The Official Publication of the Ontario Association of Chiefs of Police   17
surface potentially relevant informa-
tion – identifying patterns, grouping 
similar material, summarizing large 
datasets or highlighting items that 
may warrant closer human attention 
– so investigators can use their exper-
tise to focus on analysis rather than 
exhaustive manual review. 
For example, an investigator quer-
ies the AI to ask if there is evidence 
of narcotics distribution found on a 
cellphone extraction. The AI tool may 
return a series of images or com-
munications it believes to be about 
narcotics distribution. The investigator 
then examines the highlighted images 
and chats, interpreting the data to 
further the investigation. 
Investigators 
still 
review 
the 
underlying source evidence. They read 
the messages. They view the images. 
They assess relevance and context. 
The AI tool understands the context of 
data, but it does not interpret mean-
ing – it points investigators toward 
material that may matter.
Before AI, the investigator would 
have to pore over hundreds of thou-
sands of images and conversations 
to arrive at the same output. In both 
cases, the investigative work remains 
the same. The difference is time, 
efficiency and consistency. In other 
words, AI is being used to shrink 
the haystack, not decide what the 
evidence means.
From a leadership perspective, this 
is not a technology issue – it is an 
operational one.
EFFICIENCY AND BUDGETS
Efficiency in this context is not about 
convenience; it is about public safety 
and stewardship of limited resources. 
When digital evidence reviews take 
weeks or months longer than neces-
sary, the result is predictable: stalled 
investigations, increased strain on 
investigators, growing pressure on 
already limited budgets and delayed 
justice for victims. Digital evidence 
backlogs grow not because investiga-
tors lack skill, but because the volume 
of data outpaces human capacity.
AI-assisted review changes this 
equation by accelerating initial evi-
dence triage, allowing investigators 
to move more quickly from data 
collection to analysis, prioritizing 
cases more effectively and reducing 
unnecessary delays. For police lead-
ers managing budgets and staffing 
pressures, this matters. AI is not a cost 
cutting shortcut – it is a way to align 
investigative capability with modern 
evidentiary reality without proportion-
ally increasing headcount.
PUBLIC TRUST DEPENDS  
ON TRANSPARENCY
Community trust in policing is shaped 
not only by what decisions are made, 
but how they are made. Poorly gov-
erned AI raises legitimate concerns: 
opacity, over reliance on automation 
or technology that appears to replace 
human judgment. However, avoiding 
AI altogether does not eliminate risk 
– it creates new risks, which could 
include missed evidence, investigative 
inconsistency or investigator burnout.
An important leadership respons-
ibility is to provide investigators and 
frontline officers the tools to effect-
ively do their work. In modern times, 
this encompasses AI tools purposely 
built for investigations, as well as 
ensuring they are deployed transpar-
ently and conservatively, with clear 
boundaries and human accountability.
This includes being able to explain 
– at a high level – what AI is doing in 
investigations – and just as import-
antly, what it is not doing. AI does not 
replace investigative judgment. It does 
not operate autonomously. It does not 
remove human responsibility.
Clear messaging around this dis-
tinction is essential for maintaining 
public confidence.
RETHINKING “EXPLAINABILITY”
Concerns about “explainability” fre-
quently dominate AI discussions in 
policing, but they are often framed 
inaccurately. In practice, most officers 
and investigators cannot explain the 
internal mechanics of many technol-
ogies they already rely on in court 
– from breath analysis tools to radar 
systems to complex forensic software. 
What matters to policing – and to 
the justice system as a whole – is not 
that every investigator can explain 
how an algorithm functions internally, 
but that:
•	 investigators verify and validate 
the underlying evidence;
•	 the reliability of the AI tools has 
been tested and evaluated; and
•	 the outputs are contextualized 
and weighed by humans.
The emphasis on testing and docu-
mented performance mirrors guidance 
recently published by international 
policing and justice bodies, including 
the IACP, which consistently identify 
human oversight and empirical valid-
ation – not technical transparency – as 
the foundation of defensible AI use in 
law enforcement. 
The more meaningful leadership 
question is not “Can my investigators 
explain how AI works internally?” 
but rather:
•	 Has this technology been 
rigorously tested?
•	 Do we understand  
its limitations?
•	 Are we using it appropriately  
and conservatively?
TESTING, VALIDATION,  
ERROR RATES 
One of the strongest arguments for 
the responsible use of AI in digital 
forensics is that it can be empirically 
evaluated. AI features can be tested 
against known datasets, producing 
measurable performance metrics and 
documented error rates. This allows 
leaders to contextualize outputs and 
make informed decisions about when 
and how results should be relied upon.
By contrast, there is currently no 
empirical data measuring how accur-
ate human investigators are when 
reviewing massive digital datasets. 
Variability, fatigue and inconsistency 
have always existed in manual review 
– but have largely gone unmeasured.

View this content as a flipbook by clicking here.