Last month I wrote a blog on how technology could be used to improve the lives of our companion parrots. I assessed the ability of birds to “ZOOM” with one another as a mostly positive achievement, although I did mention possible negative issues (e.g., overuse of the same behavioral patterns as happens with humans who become “addicted” to online games). Technology, however, has many faces, and this month I want to discuss another form of technology that is mostly being used for humans but that has implications for nonhumans as well, again pointing out the pros and cons.
What is Artificial Intelligence (AI)?
So…Most of you have already heard of the advent of GPT-4: a form of artificial intelligence (AI) that mimics human communication skills. It is “trained” on what are called Large Language Modules (LLMs). What that means is that it is exposed to huge data sets made of all sorts of human communication materials—anything that can be considered public access; it then learns the most common combinations of words that exist for a particular topic so that it can generate long strings of relevant text. An extremely simple example is what happens when you use some email programs and you start to write a phrase like, “I’ll get back to you…” and the machine suggests “shortly” as the next word.
Because GPT-4’s trainers use every imaginable data source, it can not only answer simple questions and complete simple phrases, but can also be asked to write poems, songs (e.g., output akin to Shakespearean sonnets or something that sounds like your favorite pop artist) and essays in a requested style, can take a complicated scientific abstract and ‘translate’ it into terms understandable by someone outside of the specific field, etc. It’s pretty obvious to see both the pros and the cons: Pros are things like the ease of obtaining answers to questions or playing with the system to entertain oneself; cons are things like students using it to write their papers for class instead of figuring out the assignment themselves. (And, no I am NOT using GPT-4 to write this blog entry!)
AI & Animal-Human Communication
One can also see how it could be used to train “bots” to disseminate huge quantities of both true and false information. My colleagues who have been working on these projects even report how GPT-4 has invented fake references to support its not-always-entirely-accurate claims when it was asked to write a scientific report on a given project. (Remember, it is trained on huge data sets, meaning that it is exposed to whatever is public access, and the content may not always be factually correct.) Many of these colleagues are extremely concerned about how quickly systems like GPT-4 are developing and some have asked that a voluntary hold be placed on such development by its designers.
So, what does this have to do with animal-human communication? I recently took part in a fascinating ZOOM meeting, where members of a group called the Earth Species Project described their latest work: How they were beginning to use AI to develop ways to understand nonhuman communication systems by collecting tons of data on what types of actions individual subjects may perform and correlating such data with their various vocal productions—thus pairing actions with meaning.
The researchers record this information by placing very small tags on their animal subjects. The researchers discussed all sorts of extremely interesting ways in which this information could be used—not only to “crack” the code that the nonhuman was using (i.e., to get a handle on the symbolic representations of the animals’ vocalizations) but also possibly to use it as a way of directly communicating with the species in question.
How AI Can Help & How It Can Hurt Animals
Imagine really being able to understand what your parrot is saying with all its squawks and whistles, and to ‘talk’ back in that mode! Could we then directly ask our companion animals what they want or need? Or ask those in the wild how we could help them survive? Could we more effectively study their intelligence? I’d no longer have to spend years training my parrots to communicate in English but could converse with them in their own “language” to understand how they interpreted our world.
However, the downside of this research was also immediately apparent. Imagine a poacher, trying to trap an endangered species: All that person would have to do would be to broadcast information that could lure subjects to a given spot—for example, the vocalizations of a flock member calling out that a great food source was just a few kilometers away—but the poacher would be waiting with nets to entangle the birds. Poachers can already use playbacks of simple vocalizations—for example, the call of a stranger that was invading a mated pair’s territory so that the pair would want to investigate and defend their space—but this new technique would be a lot more complex, and thus potentially be a lot more attractive to a larger number of individuals. You can see my concerns.
When asked about such issues, the organizers were quick to state that they planned to put controls in place to prohibit “bad actors” from accessing their system. And I’m sure they will. However, we already know that dedicated hackers can break through such controls. Again, I’m sure you can appreciate my concerns.
I don’t have any answers and, despite my colleagues’ many suggestions of how controls could be put in place for the comparable system for human AI, none of those who are working on these issues have come up with any fail-proof method of safeguarding the results from agents who would use these systems for nefarious purposes. Sadly, I don’t see how those who are working with animals can come up with anything either.
So…although I do try to keep these blog entries upbeat and even sometimes funny, I also try to keep them informative. Thus, I felt that sharing my concerns on this topic was both worthwhile and necessary. If nothing else, I hope I’ve provided some food for thought…
1 Comment