As an engineer, I advocate a functionalist approach to assessing "understanding" in which "understanding" is measured in degrees and capabilities, rather than as a binary predicate. 2/
-
-
Prikaži ovu nit
-
Today's AI systems exhibit some understanding. For example, when I tell Siri “Call Carol” and it dials the correct number, you will have a hard time convincing me that Siri did not understand my request. 3/
Prikaži ovu nit -
But as many folks have pointed out, today's AI systems lack many forms of understanding. Such critics often claim that Siri's understanding is not "true", "real" or "genuine". But I argue that it is "real", and it can be measured functionally. 4/
Prikaži ovu nit -
Why does this matter? Because the CS community has a history of dismissing advances in AI as not being "real" and declaring the entire AI enterprise to be a failure. Funding dries up, and research progress slows. 5/
Prikaži ovu nit -
Today, despite the critics, AI R&D continues to deliver useful functionality to users. So let’s stop dismissing our successes as “fake” and not “genuine”, and let’s continue to move forward with honesty and productive self-criticism. 6/
Prikaži ovu nit -
Speaking of productive self-criticism. When we demonstrate a new capability, let's also demonstrate its limits by showing instances of functional failures. Much of the criticism is a reaction to the hype generated by researchers and the organizations they work for. 7/
Prikaži ovu nit -
It is important, both for intellectual honesty and for continued funding, that AI researchers claim credit for our successes and take ownership of our shortcomings. end/
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
-
-
reg. your example, Who did IBM’s Deep Blue system defeat?, i would argue (ala
@GaryMarcus) that it is not true understanding. Just try rephrasing the question: Who defeated deep blue? -> I get the answer "IBM". if it understood the previous qn, it shouldn't make that mistake 1/ -
current QA systems are just picking up the stastical correlations without understanding the question or context, This has been shown in prior work on robustness of current QA systems (
@percyliang &@robinomial). 2/ - Još 4 druga odgovora
Novi razgovor -
-
-
In the tradition of Drew McDermott's "How Intelligent is Deep Blue?"--that I keep making my Intro
#AI students read.. http://www.nyu.edu/gsas/dept/philo/courses/mindsandmachines/Papers/mcdermott.html … -
I hadn't seen that. It is wonderful! Thanks
- Još 1 odgovor
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.