While monitoring the logs, I’ve encountered an interesting URL being examined by Googlebot. It’s a type of pattern I’ve been seeing for last several months and decided to make a comment about it today. There’s a very little doubt that Googlebot is partially driven by machine learning algorithm, and for some reason it’s querying a date of May 14, 2143 on this instance. It’s targetting a Read the Bible in a Year service area I’ve created for my church few years ago at https://kumcabq.org/dailyreading. Apparently, Googlebot has figured out that it could traverse through time and started to query what seems to be arbitrary dates. Here are some dates it has queried in the wee hours of August 12, 2019, in the descending order of query times:
Okay, on the second look, it is actually traversing through one day at a time for years 1865, 2132, 2125, 2026, 2034, 2069, 2025, 2033, and so on. The log has many more dates. It almost looks like an artificial intelligence trying to figure out what correlation, if any, is there between the date and the Bible passages presented on the page. This seems to have gone on for a while. This is just taking up unnecessary bandwidth… not sure what to do — slap the Googlebot’s hand when it gets to this page again, or maybe take the page down?
… is like a nuclear weapon with no restraint. We probably need to be more concerned with the direction of knowledge, which is usually guided by a sense of morality in humans than simply allowing machines to obtain knowledge without such guidance system.
But it is the spirit in a person, the breath of the Almighty, that gives them understanding. It is not only the old who are wise, not only the aged who understand what is right. – Job 32:8-9
The current visual system is a dumb visual recognition system that mostly sees the world in 2D. The depth perception and the information about layers of things in between the target object and the subject has not been considered in the current models. There are phones equipped with infrared-ray sensors for this very purpose. Google’s 3D-sensing Tango project is another example. However, a visual recognition system seeing a stop sign with a Post-It note and recognizing it as a speed limit sign is a huge problem for the current models. I think this is another reason why we need to fundamentally reconsider presuppositions about how a typical neural net works. Just like the late introduction of epigenetics, there is more involved than just synapses, and neurons.
The current models deal with images fixed in time. The input data is usually too flat, so the incoming data needs to be amalgamated with more metadata to make it multi-dimensional. A parked car at a certain location is a car, but once it leaves the parking lot at one point, the image is now back to parking lot. Current models do not account this as a “missing” car. Likewise, the Post-It note should be recognized as an “appended” object to a pre-existing object in a time past. Human visual system would not be able to recognize anything if it did not scan constantly, or things were static and not moving. There’s also a factor of expectation as we anticipate in motion.
One of the simplest solution to the stop sign issue could be to train the system to pull out the best expected target, and deal with the delta between the actual target and the expected target.