Explaining Me-Too, Copycatting, and What Might be Theft
August 10, 2021
Consider TikTok. LinkedIn has found itself behind the video résumé eight ball. The Google, not to be left out of the short video game, is rolling out YouTube Shorts. Despite the hoo-ha output on a podcast with two alpha wizards, the Shorts thing is a lot like TikTok. Instead of China’s watchful eye, the Google just wants to keep advertisers happy, really, really happy. Which is better? Wow, that’s an interesting question when one defines “better.” I don’t know what better means, but you may.
I read “Is Iterated Amplification Really More Powerful Than Imitation?” For me, the write up is a logical differentiation within the adage “If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck.” Yep, that smart software has identified actor A and a match for a known bad actor. Works great. Sort of.
However, the duck analogy is of little utility in the work from home, thumb typing world. Thus, enter the phrase iterated amplification. The euphony is notable, but I am a duck kind of person.
The write up explains in a way which evokes memories of my one and only college course in semantics. Who taught this gem? A PhD from somewhere east of Russia who aspired to make the phoneme –er his life’s work. The –er in which he was planning to leverage into academic fame was its usage in words like hamburger. Okay, I will have a duckburger with pickles and a dash of -mustard. So it is not me-too; it is imitative, iterated amplification. (And the logical end point is similar, if not almost identical systems. A bit like ducks maybe?)
I learned in the essay:
iterated amplification is not necessarily the most efficient way to create a more powerful AI, and a human mimics would have the flexibility to choose other techniques. Current AI researchers don’t usually try to increase AI capabilities by iterated amplifications, but instead by coming up with new algorithms.
“New algorithms?” I thought today’s smart software recycled code chunks from open sources, friendly smart cloud outfits like Amazon, Google, IBM (excited am I), Microsoft, and others.
The innovation might be interpreted as playing with thresholds and fiddling with the knobs and dials on procedures explained in class or by textbook writers like Peter Norvig. Lectures on YouTube can be helpful too. Maybe what works best is a few smart interns who are given the message: Adjust until we get something we can use.
Duck analogy: Google’s DeepMind has been me-too’ed by other outfits. These outfits have mostly forgotten where the method originated. Upon inspection, the method may have been the outcome of a classroom assignment.
That’s why facial recognition systems and other applications of smart software often generate me too, me too outputs; that is, misidentification is more reliable than identification. With recognition ranging from 30 percent confidence to 90 percent confidence, there’s some room for error. Actually, there’s room for darned crazy errors and fascinating secondary consequences.
Just admit that “innovation” is not much different from a duck. And imitation is less costly than doing the original thinking work. Revenue, not bright ideas, are more reliable than cooking up a power from the air scheme.
Stephen E Arnold, August 10, 2021
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
