Frequently Asked Questions

What is the purpose of this app?

When AI first burst on the scene with ChatGPT in late 2022, it quickly became clear that it would have profound implications for companies and society at large. But the technology behind it – Large Language Models – is quite complicated. Business people found it hard to understand, and even harder to keep up with the rapid change. Everyone knew it would be important in their industry/company, but what is AI? And what can I apply it to?

Canine Detective is an experimental AI app built by strategic consultancy Science House using our AI Rapid Prototyping technology. We use it with clients to help them better understand AI, and to get them thinking about what parts of their business it could transform.

The Canine Detective app makes AI less theoretical and more practical/concrete, builds greater understanding of AI, and gives employees a peek at what is possible – all while encouraging questions and creating a space to engage comfortably.

Companies can work with Science House and learn from our AI Prototypes. We can also work with you to develop your own AI Prototypes that meet your specific business needs – getting working applications in the hands of employees that drive greater innovation around the applications of AI.

Is Canine Detective going to be made available to the public?

There are no plans currently to turn this into a fully public app, but we expect to make some features available to everyone in the near future.

What new features will be released?

This app will soon show contest results in ‘Featured’, and geotagged locations of interest to dog owners in the ‘Neighborhood Map.’

For the ‘Neighborhood Map,’ we will be expanding the number of geotagged locations – which was originally focused on Manhattan. In using the map feature, we tried to be as accurate as possible, though you should keep in mind that a hotel that was once dog-friendly may not be now, and a veterinary practice could close without us knowing. Always best to call in advance or check out their website before bringing your dog for a visit.

If you have suggestions for new functionality, email us at info@sciencehouse.com.

Can I participate by voting in the contents?

The ‘Voting’ and ‘Current Top 10’ tabs allow people to vote on dogs, and to see the top ranked dogs, but you need to have a voting login in order to do that. You can request a login via one of our LinkedIn posts (#caninedetective) or by emailing info@sciencehouse.com.

How can I submit a photo of my dog?

You can send a photo of your dog to info@sciencehouse.com and we will try to process it, though time constraints may make response times long.

This app currently requires a code in order to have permission to upload dog images or to have your dog matched with celebrity dogs and dogs from Film/TV.

Any tips for taking photos of my dog that are easier for the AI to process?

We recommend using these guidelines for submitting dog photos:

Close up photos of the head of the dog are better than shots taken at a distance. Make sure there is enough light – the AI will be using color of the coat and spot patterns as one factor in determining breed composition.

Keep backgrounds simple. A photo on a grass lawn or the deck of a house generally works better. A cluttered background, such as a dog on furniture or a floor with toys, tends to result in less accurate analysis. Dogs lying on a blanket or curled up in a dog bed can cause problems if the color of the dog is similar to the color of the blanket/bed.

If you use an iPhone, you can remove the background (prior to sending in a photo) by going to the ‘Photos’ app and selecting the dog picture that you want to remove the background from. Hold and long-press on the dog and the dog will flash white edges. Tap ‘Copy’ and then paste into an email with the background removed.

What do the percentages represent?

The percentages indicate how much of that breed the AI thinks is visually present for your dog. For example, the percentages might show 55.4% Poodle and 25.6% Cocker Spaniel. This means that the AI sees more Poodle features than Cocker Spaniel features.

Remember, the app is based on an image. Technically, it is testing for the ‘phenotype’ of your dog, which means the actual physical characteristics developed and displayed. It may be detecting the color of the coat of the dog, an estimated weight, length of fur, eye color, ears up/down, length of tail, etc.

Why do the percentages not add up to 100%?

We list breed percentages that are 5% or greater on the dog photo, with a minimum of three percentages – so you will sometimes see percentages as small as 1%. So the total can be several percentage points under 100%.

Mixed breed dogs are the majority of dogs, so why doesn’t the AI, for example, identify my dog as a cockapoo?

If you upload a picture of your Cockapoo, keep in mind that a Cockapoo itself is not a 'breed' (though it is a great dog!) but is a combination of two breeds – a Cocker Spaniel and a Poodle. So the AI will likely show a mix of percentages of those two breeds. And depending on the image, the AI may even identify overlaps with other breeds.

My dog is rated 82.7% German Shepherd and 14.6% Labrador Retriever. The German Shepherd makes sense to me, but the Labrador portion seems weird. What does this mean?

There is overlap of the physical characteristics among breeds. The AI may just be picking up on a physical characteristic that is shared by those breeds. It is also possible that the angle of the shot may not show a feature that is key to a German Shepherd.

We once ran into a couple of Shih Tzu dogs on the street with different coat colors. The owner mentioned that one had been sick and was on medication, the result of which was a change in the color of the coat. That color change could have resulted in the AI determining a higher percentage of another dog breed in that dog.

How consistent/repeatable are the results?

The results tend to be less consistent when the distance, angle, background, and lighting of the photo are different. Removing or simplifying backgrounds can also improve consistency.

How do the results compare with genetic testing?

While genetic testing is one way to identify dog breeds, genetic testing may be based on a small sample of genes, which can lead to inconsistent results.

AI analysis is quick and free, and can serve as a supplement to genetic testing.

The dog’s phenotype is based not only on your dog’s DNA but also the environment in which your dog has been raised. So genes can be expressed in different ways based on environmental variables over the course of your dog’s life.

Can I share the analysis of my dog on social media?

Feel free to share your results! The images that we produce conform to image size requirements of most social media platforms. If you include the hashtag #caninedetective on your post we will try to answer questions or jump in with some thoughts!

What are celebrity matches?

We have developed algorithms to compare the breed percentages of the dog photos, allowing us to search through a large dataset of dogs to find the closest matches.

We have analyzed the breed composition numbers of dogs in film, TV, celebrity dogs, historical dogs, etc. We then apply our matching algorithm to your dog and rank the ten closest celebrity matches. For example, your Cairn Terrier percentages may be a close match with Toto from the Wizard of Oz!

Why no cats?

There are around 40 cat breeds, but 200+ dog breeds. That makes the analysis more interesting. There is also more physical variation in dog breeds. Pomeranians weigh between 3 and 7 pounds, while Mastiffs range from 120 to 230 pounds. Dogs breeds can be as short as 6 inches and as tall as 44 inches. Because dogs have been bred for herding and protecting animals, there are also a lot of distinct personalities.

Does Science House have any other AI apps that I can look at?

We announced Word Golf on September 10, 2024. There are several other AI apps on the way, but viewing them may require setting up a demo with our team. To set up a demo, email us at info@sciencehouse.com.