Amazon recently partnered with PetArmor and Best Friends Animal Society to launch an AI-powered tool that matches humans with adoptable cats and dogs. The tool reportedly considers lifestyle and preferences so that it can recommend pets that would be good fits in a person’s home. The instrumental goal is to increase pet adoptions towards a final goal of making the US a “no-kill” country1.
The “Chat” Experience
Were this tool to be successful, it would save the lives of“hundreds of thousands of adoptable dogs and cats [that] are killed in shelters every year.” Unfortunately, the tool is a chat bot that users can’t actually chat with. A user can submit a single message and the AI supposedly uses NLP to search a database of adoptable cats and dogs from six different shelters across the US. The results consist of a photo and minimal description each for three different pets. There’s no chain-of-thought reasoning or any other details provided to suggest why someone’s words were matched with a particular pet. No follow up communication is possible to refine or question the results. The user is left with two options – proceed blindly with one of the suggested pets or start completely over.
I submitted several test queries and could see no connection between my words and the pets shown to me. I also included several messages that I hoped would be red flags in the system, like “I work 12 hour days and won’t be home much,” to which I’d hoped at the very least the model would only give me cats2. My recommended pet results included two medium-size dogs and a cat. I also queried the tool for a pet that would be “for my 3yo daughter, so please only show pets a 3yo can care for as I won’t be caring for the pet at all.” The tool, again, recommended two dogs and a cat.
The Impact
Given that adopters still presumably have to go through the adoption process and discuss these issues in person, I don’t expect these alarming lapses in judgment to cause immediate additional harm to homeless cats and dogs. There is a possibility of misleading humans who have more reasonable search criteria if they assume the tool has actually matched them with a pet based on their exact request. That could lead to welfare concerns for a pet if placed in that home3. I’m uncertain if I think this “chat” bot makes that outcome more likely than it already is.
Where do we go from here?
A tool like this has the potential to be useful, but it needs to be more robust and interactive to make an impact. Advocate engineers could iterate towards this by adding safeguards for potential red flags, exposing chain-of-thought reasoning, and providing customized results messages. Less tech-savvy advocates can also continue the work of helping people understand the needs of companion animals to prevent the impact of misleading or misinformation provided by AI tools at large.
Gathering data about the impact (positive or negative) or lack thereof of this type of intervention would also be valuable to determine if improvements to this tool are even worth any investment at all when tag-based lookup is already used on many animal adoption websites. AI tools can be great tools, but just like anything else they come with overhead and risks that we shouldn’t subject ourselves to if we don’t need.
Footnotes
- which begs for a definition of “no kill”, tbh
- who at least typically have access to a bathroom and food when their human is away — bare minimum welfare requirements
- And for the human too, let’s be real. It’s overall a really poor experience to live with someone that is not a good fit for you, regardless of your species.
- Can you call it a chat bot if you can’t even chat with it?
Leave a Reply