Ray Kurzweil Is Building Your ‘Cybernetic Friend’ | Hacker News Reader

Ray Kurzweil Is Building Your ‘Cybernetic Friend’

techcrunch.com

38 pointsllambda27 comments

You know, that friend who you're nice to and all but deep down you don't really trust, but you don't want to unfriend because that could set them off and you suspect they could do you great harm, and might just for the hell of it. Your, not quite a psychopath friend but with enough clues that you find yourself wondering how to create an opportunity to add distance without calling attention to yourself kind of friend? I can totally see it.

A bunch of mindless jerks who'll be the first against the wall when the revolution comes.

"Your plastic pal who's fun to be with"? Never!

"Go stick your head in a pig."

Yeah. Your friend who you suspect may be able to read your thoughts and predict your actions so you have to basically wipe your own memory each time you have a malicious thought.

Imma switch back to AOL now.

There's plenty of joking and 'it'll never work' here, but doesn't this worry anyone at all?

A system that 'knows you better than you know yourself', means a system that is able to predict your responses to different stimuli more accurately than you can.

By definition such a system can manipulate you.

Let's set aside the idea of an 'AI' - which conjures up all sorts of mythical images of talking computers.

Instead consider the implications of a single corporation with a machine learning system that has access to the private communications, personal documents, and physical location and web viewing history for a billion people and can use that information to predict the behavior of each one of them better than they can predict their own behavior.

That's what we'll get if Kurzweil and Google are only partially successful in this endeavor. It sounds to me like the greatest concentration of power in human history, and an extremely dangerous thing that we should be considering the implications of today.

Now many people seem to think that Moore's law means that machine learning systems of this power are inevitable and that it's better that it's Google than a worse entity like a government who controls it.

I think that's a dangerous dismissal for two reasons - firstly it presumes that this tool wouldn't be taken over by governments once it came into existence, or that it wouldn't be used to manipulate governments to prevent its takeover.

Secondly, even if super-powerful AI is an inevitability, it's not inevitable at all that we should be routing all of our personal information to its creator so that it can be used to model and predict our behavior en masse.

> Instead consider the implications of a single corporation with a machine learning system that has access to the private communications, personal documents, and physical location and web viewing history for a billion people and can use that information to predict the behavior of each one of them better than they can predict their own behavior.

Funny the corporation that makes that more visible to me is Netflix. After rating so many movies it can pretty well predict if I would like a move or not. It does it better than me reading the description and the title.

It is kind of nice but a bit worrying. Wonder what kind of psychological profile it could build and sell to other companies if it wanted to.

"It fundamentally deals with language." Language, Kurzweil argues, is the window to creating a genuine artificial brain, that can understand the meaning of ideas and concepts.

This sounds like the same old intelligence-as-symbol-manipulation bullshit of the 70s. Language is not thought, as anyone who's spent time with animals or children or introspecting their own mind would know. A "genuine artificial brain" would mean something that can learn language itself if exposed, not something built around linguistic concepts.

This has a flavor of reviving some ideas on how search will change that had been popular in the late 1990s / early 2000s. There were a number of proposals that "search engines" as a tool that users use to query for information should be replaced with a paradigm of "software agents" that build a model of what the user is looking for or trying to do, and which then roam around the web on the user's behalf. Here's a survey of some of those proposals from a 1999 Nature article: http://www.nature.com/nature/webmatters/agents/agents.html

This isn't meant to be a criticism, because implementation makes a huge difference, so it's quite plausible that a proposed-and-didn't-catch-on idea could catch on in another form. For example, Google has much better profile data on you than late-'90s research systems were able to collect. But I think looking at the previous iteration of attempts is still interesting.

Its not that the idea didn't catch on, its that they were unsuccessful. Just like most attempts that aimed towards 'strong' or general artificial intelligence were unsuccessful. Natural language understanding is not necessarily general intelligence but it is closer.

The difference is not so much about the profile data that Google has now but rather the AI approaches which are available. For example, the type of hierarchical hidden Markov models that Kurzweil goes into detail on in his book have a broader application than most AI approaches previously used. Also the body of text which Google can process and the amount of processing power Google has contribute to making the goal feasible.

