Reality Check: How Real Investigators Really Use (and Don’t Use) AI Today

1. In Brazil today, where digital tools are still developing, what real changes have you already seen in everyday investigative work?

C: Real change? It’s still very timid. AI, when used directly in investigations, at most helps draft a report, and that’s about it.

D: The truth is, there’s no use of AI here today that’s impactful enough to actually change investigative workflow.

C: Things are moving very slowly. It started with support in writing reports, then a bit in data analysis, but nothing really substantial yet. It should evolve gradually.

D: And in the very few agencies where I’ve seen something more modern, it’s mostly video monitoring with alerts for “unusual” behavior, facial recognition, and vehicle tracking. But all of that probably accounts for no more than 0.5% of the real work that goes into an investigation.

2. Which parts of the investigative process still function exactly as they did before technology arrived?

D: Human reasoning. The investigator’s intuition. The ability to connect the case with previous experience.

C: It’s the investigative instinct. Information that comes from informants, conversations, and signals. Nothing is going to replace that.

3. Among the tools that do exist, which ones genuinely help, and which ones create more work than they solve?

D: Data-crossing tools were a lifesaver. They replaced those spreadsheet formulas we used to rely on. And audio transcription tools help too. Now, AI for text editing and formatting? I found it barely useful.

C: Yeah, those just speed up reports, nothing deeper.

4. Can you describe a situation where technology supported the investigation, but human interpretation ultimately made the difference?

[Neither of them could think of an example]

5. What risks do you see in relying on systems that are still immature or fed with incomplete or low-quality data?

D: Baseless accusations. You could send someone to jail based on inconsistent results.

C: For example, AI creating fictitious information, even fake jurisprudence. You always need to verify everything.

6. Many countries adopt AI faster than Brazil. In your view, what slows adoption here—budget, infrastructure, culture, procurement, or something else?

C: There are several factors.

D: Budget is definitely the main one.

C: Yes, but it’s no use buying technology without training people. So the lack of police training to actually use AI is also a big problem. While the Police Academy provides some intelligence and technology courses, it isn’t able to train the required number of officers.

7. In your daily work, what is the biggest barrier preventing AI or automation from being genuinely useful—data fragmentation, lack of integration, training gaps, or trust issues?

D: Data fragmentation, lack of integration, and trust issues, in that order.

C: And like I said, lack of training for officers to use the tools.

8. When you look at current “AI” tools in the market, where do you see the biggest gap between what they promise and what they actually deliver?

D: Operators often struggle to extract what they really want from AI. Prompt engineering is still too complex to reach the same result as a human would.

C: Yes. Sometimes it takes longer to get the same result using AI than doing it “by brute force”.

D: Like with audio transcription tools.

C: They help, but they always require correction and prior knowledge of the audio. Usually, the transcription is confusing, leading to ambiguous interpretations.

9. What kinds of knowledge gained from field work, community interactions, and lived experience cannot be captured by any model today?

D: Body language and the sociological-environmental context around the suspect.

C: It’s the ability to analyze the context as a whole: the psychological motivation of the criminal, the environment where the event occurred. AI, as far as I know, can’t do that. For example, it can tell you a chair is knocked over, but not why that matters.

10. Which human skills become even more important precisely because automation is still limited—intuition, context reading, interviewing, analysis, or something else?

D: Exactly those: intuition, context reading, and the experience needed to “shape” an interview.

C: It’s the lived police experience. Only a police officer knows what life inside a station is like. Nowadays, newer, inexperienced officers either rely too much on technology or think police work is like a game of cops and robbers. Working inside a station teaches you. And once they learn that well, then they can make technology work in favor of the investigation.

11. Given the tools available today, how have technology, data checks, and financial or digital intelligence changed the pace and priorities of your investigations?

D: Before, analyzing banking data or cyber information was a slow, highly specialized process. That meant crimes dependent on this type of analysis were pushed aside because it took too long to get results. Priority went to “common” crimes that were faster to solve. Today, with modern technology, that’s changed. These investigations became more feasible, more “digestible”, and that’s already shifting team priorities.

C: Yes, it became much faster to break down banking data, which is essential in fraud cases, for example. But then we hit the same issue again. We lack people trained to operate these tools. Without that, the technology can’t deliver its full potential.

12. In your view, what should a “modern Brazilian investigator” be trained in, considering our constraints in infrastructure, integration, and resources?

D: Computer networks, hosting infrastructure, banking systems, and communicating with big tech companies. Without this knowledge, you can’t solve most crimes. Today, with everyone connected, there’s always some digital trail to follow.

C: Exactly. The modern investigator needs broad training and knowledge in information technologies. That’s the area where we’re still most lacking.

13. What do you see as the ideal working relationship between human investigators and AI tools—who should do what?

D: AI can and should replace any task involving calculations or large-scale data analysis performed by humans, but always with human validation afterward.

C: AI can assist with several tasks and speed things up, but human verification is always necessary.

14. When is it essential to rely on technology for support, and when is it better to rely strictly on your own judgment and experience?

C: In daily practice, technology is welcome in everything. But in cybercrime cases, for example, it’s absolutely essential.

D: But if there’s a conflict between the system and my own experience, I’ll go with human judgment.

15. Looking ahead, what improvements in technology or AI would truly make your work more efficient—without replacing the human investigator at the center?

D: In an ideal, futuristic scenario, feeding a case database into an AI that suggests steps and solutions, automates reports, and then gets validated by a human. That would be a dream.

C: And if that system were integrated across all police forces at every level of government, even better. But again, none of that works without training officers to use the tools, which, in my opinion, is the biggest challenge.