Emerging AI Topics: Hallucinations Linked to Serotonin and AI Applications for Crime Prevention

Scientists propose that introducing hormone‑like systems, similar to those in the human brain, could enable artificial intelligence to reason and make decisions like humans. Recent research suggests that human emotions are partly a byproduct of learning, implying that for machines to think or feel they must be able to interpret emotions.

Neuroscientist Zachary Mainen from the Champalimaud Center discussed recent experimental findings on serotonin’s impact on decision‑making at a brain‑machine computing symposium.

According to Mainen and his team, serotonin may not be directly related to "mood" such as happiness, but functions as a neuromodulator that updates and alters the brain’s learning parameters.

Mainen even speculated that, despite potentially unsettling side effects, machine‑learning systems might need such a mechanism. In an interview he said:

"Depression and hallucinations seem to depend on the chemical serotonin in the brain. Serotonin is just a biological quirk. However, if serotonin can help solve emotional issues in intelligent systems, machines might use it to achieve brain‑like functions, and errors in serotonin in humans could have analogous errors in machines."

The research is still in its early stages; mouse experiments indicate that serotonin plays an important role in memory and bodily functions. The results also suggest that dopamine may be an intrinsic factor promoting intellectual development.

To determine how serotonin influences decision‑making, scientists placed mice in a maze with a water reward at the end of a branch. After the mice learned the reward location, researchers moved the water to surprise them, triggering a 5‑HT response. The presence of water itself was not the primary factor driving serotonin level changes.

Whether this insight is useful for machine‑learning developers depends on how they intend to emulate the human brain. Some scientists view chemical imbalances as pathological, but Mainen’s findings appear to challenge that view.

The future use of serotonin in AI remains to be seen.

PCMag interviewed Professor Tambe at USC’s Salvatori Computer Science Center AI Lab to learn more about AI’s role in solving societal problems, as co‑founder of the USC Center for Human‑Centric AI.

Dr. Tambe recounted his early fascination with AI, inspired by the depictions of AI in Star Trek and Star Wars during his childhood in India, which eventually led him to earn a PhD at Carnegie Mellon University.

The original research behind the ARMOR software was funded by the U.S. Department of Homeland Security, the Department of Defense, and the Army Research Office, in collaboration with the University of Southern California.

The ARMOR platform was built in‑house; its algorithms randomize security patrols and provide risk‑assessment scores. These algorithms are not plug‑and‑play but can be deployed in a new organization within about 30 days.

The team now employs machine‑learning techniques, leveraging increasing amounts of historical data. They assume adversaries act strategically, and by modeling known weaknesses—such as in wildlife poaching scenarios—they can develop predictive models to anticipate and mitigate threats.