Boston Dynamics has found a way to make its beloved Spot robot even more cursed – this time, by giving it a voice. The loosely dog-like robot is capable of a multitude of feats, from sniff tests and safety enforcement to initiating ‘dance party’ mode, but has lacked the ability to communicate – until now.
While it might feel like a Black Mirror episode has come to life, Boston Dynamic’s engineers have ensured that the experiment is simply an internal exploration into the capabilities of AI Foundation Models, so currently, it seems that they won’t be initiating a robodog uprising any time soon. (AI advancement isn’t the end of the world, take a look at our collection of the best AI productivity tools to make AI work for you).
Using OpenAI’s ChatGPT model, Spot’s team developed a number of prompts to create a series of ‘personalities’ for the robot to assume when giving a guided tour of the complex. From a fancy English butler to “Josh” (yes, literally just Josh), the personas all have a distinct flair, with ChaGPT’s AI capable of emulating a sassy sarcastic tone or even performing Shakespearean-esque sonnets.
I was prepared to be horrified by Boston Dynamics‘ (BD) latest experiment, but honestly, it’s actually quite endearing. Each persona has a distinct, non-robotic voice that appears pleasantly conversational and by incorporating Visual Question Answering (VQA), Spot was able to scan its surroundings and use the context to answer basic questions with surprising sophistication.
It’s safe to say I’m a little enamoured by Spot, but that doesn’t detract from BD’s odd choice to make its mouth move “to look like it was in conversation with the audience.” Instead of appearing to converse, Spot’s mouth awkwardly flaps like a sock puppet but BD insists in a blog post that the “illusion was enhanced by adding silly costumes to the gripper and googly eyes,” so perhaps realism wasn’t the primary concern here.
Notably, Spot was able to answer more abstract questions such as who its “parents” are. It responded by leading the team to the location where old Spot models (Spot V1 and Big Dog) currently reside, stating that the previous versions were its “elders” – because that’s not creepy at all. Despite Spot’s eerie response, it points to the impressive advancement of emergent behaviours, leading Spot to formulate a connection between the concept of “parents” and its old models.
Boston Dynamics has shared that it’s “excited to continue exploring the intersection of artificial intelligence and robotics,” so we could be seeing more ‘sentient’ Spot content soon. “A world in which robots can generally understand what you say and turn that into useful action is probably not that far off,” the company adds, and I for one, am excited to get my paws on my own Spot (“Josh” edition, obviously).