I’ve long been an aficionado of Species Appropriate Interfaces, which are both important experimental tools and deeply interesting philosophically [2].
This winter researchers at the University of the Negev report yet another very interesting experiment: a goldfish navigating on land [1].
What? Huh?
The project created a motorized cart with a species appropriate control system. The “cab” is a fish tank with clear sides so the occupant / driver / terranaut fish can see the area. An overhead camera detects the movements of the pilot, and, basically, moves the cart in the direction the fish wants to swim (or stops if the fish idles).
Cool.
This rig enables a fish to direct the cart in any direction, using natural piscine locomotion and navigating ability. The question is, can a fish actually maneuver this cart?
Now, a fish has no experience on land, and properly has no reason to even try to navigate on land. So the experiment created a version of a task that is relevant and familiar to the fish: a reward for swimming to a specific target. In this case, the target is on dry land outside the cart, and the reward is dropped into the cab. The task is to “swim” the cart toward the target visible through the tank.
So, a species appropriate interface and species appropriate task.
The results are clear. The goldfish can successfully execute this version of the classic target learning task.
“These results demonstrate how a fish was able to transfer its space representation and navigation skills to a wholly different terrestrial environment, thus supporting the hypothesis that the former possess a universal quality that is species-independent. “
([1], p1)
This experiment does seem to demonstrate a successful domain transfer, using natural swimming and underwater navigation to navigate and move on land. The researchers further conclude that this transfer demonstrates that the fish’s navigation abilities are “universal”, i.e., there is a general core of ability that is not tightly tied to the properties of the usual watery environment.
Hmm. I’m not sure I entirely agree that last part.
Here’s the thing. The “domain transfer” is behind the scenes, it’s not clear that the fish can even tell that it isn’t navigating underwater, i.e., as usual. This experiment works by essentially faking the experience of swimming around the room by moving the small tank to simulate moving through a pool. In this way, the apparatus maps natural fish behavior (“swim to target to get reward”) onto a different domain (“drive cart to target to get reward”), but the fish neither “knows” nor learns anything about the alternative domain.
So, in this sense there isn’t actually a “domain transfer” for the fish, all the magic is done by the humans.
(Note that this experiment would seem the same to the fish if the room is filled with water, and if the cart was actually a submarine.)
Let’s grant that the pilot knows that there is something fishy going on, because there is a window distorting his or her view and a lack of normal currents, as well as unusual vibration and sloshing from the mechanical system. But for him, this isn’t necessarily anything other than “a very strange underwater place”.
Philosophically, we can wonder if a fish is capable of understanding the concept of a room, a cart, or moving around the room. Likely not, at least in the ways we do.
What does the hypothesized “universal” capability actually mean? Minimally, the researchers showed that “swim to target for reward” definitely can be mapped to other “move to target for reward”. The fish sees the target through the cab window, groks that it is “over there” in the same way that normal underwater targets are, and groks that “swimming towards target” will have the expected effect.
OK, sure. “Move toward target” is very generic.
But the mapping is done by humans, not the fish. The fish is just doing “swim toward target”, which is the only instance of the generic “move toward target” that the fish can actually do. It’s the cart that translates swimming into movement, which is where the generic capability lies. And that was built by humans.
To me, this shows that the simulation was successful, and the weird, fictional world was close enough that the fish could adapt existing behavior to work as both the fish and the people wanted it to.
So—very cool project. But probably not as deeply meaningful as suggested.
I’d definitely give ‘em an ‘A’. : – )
- Shachar Givon, Matan Samina, Ohad Ben-Shahar, and Ronen Segev, From fish out of water to new insights on navigation mechanisms in animals. Behavioural Brain Research, 419:113711, 2022/02/15/ 2022. https://www.sciencedirect.com/science/article/pii/S0166432821005994
- Robert E. McGrath, Species-appropriate computer mediated interaction, in Proceedings of the 27th international conference extended abstracts on Human factors in computing systems. 2009, ACM: Boston, MA, USA. http://delivery.acm.org/10.1145/1530000/1520357/p2529-mcgrath.pdf
Robert McGrath's Blog 