Concerns about the invasive potential of advanced AI technology are becoming increasingly justified. Veritone has introduced an AI model named Track AI, capable of identifying and monitoring individuals in various types of video footage.
You might assume that such technology would be illegal, especially since facial recognition for surveillance purposes is prohibited in numerous regions across the United States. While that assumption is correct, Track AI operates without relying on facial recognition. Instead, it distinguishes individuals based on physical characteristics such as body size, gender, hair colour, clothing, and even gait.
The ability to track individuals by their physical attributes while bypassing facial recognition may make Track AI particularly appealing to law enforcement agencies. Reports suggest that some agencies are already employing early versions of this surveillance technology in their investigations. However, like facial recognition used for mass surveillance, Track AI raises significant privacy concerns.
Nonetheless, Track AI could serve as a valuable asset for law enforcement, provided it is used within legal boundaries. The primary concern lies with human misuse of this powerful AI for personal motives. Additionally, there is the looming threat of authoritarian regimes potentially developing similar technologies or adopting Track AI for near real-time monitoring of citizens.
According to a report from MIT Technology Review, Veritone’s AI technology has already attracted 400 clients, including various public sector organisations in the US, such as state and local police departments, universities, the Department of Justice, the Department of Homeland Security, and the Department of Defense.
Veritone has not specified which divisions within these agencies are utilising Track AI or the specific purposes for which it is being employed. The departments did not provide comments for the MIT Technology Review article. However, Veritone did showcase the technology, and the report features a brief video demonstrating Track AI in action.
Imagine attempting to locate a suspect within a collection of surveillance footage, with facial recognition off the table due to local restrictions. With Track AI, investigators can select various attributes of the suspect. In the demonstration shared by MIT, the investigator specifies an accessory (a backpack) and describes clothing details (grey colour and long sleeves).
This information is sufficient for the AI to generate a list of potential matches. The investigator can then review the footage to confirm the correct individuals. While Track AI does not utilise facial recognition, the human reviewer may still recognise the suspect’s face or other distinguishing features.
Once a match is established, the AI compiles a timeline of the suspect’s movements. In this way, Track AI aggregates a substantial amount of video data that could aid in advancing an investigation. Rather than having humans spend hours scrutinising surveillance footage, the AI handles the preliminary work, allowing the human to simply verify the findings.
Despite privacy concerns, this technology could prove beneficial in everyday situations. For instance, it could assist in resolving cases of iPhone theft, where criminals deceive owners into entering passwords, subsequently stealing their data and locking them out of iCloud. Tools like Track AI could help solve such crimes, along with many others captured on video.
Currently, Track AI is compatible with any recorded video, whether from surveillance cameras or other sources. However, Veritone’s CEO, Ryan Steelberg, informed MIT that the company anticipates the technology will be able to work with live video surveillance within approximately a year.
Nevertheless, the privacy risks cannot be overlooked, regardless of the technology’s potential benefits. There is a genuine concern that individuals with access to Track AI could misuse it. One can easily envision how someone with access to surveillance footage might track an individual for personal reasons.
Moreover, there is a broader apprehension that governments could employ Track AI for mass surveillance. The ACLU is closely monitoring the situation, cautioning MIT that this represents “potentially authoritarian technology” emerging from Veritone.
Jay Stanley, a senior policy analyst at the ACLU, remarked to MIT that Track AI provides police and government agencies with powerful new tools. While it may assist in solving crimes, it also increases the risk of overuse or abuse of the technology.
ACLU attorney Nathan Wessler highlighted that Track AI lacks the limitations associated with facial recognition, which requires clear images of faces. In contrast, Track AI can identify individuals based on their gait, enabling law enforcement to utilise video footage that may have previously been deemed ineffective.
The report also indicates that the majority of Veritone’s clients are in the private sector, with only 6% from the public sector. However, this public sector segment is currently the fastest-growing market for the company.
As previously mentioned, the emergence of AI in surveillance is not surprising, given recent advancements from AI firms. Technologies like ChatGPT and Gemini can now identify objects and locations within images, making it inevitable that companies would begin developing AI tools for law enforcement, raising privacy concerns in the process.
Track AI is unlikely to be the last tool to provoke such worries. The positive aspect is that democratic governments will likely implement privacy protections to ensure that the technology is used for criminal apprehension rather than widespread surveillance.
In the meantime, it is crucial to remain informed about the capabilities of tools like Track.
