AIs are Ruining Everything Part 1: Secret Animal Languages

Artificial intelligence (AI) has not quite advanced to the point where you could speak complete sentences and have them translated to your dog. In fact, animal language is a bit more complicated than "speech" alone and is often a combination of behavior and sounds. So, of course as technology improves, there is an effort to understand the world around us. AIs are a core component of Machine Learning (ML) algorithms and models, and in this case are being used to better understand animals. More specifically, there are five groups of researchers using machine-learning algorithms to analyze the calls of a select few mammals.

If you have worked with mathematical models, then you know that in most cases, the AI needs training data to learn from and the model is then refined as the AI gains experience through new data. So, in order to build a good mathematical model, there has to be plenty of good data for the model to "learn" and then be tested on. AI-based models are unique in that they can continue to learn as new data points are provided. The crux of the issue in the case of animal language translation is that scientists are not 100% sure of all of the data points that need to be collected in order to adequately make predictions. It's similar to trying to predict the occurrence of breast cancer given grocery related data - is it possible? Sure, it's possible. Will it be easy? Probably not, especially with no available background research on shopping patterns of those with breast cancer versus those without. Without that sort of information, it is difficult to determine what is useful/helpful information and what is not.

While we are pretty far off from having "Google Translate''-like phone applications for animals, some of these models are currently accurately able to detect the presence of a specific species of animal by call. Don't believe us? Test out Connell's Merlin app and let us know how it does in the field if you're an avid bird watcher or ornithologist. While it's not quite a translator it is a good step one; you wouldn't want your translation software replying in Spanish to someone speaking Italian, right? You would want your translation software to correctly recognize the language being spoken as Italian the first time. That being said, detection of different animal languages is a good step one. If this feat still sounds easy and not at all ambitious, here is one last interesting piece about the development of a model that can translate animal sounds: language.

Analyzing language is far different from analyzing "regular" data points. Now add the sound of spoken language to the mix and that's more information for the model to ingest. Now instead of static, hard coded data points the data points must be quality captured vocal recordings. For animals, in addition to the vocal queues, the behaviors are important in communication as well so a video component must also be analyzed. Again, this is something completely doable, but it's about as difficult as trying to separate sand by hand. Although in this case, the hand never gets tired and is aided by tweezers, building a model to translate animal languages will by no means be a walk in the park. Hopefully within the next decade or so we will see the first versions of an animal translator app, but only time will tell.

For the latest in news and stock picks, don't miss our podcast at https://www.arbitragetrade.com/podcasts/