Artificial intelligence has been used to identify live streaming of child sexual abuse (CSA) by analysing online financial transactions.
The new research, from the Australian Institute of Criminology (AIC), used a machine learning model to compare CSA live streaming transactions with a comparison sample of non-CSA live streaming transactions.
The AIC paper stated its model was successful at identifying those who live streamed CSA, while making few errors in identifying those who did not.
“As a proof of concept, this model performed exceptionally well, providing evidence that it may be possible to predict CSA live streaming offenders using financial transactions data,” the paper noted.
“It is possible that, in collaboration with law enforcement and regulatory agencies, a model such as this could be developed to detect suspicious transactions for further investigation.”
The research paper said this type of approach could also be used for real-time monitoring of suspicious transactions, assisting in identification, referral and investigation procedures.
“The performance of this proof-of-concept model supports consideration as an approach to assist the detection and triage of cases.”
It stated the findings suggested the tool could be used to identify high-risk individuals and cases, with very few false positives, thus avoiding misidentification of those who were not involved in this type of offending.
The report did note some limitations to the research, stating “while the model demonstrated success in making predictions, the data were current up to 2019”.
“It is possible that the CSA live streaming environment has changed in the intervening time, meaning our model is less useful.”
It also stated while the model very rarely predicted that someone had procured live streaming of CSA when they did not, it only successfully identified 54 percent of offenders.
“For every individual found to be procuring live streaming of CSA, the model misses another such individual. Put another way, the model successfully finds about half the needles in the haystack.”
The report also stated it was uncertain if the use of cryptocurrency would impact the model’s effectiveness.
“While we are unable to make this assessment using the present data, future research should seek to understand the role of cryptocurrency in supporting these offences, and whether the type of currency influences the trends in procuring offences.”
Read the full paper: Machine learning model identifies payments for live streamed child abuse.
