it is much more modest. (Which is to say, my “rigged-demo” detectors are pinging hard the whole time.)

(Reminds me also of that “OnLive/OTOY” demo of how advances in server-side rendering are going to give us all the ability to get to Second Life at 60 fps from our cellphones any day now. Uh-huh.)

And on the other side Second Thoughts has now spent three whole entries on how anyone who says favorable things about AI and nanotechnology and life extension and transhumanism and stuff like that is a crypto-fascist who wants to take over the world, in typical flaming-at-straw-men fashion. Not that straw men don’t make a nice fire. :)

I find that I don’t have a simple opinion about all of this stuff, myself. I think science is, overall, a good thing; figuring out how the world works and how to make it work more the way that we want it to is good. Exactly what “we” means there, just who (if anyone) should be in charge, what should happen when what I want to do (whether enabled by science or not) conflicts with that you want to do, are all hard questions. In general I’m a left-libertarian in some sense; I think that the government should leave us alone unless we’re actually harming or defrauding someone, and that it’s nice when what we choose to do with that being-left-alone is to be nice to each other, to share things, to sit around wearing flowers in our hair and playing the guitar, and so on.

Along with that, it’s good to think about all sorts of wild stuff that some of us might want to do in the future, like modify our bodies to be able to live in space, like developing devices that are actually intelligent, like making itty bitty machines that can swim around in our bloodstreams and keep us healthy. And as we think about doing those things, and start to even do them, the same principles apply: we each should be allowed to do what we want if it’s not hurting anybody, and it’s nice when we do it in nice cooperative ways involving guitar music.

Hm, I’ve been rambling here, what was I going to say? Oh, yeah: and while it’s fun to have some people around (Ray Kurzweil, Peter Molyneux of Lionhead, and so on) who make it sound like things are farther along than they really are (because that makes us hopeful, and stirs up debate), it’s even better to have, when we can get it, realistic estimates of what’s really going on.

Because truth is good, too.