As some non-AI savvy individual, I take this headline to mean "what I think will happen in the future". But then after reading, it turns out you are asking me very narrow questions having nothing to do with me, and everything to do with AI. So .... I - do - not - get - it.
Sorry if this is going over my head, but its on the front page, and I always take time to upvote and view "review my _" posts. But this one I am completely in the dark as to why its only about AI. What if I don't care about AI? Maybe just be specific as to your motives on the front page then.
Introduction: The Uncertain Future is a future technology and world-modeling project by the Singularity Institute for Artificial Intelligence (http://www.singinst.org). Its goal is to allow those interested in future technology to form their own rigorous, mathematically consistent model of how the development of advanced technologies will affect the evolution of civilization over the next hundred years. To facilitate this, we have gathered data on what experts think is going to happen, in such fields as semiconductor development, biotechnology, global security, Artificial Intelligence and neuroscience. We invite you, the user, to read about the opinions of these experts, and then come to your own conclusion about the likely destiny of mankind.
Unfortunately I had to force-quit Safari after about 3 minutes of beachballing. Looks like your Java Applet had some trouble loading. This is on Mac OS X 10.6.2, Safari 4.0.4.
The problem is, you've created both a bayesian decision-making tool AND a dataset for the tool for estimating human-level AI. Forcing a user to learn both of these at the same time is asking too much from your users.
Also, your thesis is clearly "nothing matters about the future other than human-level AI." This is a defendable thesis, but you can't use this as a starting assumption without confusing your users immensely. Maybe you should also ask questions about other things and then have them discover the importance of human-level AI naturally. (assuming it really is as important as you think.)
Finally, your java-ish implementation is really buggy and the implementation/documentation in general needs a lot of work. I couldn't get the tool to really work at all on Ubuntu with Firefox 3.5 (The edit fields keep freezing up)
(Dollars to Donuts, the java applet is written in a Lisp dialect and then translated/compiled into a very fat java applet :-)
Also: I found that almost all my answers to the questions require highly lopsided distributions, which the app doesn't support... this makes it hard to get meaningful answers.
Worked well for me (shocking for Ffox 3.5 on Ubuntu). I'm particularly impressed with the interface for manipulating the distributions directly by interacting with the graph.
However, the 'expert opinions' are not very discoverable (they were off the screen for me). Perhaps offering them on the side could work? (Since the input data is pretty fuzzy, the graphical area doesn't need to be so large).
Also, I find the 'rate-increase' graphs are less intuitive than the normal distribution graphs. Unfortunately, I don't know how to improve upon them.
Really cool. Would be interesting to break some of these things down further, into easier-to-estimate probabilities -- probably using some sort of more detailed graphical model. Seeing some sort of consensus view of the future according to your users would be neat, too (especially if it were broken down by demographic).
The interface is a little quirky. I sometimes found it difficult to determine how to set means / deviations.
This isn't a webapp, it's a wizard for naive prediction of the date of the singularity (you seem to mix AI with a very narrow form of IA). You're not quite "modeling" a future world. Java sucks for this - why use it, given the current state of javascript and canvas?
This would make an interesting blog post if you distributed the various "viewpoints" on the various questions into a bunch of images.
What is the goal of my interaction with this? You're presenting me with a very short blob of technical gibberish on each item and then you're essentially asking me to guess at the answer.
The best-case result of this is a scientifically (i.e. in the sense of empirical social sciences) awkward set of answers but I fail to see how they would carry any predictive value.
Just because everyone believes the world will be hit by a meteor in 2012 does not make it happen.
19 comments
[ 2.5 ms ] story [ 51.9 ms ] threadSorry if this is going over my head, but its on the front page, and I always take time to upvote and view "review my _" posts. But this one I am completely in the dark as to why its only about AI. What if I don't care about AI? Maybe just be specific as to your motives on the front page then.
- maybe making AI explicit in the title would be good.
Sorry, guess I'm not the target.
Second, I didn't understand the questions and answers even though I know AI techniques quite well.
The problem is, you've created both a bayesian decision-making tool AND a dataset for the tool for estimating human-level AI. Forcing a user to learn both of these at the same time is asking too much from your users.
Also, your thesis is clearly "nothing matters about the future other than human-level AI." This is a defendable thesis, but you can't use this as a starting assumption without confusing your users immensely. Maybe you should also ask questions about other things and then have them discover the importance of human-level AI naturally. (assuming it really is as important as you think.)
Finally, your java-ish implementation is really buggy and the implementation/documentation in general needs a lot of work. I couldn't get the tool to really work at all on Ubuntu with Firefox 3.5 (The edit fields keep freezing up)
(Dollars to Donuts, the java applet is written in a Lisp dialect and then translated/compiled into a very fat java applet :-)
However, the 'expert opinions' are not very discoverable (they were off the screen for me). Perhaps offering them on the side could work? (Since the input data is pretty fuzzy, the graphical area doesn't need to be so large).
Also, I find the 'rate-increase' graphs are less intuitive than the normal distribution graphs. Unfortunately, I don't know how to improve upon them.
The interface is a little quirky. I sometimes found it difficult to determine how to set means / deviations.
This would make an interesting blog post if you distributed the various "viewpoints" on the various questions into a bunch of images.
The best-case result of this is a scientifically (i.e. in the sense of empirical social sciences) awkward set of answers but I fail to see how they would carry any predictive value.
Just because everyone believes the world will be hit by a meteor in 2012 does not make it happen.