Short Two Minute Papers video (hosted by Dr. Zsolnai) describing experiments where real goldfish were placed in a Fish Operated Vehicle (FOV) and trained to navigate to targets. The video presents single-fish demonstrations, randomized-start tests, and a six-fish study showing decreased time-to-target and improved success after training; the host highlights the possibility that navigation skills may be more universal across species. The video includes an ethics/regulatory note about university approval and a sponsor message for Lambda GPU Cloud.
Chapters
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today we are going to try to teach a goldfish to drive a car of sorts. Now you may be asking, are we talking about a virtual fish like one of those virtual characters here? No, no. This is a real fish. Yes. Really? I am not kidding. This is an experiment where researchers took a goldfish and put it into this contraption that they call an fov, a Fish operated vehicle. Very good. I love it. This car is built in a way such that it goes in the direction where the fish is swimming. Now let's start the experiment by specifying a target and asking the fish to get there and give them a treat if they do. After a few days of training, we get something like this. Wow, it really went there. But wait a second. We are experienced fellow scholars here, so we immediately say that this could happen by chance. Was this a cherry picked experiment? How do we know that this is real proficiency, not just chance? How do we know if real learning is taking place? Well, we can test that. So let's randomize the starting point and see if it can still get there. The answer is yes. Yes it can. Absolutely amazing. This is just one example from the many experiments that were done in the paper. So learning is happening. So much so that over time our little friend learned so much that when it made a mistake, it could even correct it. Perhaps this means that in a follow up paper, maybe they can learn to even deal with obstacles in the way. Now note that this was just a couple of videos. There are many, many more experiments reported in the paper. At this point we are still not perfectly sure that learning is really taking place here. So let's run a multi fish experiment and assess the results. Let's see. Yes, as we let them train for longer, all six of our participants show remarkable improvement in finding the targets. The average amount of time taken is also decreasing rapidly over time. These two seem to be extremely good drivers. Perhaps they should be doing this for a living. And if we sum up the performance of every fish in the experiment, we see that they were not too proficient in the first sessions. But after the training, wow, that is a huge improvement. Yes, learning indeed seems to be happening here. So much so that yes, as a result, I kid you not, they can kind of navigate in the real world too. Now note that the details of the study were approved by the university and were conducted in accordance with government regulations to make sure that nobody gets hurt or mistreated in the process. So this was a lot of fun. But what is the insight here. The key insight is that maybe navigation capabilities are universal across species. We don't know for sure, but if it is true that is an amazing insight. And who knows a couple papers down the line if the self driving car projects don't come to fruition, maybe we will have fish operated Teslas instead. What a time to be alive.
This episode has been supported by Lambda GPU Cloud. If you're looking for inexpensive cloud GPUs for AI, check out Lambda GPU Cloud. They've recently launched Quadro RTX 6000, RTX 8000 and V100 instances and hold on to your papers because Lambda GPU Cloud can cost less than half of AWS and Azure.
Plus they are the only cloud service.
With 48 gigabyte RTX 8000. Join researchers at organizations like Apple, MIT and Caltech in using Lambda Cloud instances, workstations or servers. Make sure to go to lambdalabs.compapers to sign up for one of their amazing GPU instances today. Our thanks to Lamda for their long standing support and for helping us make better videos for you.
Thanks for watching and for your generous support and I'll see you next time.
