[ExI] AI milestones

[Anders Sandberg]

On 01/03/2012 05:51, Kelly Anderson wrote:
> Asimo 

Hmm, the first walker might be a milestone. But what is Asimo besides 
amazingly cute?


> Expert systems for some kinds of medical diagnosis were perfected long 
> ago. Statistical methods for treating patients with heart attacks 
> (discussed by Seth Godin, not sure how much AI was in there)

Some of the most amazing statistical methods for treatment are utterly 
simple, just a three level decision tree. Here we have a case of 
interesting and important work in AI/machine learning producing a 
conceptually important result, but the result itself is trivial. Goes to 
show that AI breakthroughs can be simple.

The real question is how good the average expert system is. Presumably 
it has to be cost-effective, which leads to the question of how common 
they are in different domains.


> The recent flying machines with dynamic controls that learn to fly 
> through hoops certainly should make it onto the list, as well as the 
> first bipedal robots of a few years back.

Why are the flyers worth to be on the list? (still trying to figure out 
the criteria myself)


> There are commercial OCR engines that scan well over 99% accuracy for 
> typeset text... so I'm thinking that the problem is old newspapers and 
> bad image acquisition here... I've seen some marks in the 80% range 
> for hand printed text, for heaven sake!!

Any references would be welcome. Yes, it was old newspapers.

-- 
Anders Sandberg,
Future of Humanity Institute
Philosophy Faculty of Oxford University