Wired magazine has an article on the cost of AI to play games - particularly Go - and what that suggests for the future.
Wired magazine wrote:
DeepMind, likely the world’s largest research-focused artificial intelligence operation, is losing a lot of money fast, more than $1 billion in the past three years. DeepMind also has more than $1 billion in debt due in the next 12 months.
The writer notes that current AIs are very good at functioning within precisely defined domains, but seem unable to generalize sufficiently to transfer their skills to other domains:
Wired magazine wrote:
What works for Go may not work for the challenging problems that DeepMind aspires to solve with AI, like cancer and clean energy. IBM learned this the hard way when it tried to take the Watson program that won Jeopardy! and apply it to medical diagnosis, with little success. Watson worked fine on some cases and failed on others, sometimes missing diagnoses like heart attacks that would be obvious to first-year medical students.
Fun citation from that webpage: "Go—neither the board nor the rules have changed in 2,000 years". Uhm.
Other fields may have AI difficulties in particular because
- the fields are more complex or complicated in their description than go,
- errors in AI go might occur but so far it seems only for rare positions (e.g. highly multiple kos) irrelevant in playing practice; errors in other fields might occur more easily or have more dramatic consequences (such as killing human beings).
The complexity will be overcome to the extent of letting AI learn like a human being during his life and then beyond. However, all errors can never be avoided because of extreme decision complexity.
Joaz Banbeck wrote:Wired magazine has an article on the cost of AI to play games - particularly Go - and what that suggests for the future.
Wired magazine wrote:
DeepMind, likely the world’s largest research-focused artificial intelligence operation, is losing a lot of money fast, more than $1 billion in the past three years. DeepMind also has more than $1 billion in debt due in the next 12 months.
The writer notes that current AIs are very good at functioning within precisely defined domains, but seem unable to generalize sufficiently to transfer their skills to other domains:
I hope they are good at printing money.
The Adkins Principle: At some point, doesn't thinking have to go on?
— Winona Adkins
It's worth noting that the author, Gary Marcus, is pretty much the world's preëminent deep learning skeptic. I started reading the article and wondered when the inevitable quote from Gary Marcus was going to show up... and then saw that the article was written by Gary Marcus. Anyway, he's a smart and knowledgeable guy, but he has a definite agenda, so if you want a balanced appraisal of the current state of deep learning, you may want to read some other articles as well.
dfan wrote:It's worth noting that the author, Gary Marcus, is pretty much the world's preëminent deep learning skeptic. I started reading the article and wondered when the inevitable quote from Gary Marcus was going to show up... and then saw that the article was written by Gary Marcus. Anyway, he's a smart and knowledgeable guy, but he has a definite agenda, so if you want a balanced appraisal of the current state of deep learning, you may want to read some other articles as well.
As far as I understand, his skepticism is about whether the current approaches in Deep / Reinforcement Learning have anything to do with "real AI" or not.
Researchers in machine learning now often ask, “How can machines optimize complex problems using massive amounts of data?” We might also ask, “How do children acquire language and come to understand the world, using less power and data than current AI systems do?” If we spent more time, money, and energy on the latter question than the former, we might get to artificial general intelligence a lot sooner.
Or not. That's a huge if. Besides, what's wrong with asking both questions and others concurrently ? ( He mentioned limited resources as one argument, which is kind of a given. ) I thought some folks ( at IBM? ) have been researching to simulate a small brain ( like a mouse's ) already. People are working hard, on multiple fronts.
dfan wrote:It's worth noting that the author, Gary Marcus, is pretty much the world's preëminent deep learning skeptic. I started reading the article and wondered when the inevitable quote from Gary Marcus was going to show up... and then saw that the article was written by Gary Marcus. Anyway, he's a smart and knowledgeable guy, but he has a definite agenda, so if you want a balanced appraisal of the current state of deep learning, you may want to read some other articles as well.
As far as I understand, his skepticism is about whether the current approaches in Deep / Reinforcement Learning have anything to do with "real AI" or not.
The search for intelligent life on earth.
The Adkins Principle: At some point, doesn't thinking have to go on?
— Winona Adkins
If the goal is to make money, the researchers better emulate man's capacity for taking investor capital to a level where it's too big to fail, than man's intelligence.
