Pet owners are a special breed of people, they willingly take responsibility for another life and take care of them. I personally like pets as long as I can return them to the owner at the end of the day (or hour, depending on how annoying the pet is). I had to take care of a puppy for a week when Surabhi & Vinit were out of town and that experience was more than enough to confirm my belief in this matter. Others however feel differently and spend quite a lot of time and effort talking to the pets and some of them even pretend that the dog is talking back.
Now leveraging the power of AI there is a new app created that analyses & interprets the facial expressions of your pet. Folks over at the University of Melbourne decided to build an Convolutional Neural Networks based application called Happy Pets that you can download from the Android or Apple app stores to try on your pet. They claim to be able to identify the emotion the pet is feeling when the photo was taken.
While the science behind it is cool and a lot of pet owners who tried out the application over at Hacker News seem to like it, I feel its a bit frivolous and silly. Plus its hard enough for us to classify emotions in Humans reliably using AI so I would take the claims with a pinch of salt. The researchers themselves have also not given any numbers around the accuracy percentage of the model.
When I first saw the post about the app it reminded me of another article I had read a few days ago which postulated that ‘Too many AI researchers think real-world problems are not relevant’. At first I thought that this was an author trolling the AI developers but after reading the article I kind of agree with him. AI has a massive potential to advance our understanding of health, agriculture, scientific discovery, and more. However looking at the feedback AI papers have been getting it appears that AI researchers are allergic to practical applications (or in some cases useful applications). For example, below is a review received on a paper submitted to the NeurIPS (Neural Information Processing Systems) conference:
“The authors present a solution for an original and highly motivating problem, but it is an application and the significance seems limited for the machine-learning community.”
If I read this correctly then basically they are saying that this AI paper is for a particular application so its not interesting enough for the ML community. There is a similar bias in the theoretical physics/mathematics world where academics who talk about implementing the concepts/theories are looked down upon by the ‘purists’. I personally believe that while the theoretical sciences are all well & good and we do need people working on them to expand our understanding, at the end of the day if we are not applying these learnings/theorems practically they are of no use. There will be cases where we don’t have the know-how to implement or apply the learnings but we should not let that stand in the way of practical applications for things we can implement/use.
To quote a classic paper titled “Machine Learning that Matters” (pdf), by NASA computer scientist Kiri Wagstaff: “Much of current machine learning research has lost its connection to problems of import to the larger world of science and society.” The same year that Wagstaff published her paper, a convolutional neural network called AlexNet won a high-profile competition for image recognition centered on the popular ImageNet data set, leading to an explosion of interest in deep learning. Unfortunately, the disconnect she described appears to have grown even worse since then.
What do you think? Do you agree/disagree?
Source: HackerNews
– Suramya
