At its public meeting on February 28, 2022, the Toronto Police Services Board (Board) approved the new Board Policy entitled “Use of Artificial Intelligence Technology.”
This Board Policy establishes governance for the use of new and/or enhanced Artificial Intelligence (AI) technology, and for previously approved AI technology that will be used for novel purposes or circumstances. Additionally, this policy establishes an assessment and accountability framework regarding the acquisition and use of AI technology.
In response to this Board policy, the Service is undergoing the creation of the AI procedure to provide Service members with a framework for the acquisition and use of AI technology. The draft AI procedure was published for feedback on November 24th, 2022 and public comment was crucial in the continued development of the measured approach to risk assessment, costs, and benefits related to the Service’s use of AI technology
As new information, technological advances and legislative requirements arise, the AI procedure is continuing to be developed.
This Board Policy establishes direction for the risk analysis of all AI technology used by the Service and to post the list on the TPS website. On the public meeting of January 11, 2024, the Board published the list of AI inventory.
High Risk Technologies
Application / Vendor | Purpose of Technology / Intended Use | What information is collected by the technology | Authorized Users | Steps towards Mitigating Risks |
---|---|---|---|---|
NeoFace Reveal by NEC |
The system assists investigators by comparing an criminal suspect image captured in relation to a criminal occurrence to a database of lawfully obtained criminal record images (booking photographs). The system uses a fixed algorithm to provide images from the database that may match the captured image for further review. The potential image matches (if any) are examined and reviewed by a Facial Recognition Analyst. Potential matches are then provided to the investigator for additional review. |
None. The system compares the newly acquired image to a database of known images. |
Two (2) Facial Recognition Analysts members within the Forensic Identification Services Unit |
Database used is a highly controlled set of images. Database is populated with lawfully obtained criminal record images. Request to use application is supported by governance and various forms and documentation. Application conducts image matching through the image only and not through any other data to reduce bias. Two (2) authorized and trained users. Investigators must conduct own actions to continue investigation. Use of application and following investigative steps are disclosed for purposes of prosecution. |
Low Risk Technologies
Name and manufacturer | Description of type of technology |
---|---|
Automated Fingerprint Identification System (AFIS) |
The system assists investigators by providing the tools to compare fingerprints located at a crime scene with a database of known criminal offenders. The system uses a fixed algorithm to provide specially trained Fingerprint Examiners with potential candidates for manual comparison. Results are manually examined and compared by a Fingerprint Examiner |
AutoVu by Genetec |
This is an Automated Licence Plate Reader (ALPR) which is mounted on a Parking enforcement vehicle. Uses Optical Character Recognition technology to observe and identify characters on licence plates. The characters are then prepared for comparison to known reported stolen vehicle and plates with associated issues. Results are displayed for operator to assist in determining next actions. |
AI technology currently being testing or piloted
Name and manufacturer | Description of type of technology |
---|---|
Automated Licence Plate Recognition by Axon Pilot Phase Risk Category: Low Risk |
The system is installed on front line police vehicles and conducts automated licence plate recognition and comparison to a known list of licence plates associated to stolen vehicles, missing persons and AMBER alert. The system will notify the user for a possible match – and user has the ability to evaluate the result accuracy. The user is accountable and responsible for any course of action arising from the information received from the system. Licence plate reads that do not match the comparison list are not retained. |
BriefCam by BriefCam Ltd. Testing Phase Risk Category: Low Risk |
Assists investigators in reviewing large video evidence files to identify and flag for specific objects, or recognize specific licence plates through optical character recognition. The application will scan throughout the video in accelerated time and flag items specified at time-marks. The investigator will then go to the flagged time-marks to review the video and to determine next actions. |
In response to this Board policy, the Service drafted a procedure and associated appendix to provide Service members with a framework for the acquisition and use of AI technology. This draft governance aims to use a measured approach to risk assessment, costs, and benefits related to the Service’s use of AI technology.
The Service would like to thank all members of the public that provided comment on this draft governance during the feedback period of November 24, 2022 – December 23, 2022.
After reaching out to those who commented on the governance to obtain the necessary permission, we have shared the feedback below. In some cases we have removed personal identifiers as requested.
AI Draft Procedure Public Response Concordia-McGill
AI Draft Procedure Public Response Newman
AI Draft Procedure Public Response Timms
AI Draft Procedure Public Response Saulnier and Konikoff
AI Draft Procedure Public Response Anonymous 1
AI Draft Procedure Public Response Anonymous 2
Please note that some documents might not render correctly in your browser. For best experience, download/save the document to your desktop first.