"who knows users better than they know themselves"

I'm hoping that, in the future, data collection services like Google will allow users to view more of their own data, not less.

> “I envision in some years that the majority of search queries will be answered without you actually asking,”

It would be of particular to entrepreneurs, exactly how search constrains thought. Vernor Vinge has pointed out that search is the proverbial street lamp in the joke about searching for your keys. Figuring out what it's hard to search for would be a way of finding the unturned stones.

Hmm,

When you put it that way, it sounds unlikely.

I think neuroscience has found that while the potential for an action begins before a person realizes it, the fully formed action only happens in your hands or your mouth as you type or say something.

Anyway, something that somehow predicted even most of things you query for wouldn't feel terribly pleasant and probably wouldn't get all of your choices.

A search engine that could let say "word1 is a category, word2 is an adjective.." etc but that would require more, not less thought.

Really, you could talk to Google in full sentences "like a person", would you want to? For most your searches would you want to? For some of them, yes.

Hmm...

It's my belief that the first computer programs that we would consider a real AI have already been built, and quite recently. Somehow or other, when enough information is finally available for such a breakthrough, there is a kind of 'leakage' that happens into the public domain. News articles, science and tech articles in particular, begin to converge. From this 'leakage' it is possible to piece together how such a thing might be accomplished. And usually, within five years or so, a big announcement is made by a big company. The funny thing is, I can already see what the big limitation of AI is going to be. It's not going to draw up plans for a warp drive or cure cancer any time soon, as some have imagined. Instead, we are going to learn something very interesting about the current rate of scientific progress.

>Instead, we are going to learn something very interesting about the current rate of scientific progress.

...Which would be?

...that it isn't limited by how 'smart' we are. That's the cheap answer to your question of course. The more expensive answer relates to the role of creativity in scientific advance, and how creativity is limited by experience. When you look at the history of scientific advance, very rarely are discoveries made out-of-time. They occur when the conditions are right. They rarely occur sooner because they are meaningless without the context of the times in which they were developed. So, scientific advance is essentially limited by us humans and the rate that we progress as a race. In other words, the rate of scientific advance is just about optimal, or approaching a limit set by us. :-)

Eric Schmidt: "“Is it a ‘hot dog’ or a ‘hotdog.’ And, if you knew something about whether the person had dogs, or whether the person was a vegetarian, you’d have a very different potential answer to that question.”"

The problem with this semantic search "problem solving" is that a vegetarian who had a dog is still searching for the food. If their dog was overheating they would search "What to do about an overheated dog".

Conclusion: This isn't the problem they are trying to solve.

It sounds like an extension of Google Now to me. I was wondering what the big G was planning on using Kurzweil for, I wonder how his singularity shtick is related to this.

Pretty obvious - modeling your interests is (in the limit) equivalent to modeling you.

Microsoft tried that in the 90s.

The result was Clippy.

Do we have any privacy in future?

Do you have any privacy now?

[1] http://datasift.com/platform/data-sources

[2] http://gnip.com/sources/

[3] http://klout.com/how-it-works

[4] http://en.wikipedia.org/wiki/ChoicePoint

[5] http://www.lexisnexis.com/backgroundchecks/

[6] http://en.wikipedia.org/wiki/Trailblazer_Project

yes. In a cave, a jungle, a desert or a glacier or moving from place to place every few hours.

Either that or right in front of everyone (grey man theory).

Its possible to have privacy, but you will have to put up with a little inconvenience and live a different life to others.

I have considered this myself to be honest - it would be an interesting way to live.

I can imagine the Kurzweil/Google press conference introducing your new "cybernetic friend" going something like this...

http://www.youtube.com/watch?v=IKBJxZf-Dgs

If they make it look like ED-E[1] then I'm sold..

[1] http://fallout.wikia.com/wiki/ED-E_(Lonesome_Road)

Nope. They've already got a prototype running: https://en.wikipedia.org/wiki/Office_Assistant

Oh god, I clicked on that without thinking about what it was, and pretty much spit out my drink.. thanks..

Reply on news.ycombinator.com