Truth in the Age of AI (or another accidental insider threat to reality)

Supt. Denis Desnoyers (retired)

Chief Partnerships Officer and Associate Instructor, CInRM CoE

There is a popular saying which claims history is written by the victors. It is generally accepted that “historical truth”, was tempered by those in power, those who survived and those who were literate. Historians’ sources for historical truths are limited by what has been passed down from previous generations.

With the increase of literacy in the 20th and 21st centuries and the advent of the Internet, there has been a paradigm shift in who and what is being passed down. A giant firehose of data has been turned on and is running full blast 24/7, 365 days a year. Our ability to search through the noise of the ever-present data stream has also evolved. Libraries, encyclopedias, hardcopies of newspapers, magazines, trade publications, and registries (to name a few) has given way to the Internet which is now in the process of giving way to artificial intelligence (AI). Now, all you need to do a search in your browser, is to write your query in the form of a question and out pops “an” answer which can save countless hours of reviewing web pages.

For those who read my previous blog, you’ll know that I am a retired Royal Canadian Mounted Police white collar crime investigator. I also spent time in the intelligence world. Something I learned there, was the circular nature of intelligence, or in short, “garbage in, garbage out”.

Law enforcement agencies spend a lot of time imputing information gleaned from their various investigations into databases. “Criminal analysts” working in law enforcement across the country mine those data bases endeavouring to uncover links which will advance ongoing investigations. In contrast, “strategic analysts” use the same data bases to look for information which senior decision makers can use to determine where best to allocate finite resources. They do this, in an effort to be intelligence-led, aka: getting the biggest bang for the buck.

To paraphrase something United States Secretary of Defense Donald Rumsfeld (circa 2002) said, intelligence helps us determine the “known knowns”. Although Donald Rumsfeld comments were associated to the Iraq war, much of what he said applies to criminal intelligence and by extension to AI. When a data set or an investigation becomes too large for one person to fully comprehend, it becomes too convoluted to link relevant minutia. This is an area where AI can have tremendous advantages. Pattern recognition AI tools can assist an analyst find links that may otherwise have been missed; all within a world of known knowns. To my mind, pattern recognition within a data base, is perhaps AI’s most significant advantage. Donald Rumsfeld’s full quote was “there are known knowns. There are things we know we know. We also know there are known unknowns; that is to say, we know there are things we do not know. But there are also unknown unknowns, the ones we don’t know we don’t know.”

Not surprisingly, intelligence databases cannot enlighten us about the unknown unknowns, mostly because nothing about the things we don’t know about were ever put into the database. However, what is in the intelligence databases, was put there by analysts who obtained their data from sworn law enforcement officers. With regards to reliability, if someone within law enforcement is dishonest, they are fired, perhaps face financial discipline, and in some cases, may even be charged criminally. Why the high ethical standards? Because truth matters. Law enforcement is in the business of sending offenders to jail. Whatever evidence is presented at trial, must be 100% true and the source of the evidence must be verifiable. As such, it is vital that the information contained in those databases are accurate.

Consider the concept of a law enforcement intelligence database, with all its checks and balances for truth, accuracy, and source verification, and expand it to envelop the Internet. When you replace the criminal and strategic analysts with AI, instead of having sworn professionals generating reports and administering the database, which now encompasses the world wide web, you have every “Tom, Dick or Sally” who has ever had an opinion or bias about anything (regardless of their political, religious, gender beliefs, or ethics) and has imputed something on the Internet. Not only is the information imputed on the Internet questionable, but the stand-in for the criminal and strategic analysts, the AI agent, is an alien entity of unknown allegiance. Bearing in mind, whatever the AI’s allegiance, it was incubated in a capitalistic or profit driven environment and not an altruistic one.

At this point, we start to appreciate the nature of truth in the age of AI. The expression, “garbage in, garbage out”, takes on a new meaning. It is the same as a database but expanded by an exponential factor of a billion. To borrow a common law phrase, buyer beware, which is a legal principle that places responsibility on the buyer to thoroughly examine a purchase before completing a transaction. If you are going to use AI, you better be aware what it is you are buying as the “truth”. Here are a few anecdotes to drive this issue home. First, should I succeed in having this blog posted on the Canadian Insider Risk Management Centre of Excellence (CoE) web site, I will no doubt be asked to produce a French version for our Francophone friends. To produce my French version, I will likely go to DeepL, my favorite translation site and an early example of an AI language model. Although my content is already set in English, DeepL will assist me in finding the correct spelling and conjugation and allow me to pick the turn of phrase that sounds right to my ears. AI in this instance is a big plus.

As a quick insider risk management side comment, I am not sure where the information I enter into DeepL goes. Fortunately, this blog is not of a sensitive nature. If it were sensitive data, I would think twice about openly sharing on a non-proprietary site. If I was translating sensitive data, I would use a software which would reside on my CPU as opposed to the cloud. The same goes for Google Translate, ChatGPT, Google Gemini, Microsoft Copilot and other cloud-based AI sites.

A more current use of AI was my efforts to improve the appearance of the slide deck I created for my CoE Investigators course. Although I was confident of the content, I felt my slide deck was folksy and lacked visual finesse. By using AI tools, I was able to tweak some of the graphics and realistic images in a way which fit my content. The outcome is sleek and professional looking. AI in this instance was again beneficial. In another side note, I will be upfront about my use of AI when I deliver my courses which is in accordance with the CoE policy on the use of AI.

Finally, there is the true story told to me by a close friend mine, of a person they worked with who had been hired to work on human resources issues. After a few months, the person had to be let go because their work was sub-par. The person’s supervisor and my close friend found themselves having to take over that person’s job function and clean up the mess left behind. Upon delving into the person’s work, they came to realize their work computer contained a number of unauthorized, commonly used, free AI accounts, such as ChatGPT. There were also a number of streaming sites which suggested the person was watching entertainment shows when they should have been working. It appeared the person’s work products had been AI generated with little to no filtering process. This helped explain why the person’s work was so unsatisfactory. Smoke and mirrors is another concept I encountered while working in the white collar crime world. Apparently, smoke and mirrors will only get you so far before someone realizes that what you are trying to sell is a load of nonsense.

The lesson learned, I believe, is to emphasize the need to have a human in the loop who can best judge what is and what is not pertinent. If you are using AI, the person in the loop, should use it only in ways to assist that person, not replace them. Be upfront about your AI usage and how you used it. Personalize your content, be sure of the source and accuracy of the information. In the final “analysis”, AI is no match for a person’s individual training, work and life experiences, creativity, research and common sense.