atypicalset -- 亦云?

Tuesday, March 07, 2006

an enjoyable conversation with machines

A neat translation robot that exceeds expectation :)
He is at http://us.dict.cn/

Here is my conversation with its MSN variant:



Date
3/7/2006

Time
1:01:07 AM

from
xlx - *-less

to
dict.cn

Message
hello
3/7/2006
1:01:07 AM
dict.cn
xlx - *-less
['heləu, he'ləu] int. 喂,哈罗
3/7/2006
1:01:11 AM
xlx - *-less
dict.cn
how are you
3/7/2006
1:01:12 AM
dict.cn
xlx - *-less
你好
3/7/2006
1:01:21 AM
xlx - *-less
dict.cn
happy
3/7/2006
1:01:22 AM
dict.cn
xlx - *-less
['hæpi] a. 快乐的,幸福的
3/7/2006
1:01:24 AM
xlx - *-less
dict.cn
cool
3/7/2006
1:01:25 AM
dict.cn
xlx - *-less
[ku:l] a. 凉的,凉快的,酷 ad. 冷静地 n. 凉快 v. 使变凉 vt. 使变冷
3/7/2006
1:01:28 AM
xlx - *-less
dict.cn
fun
3/7/2006
1:01:29 AM
dict.cn
xlx - *-less
[fʌn] n. 乐趣,玩笑
3/7/2006
1:01:35 AM
xlx - *-less
dict.cn
郁闷
3/7/2006
1:01:36 AM
dict.cn
xlx - *-less
[yù mèn] vapor
3/7/2006
1:02:17 AM
xlx - *-less
dict.cn
news?
3/7/2006
1:02:19 AM
dict.cn
xlx - *-less
新闻中心-新闻要闻
3/7/2006
1:02:19 AM
dict.cn
xlx - *-less
上海市长称迪斯尼乐园筹建工作正在进行 (http://news.sina.com.cn/c/2006-03-07/12158382708s.shtml)
吴敬琏:政府资源配置权力过大阻碍经济体制转变 (http://news.sina.com.cn/c/2006-03-07/11209286726.shtml)
教育部谈助学贷款问题将曝光违约生名单 (http://news.sina.com.cn/c/edu/2006-03-07/10458381713s.shtml)
政协委员质疑大学高收费:学费20年上涨25倍 (http://news.sina.com.cn/c/2006-03-07/10398381642s.shtml)
3/7/2006
1:03:33 AM
xlx - *-less
dict.cn
bye
3/7/2006
1:03:34 AM
dict.cn
xlx - *-less
[bai] int. 再见

"Dear Amazon.com Customer," ... an acute dumb outcome of machine learning

This is not a technical note regarding machine learning.

"Dear Amazon.com Customer,"
"We've noticed that customers who have purchased Halo: Combat Evolved are also interested in The Godfather (DVD-ROM) for the Windows XP. For this reason, you might like to know that this game will be released on March 21, 2006. You can pre-order your copy by following the link below."


I am truly impressed that machines accurately capture this correlation to intuitive typical human nature. However the conclusion is wrong, as it so focuses on the data that it blatantly ignores:

1) Compared to the rest of a user's entire profile, a few entries in the transaction history can be outliers.
2) Predictions should be context-dependent, where context can change, for the good or bad.

After all, machines doesn't know all, and they aren't smart enough yet to fortune-tell on people -- it'd be boring / sad if they are.