Scientists believe the introduction of a hormone-like system, such as the one found in the human brain, could give AI the ability to reason and make decisions like people do. Recent research indicates human emotion, to a certain extent, is the byproduct of learning. And that means machines may have to risk depression or worse if they ever want to think or feel.
Zachary Mainen, a neuroscientist at the Champalimaud Centre for the Unknown in Lisbon, speaking at the Canonical Computation in Brains and Machines symposium, discussed the implications of recent experiments to discover the effects serotonin has on decision making.
According to Mainen and his team, serotonin may not be related to ‘mood’ or emotional states such as happiness, but instead is a neuro-modulator designed to update and change learning parameters in the brain.
He even opines that such mechanisms may be necessary for machine learning, despite some potentially disturbing side effects, namely the ones people suffer from. In an interview with Science, he said:
Depression and hallucinations appear to depend on a chemical in the brain called serotonin. It may be that serotonin is just a biological quirk. But if serotonin is helping solve a more general problem for intelligent systems, then machines might implement a similar function, and if serotonin goes wrong in humans, the equivalent in a machine could also go wrong.
The research is still fairly nascent and requires further testing, but experiments conducted on mice indicate serotonin plays a large role in what ‘data’ the brain chooses to keep and how much weight it’s given. In essence, the results of the research show serotonin and dopamine may be intrinsic to the facilitation of a developing intelligence.
In order to determine how serotonin affects decision making, scientists gave mice a choice between two paths, left or right. At the end of one path they placed a reward in the form of water. Once the mice were familiar with the location of the reward the team was able to trigger a serotonin response in the rodents by moving the water and surprising them. Whether the mice found the water wasn’t much of a factor in whether serotonin levels spiked or not, but whether it was surprised was.
When Mainen’s team conducted further experiments, including manually activating serotonin production in an animal running around in a field, they found subjects would slow down and consider the situation almost immediately after a spike. This, according to Mainen, indicates serotonin causes a learning system to place less value on things that just happened (the previous input), instead working to change previous assumptions. This is something that could greatly benefit AI.
The researchers also injected the same mice with a serotonin inhibitor and found that learning became delayed. With the hormone it only took a couple of days for their brains to normalize new data. That time was increased when the mice weren’t able to naturally release serotonin. And that means serotonin (and its effects) may be crucial for human learning.
Whether or not this is useful to machine learning developers depends on how closely they intend to mimic the human brain. Some scientists argue that chemical imbalances in an organic brain are anomalous, but Mainen’s research seems to indicate otherwise. His team hypothesizes that hyper-modulators, similar to serotonin, could be used as ‘shortcuts’ to keep autonomous systems from becoming stuck in outdated models.
Designing robots to deal with a static environment using supervised learning likely won’t prepare machines to deal with the constantly changing real world. But giving them emotions and the capacity to hallucinate things that aren’t real doesn’t seem like a good idea either.
The Next Web’s 2018 conference is almost here, and it’ll be 💥💥. Find out all about our tracks here.