Application of AI has made bird
watching quite interesting. Merlin (developed by Cornel lab) is quite amazing. Open
the app and switch on the listening option and it looks for bird call patterns
from myriad of noises and comes out with picture of the bird with common name and
scientific name as also the recording of the call for reference! Isn’t that
wonderful?! Birding is not going to be the same again. The other day I went for a late
morning walk nearby and kept the app on for half an hour or so, and it caught
many birds that I was also able to tag with calls but Shikra (Tachyspiza badia)
was quite a pleasant surprise (the number of raptors is going down at an
alarming rate) so was Jungle babblers (Yellow-billed Babblers are common in this habitat, while Blue-tailed Bee-eaters, the local migrants, are exiting with rising heat). Once alerted by Merlin I was searching nearest tall branches
for this now uncommon elusive bird. For a raptor it has a weak flight and so
was easy to spot as it flew away. This app is particularly useful when you hear
calls from thickets that makes it difficult to locate small sized birds. Too much
noise is a problem. Lacking civic sense people tend to put loud music and
whatnot that not only stresses other species but makes it impossible to record their
presence. Thankfully there are pockets of silence and where I am located has
high level of biodiversity with lots of trees and ponds, and relatively clean
spaces (unlike horrors of big cities). Few days back I was in a metropolitan,
with low air quality and persistent honking and noise, despite these I was able to listen
Brown-headed Barbet (Psilopogon zeylanicus). While travelling in train in isolated
patches I even spotted the charming Indian Roller (Coracias benghalensis), what
was once quite common in Bangalore-Mysore belt and now pushed into ‘Near Threatened’
in IUCN red list. So when the train stopped at isolated places along the western ghats I went to the exit to record bird calls. The app can also tag with binocular to identify birds realtime
and recognize photos taken. What is quite disturbing, and I take serious exception
to this, is that this app refuses to identify ubiquitous common crow calls. It seems crows
are not birds!
Putting Deep Learning into any
sensory device makes it AI. It’s like our senses and how brain interprets it.
Though deep learning neural network tries to emulates human brain it is nowhere near it.
It is a different kind of intelligence that is quite impressive. Again, I am quite
shocked that brilliant Transformer architecture that made generative AI
possible, hence significantly contributing to AI revolution that is unfolding
realtime, is being sidelined (and as expected some high-end Indian crabs are
also working so as to usurp AI space to represent India as entitlement, the pettiness is overwhelming
in this part of the world). It is simply miraculous human ingenuity how they
were able to bring meaning from words by embedding in number space. Predicting next
word into meaningful sentence and creating context for complexities and emergent
ideas is indeed brilliant. LLM is no longer restricted to language and is multimodal
hence encompass much more reality arbitrated by tapping consolidated human knowledge and abstractive emergent complexity herein.
Foundational Model therefore is significantly important to emergence of AGI. Hopefully
those who understand value of this brilliance recognize these. Whether transformer
architecture is replaced or not is not the point. What is clear is that it is
an important benchmark for AI development and significant step for computing
with lasting impact.
