> For example, we aren't using millions of mammals to sit and label machine learning models all day. It is not because they can't create those labels, mammals are really good at a lot of tasks, it is because we can't make mammals sit and repeatedly do a task all day even when rewarded. They just lack the reasoning capacity to override their intuition to do it.
We can. There are chimpanzees in neuroscience laboratories that perform repetitive tasks all day, every day, while they have electrodes hooked in their brain. It's actually disturbing that we submit them to this.
> The experiment consisted of training and test sessions. We used ten versions of the pictures described above. Five pairs (1 and 32, 2 and 27, 3 and 23, 5 and 19, 8 and 15 hours) were used in the training sessions. When the chimpanzees touched the start key at the centre of the touchscreen monitor, images from one of the pairs were displayed side by side. The chimpanzees’ task was to touch the image of the fresher cabbage out of the two alternatives. Once the chimpanzees had chosen one of the two images, the stimuli disappeared. A food reward and chime were given for the correct response. A buzzer sounded when the chimpanzee made an error. Following a trial where the chimpanzee made an error, a correction trial, where only the correct image was presented, took place. The correction trials were included in the methodology to keep the chimpanzees motivated.
We could absolutely use chimpanzees to perform labeling of data for ML. But we shouldn't.
They didn't get the monkeys sit and do those for hours, just a short while per day.
> Each session consisted of 30 trials.
> The experiments took place over eight days, with the chimpanzees participating in five test sessions per day in order to cover all of the novel combinations within a day.
The study just affirms my point that it is hard to make monkeys do stuff for extended amounts of time or the experiment would have progressed much faster. The only way we have to motivate them is to give them food directly after completing a small task, and that motivation doesn't really work in the long run. You can't make them do an hour long task and then give them something.
Edit: Humans however have been shown capable to toil away with no reward whatsoever for years just to achieve some arbitrary goal or reward they believe exists. If this was just a question of degree then it shouldn't have been hard to get monkeys to do a days work with no rewards. Humans aren't a thousand times smarter than monkeys.
We can. There are chimpanzees in neuroscience laboratories that perform repetitive tasks all day, every day, while they have electrodes hooked in their brain. It's actually disturbing that we submit them to this.
Here is an example: https://www.nature.com/articles/srep34685
> Procedure
> The experiment consisted of training and test sessions. We used ten versions of the pictures described above. Five pairs (1 and 32, 2 and 27, 3 and 23, 5 and 19, 8 and 15 hours) were used in the training sessions. When the chimpanzees touched the start key at the centre of the touchscreen monitor, images from one of the pairs were displayed side by side. The chimpanzees’ task was to touch the image of the fresher cabbage out of the two alternatives. Once the chimpanzees had chosen one of the two images, the stimuli disappeared. A food reward and chime were given for the correct response. A buzzer sounded when the chimpanzee made an error. Following a trial where the chimpanzee made an error, a correction trial, where only the correct image was presented, took place. The correction trials were included in the methodology to keep the chimpanzees motivated.
We could absolutely use chimpanzees to perform labeling of data for ML. But we shouldn't.