+1 to AI not being analogous to statistics.
Rather than view these terms: Data Science, Machine Learning, or Artificial
Intelligence, as interchangeable, I see them as existing on a spectrum.
Data Science produces Insights, it still requires a scientist to interpret
data and have subject expertise. Somewhere in the middle is Machine
Learning, which Predicts. Programmers who train algorithms are considered to
"harness the power of AI", but really, AI is Actionable.
Self-driving cars are a good example of the three concepts: one might train
an algorithm to recognize stop signs. The ML aspect can then attempt to
Predict whether an image is a stop sign or not. In practice the car uses AI
to recognize the predicted stop sign in the world, to apply brakes
considering the road, weather, and brake conditions, and make decision based
on the Machine Learning predictions. A Data Scientist may study the
collected data from the car, in order to produce Insights about accuracy of
the stop sign prediction algorithm. They may notice that Stop signs are
being missed at certain times of day, and discover the training set is all
during bright daylight hours, and can then retrain the ML algorithm.
In this sense, AI isn't really another name for statistics. Statistics is
still statistics, which can be utilized for many purposes including training
algorithms, or creating decision making processes. The old saying is:
When Fundraising, it's "AI",
When Hiring, it's "ML",
When Implementing, Linear Regression,
When Debugging, printf()
Thanks,
-Sean
Digital Innovation & Information Architect
CSU-Global
-----Original Message-----
From: Code for Libraries [mailto:[log in to unmask]] On Behalf Of Péter
Király
Sent: Tuesday, December 11, 2018 7:11 PM
To: [log in to unmask]
Subject: Re: [CODE4LIB] ai in libraries
""AI" now is just another name for statistics and statistics (especially in
disguise) is very dangerous given that untrained human beings are very bad
at statistics."
I agree with the second part of this sentence, but I don't think that AI is
just relabelling statistics. There are lots of discussion about it, my
favorite paper on this is David Donoho's 50 years of Data Science
(https://courses.csail.mit.edu/18.337/2015/docs/50YearsDataScience.pdf).
Regarding to metadata: I think there might be lots of cases, where we could
use metadata as labels in supervised learning or as a gold standard in
information extraction experiments, however I never run any real test, so I
can imagine different scenarios where it may or may not be misleading (I
hope that properly curated metadata would more often be proved to be
unbiased than not).
Best,
Péter
Chris Gray <[log in to unmask]> ezt írta (időpont: 2018. dec. 11., K,
17:34):
>
> 1. I would add that what is called "AI" nowadays is not what was meant
> when the term was invented. "AI" now is just another name for
> statistics and statistics (especially in disguise) is very dangerous
> given that untrained human beings are very bad at statistics. I would
> rather that people spent less time promoting "AI" and more time
> becoming aware of cognitive biases and how they infect everything we
> think. I love the book title "Don't believe everything you think".
>
> 2. We are not even close and most of that technology should be
> classified as torturing the data until it confesses. I was very
> disappointed when I started to read a book on deep learning with
> Python and discovered it was just about choosing canned statistical
> analyses to get what you were already looking for. I recommend
> googling for tests that show that Alexa, Siri, and Hey, Google aren't
> as good as the commercials might lead you to believe, or why
> self-driving cars are not coming any time soon.
>
> 3. Technology is no substitute for serious human thought. Metadata
> will not save us. On this Cory Doctorow's Metacrap is my touchstone.
> Also Fred Brooks's "No Silver Bullet" was right in 1986 and it is still
> right.
>
> I'm not saying that techniques that come out of AI research aren't
> worth using, but that we should use them instead of being used by them
> and those that hype them.
>
> Chris
>
> On 2018-12-10 4:52 p.m., Eric Lease Morgan wrote:
> > Last week I attended an artificial intelligence (AI) in libraries
> > conference, and I've written the briefest of travelogues. [1] Some of my
> > take-aways include:
> >
> > 1. Machine learning is simply the latest incarnation of AI, and
> > machine learning algorithms are only as unbiased as the data used
> > to create them. Be forewarned.
> >
> > 2. We can do this. We have the technology.
> >
> > 3. There is too much content to process, and AI in libraries can
> > used to do some of the more mechanical tasks. The creation and
> > maintenance of metadata is a good example. But again, be
> > forewarned. We were told this same thing with the advent of word
> > processors, and in the end, we didn’t go home early because we
> > got our work done. Instead we output more letters.
> >
> > 4. Metadata is not necessary. Well, that was sort of a debate,
> > and (more or less) deemed untrue.
> >
> > If you want to participate in AI for libraries-like discussions, then
> > consider subscribing to ai4lib.
> >
> > [1] travelogue -
> > https://sites.nd.edu/emorgan/2018/12/fantastic-futures/
> >
> > [2] ai4lib – https://groups.google.com/forum/#!forum/ai4lib
> >
> > --
> > Eric Lease Morgan
> > University of Notre Dame
--
Péter Király
software developer
GWDG, Göttingen - Europeana - eXtensible Catalog - The Code4Lib Journal
http://linkedin.com/in/peterkiraly
|
