How Do Companies Use AI to Detect AI?

With the advent of readily accessible large language models like Chat GTP, everyday people can utilize the power of AI to ask questions or queries, or even to prompt the software to perform entire tasks.

When it comes to content creation, this can be a problem for companies who are looking for high-quality wordsmithing. AI content might be factually inaccurate, unoriginal, or even plagiarised. Google search engines allegedly don’t penalize AI content specifically, but they will penalize mistakes and bad quality that derive from the improper use of AI.Content creation is not the only area in which AI can be used by the masses to get ahead. In recruitment and education, students and job applicants have been using technology to write CVs and even academic essays, throwing up a whole number of problems when it comes to grading and selection. Online gaming, particularly where it concerns real money, is also an area of concern. AI programs, known as bots, can be created by those with the technical knowledge and then used to attempt to cheat. In all of these cases, companies have the option of working with AI to combat these negative effects. By using programs that spot AI text or bot gameplay, companies are attempting to get ahead of the curve. 

AI detection in online gaming

The first category we’re going to explore is online gaming, particularly online casino gaming, and poker. This is an industry worth hundreds of billions of dollars globally. With so much money involved, it’s imperative that companies detect frauds and scams and quickly shut down AI accounts.This isn’t so much of a problem for casino games that are based purely on chance, as an AI bot cannot overcome the house edge that is built into the games. In other words, it’s virtually impossible to cheat at online roulette or slots using an AI bot. The use of AI bots is much more concerning in games of part chance and part skill, such as poker. In this case, it is possible to create an AI program that can play without human participation and has at least a chance of competing profitably in the pool.Fortunately, it’s very difficult to create an AI bot with a big edge in poker. Most of the time, scammers have to settle for very marginal gains. To tackle the issue, major online poker sites, have security teams dedicated to finding these bots and banning the accounts. For instance, Ignition Poker uses top-notch encryption methods, which protect player data and transactions to the highest level and also incorporate card-shuffling algorithms that craft entirely unpredictable card combinations. Overall, the platform provides a reasonably safe environment for players, as attempts to cheat are usually quickly exposed. Online gaming sites may employ the use of AI detection systems as part of their security. These systems will scan the site much more efficiently than humans to find unusual account activity, anomalous behaviors and strange, overly consistent betting patterns. Once flagged, these can then be monitored by humans who will make the final decision on whether to take further action. 

In recruitment and education

Now large language models like Chat GTP, have changed the internet and provided a near-instant solution for the creation of any text-based documents. This includes CVs, job applications, and academic essays, all of which are very problematic for the company or educational institute. Applicant tracking systems (ATS) are used by recruitment companies in the first stage of screening to scan hundreds of potential applicants. Most of these systems, for now at least, focus on analyzing the content of the application and matching it to the job description. They do not tend to include AI detection tools.This means that, technically speaking, an AI application will probably get through this initial screening, if it has been hand checked for quality. After this, however, the application will be manually assessed and the applicants themselves will still have to undergo further interviews and in-person tasks. Although an AI-written application may pass screening, the individual will still have to have the necessary skills and experience to get the job. This may change in the future if companies decide to integrate AI detection into their ATS screening stage. Or it could go the other way. Forward-thinking companies, especially those in the tech industry, may see the use of language models as a sign of innovation and experience with AI.The issue is much more problematic in education, as students can potentially use the tool to cheat their grades. Universities and colleges have been combatting this using AI detectors. This software compares the essay to a database of text and analyses the language used and the way it is used. In some cases, universities have reported a very low false positive score of less than one percent and claim that their AI detectors are remarkably accurate. However, Open AI themselves admitted in 2023 that detection tools are not accurate enough and even discontinued their detector.Again, a mixture of AI tools and human intervention is necessary to tackle the issue. 

Content writing

Many companies outsource their content writing, both for real-world advertising and everything to do with their website and online campaigns. So-called writers can use Chat GTP to their advantage, drastically reducing the amount of time it takes to create an article or blog post. If used with permission and for a purpose such as editing or refining text, AI can improve productivity. The problem is that AI writing can be generic, inaccurate, and potentially illegal in terms of copyright. It is not as original as human-written content and devalues the service for the companies who have paid for it.Some of the telltale signs of AI writing include the use of overly sophisticated jargon, the lack of personalization, and repetitions of particular words and sentence structures – although these can also be the signs of an inexperienced writer!In any case, AI content detectors can again be used. They will scan for features such as perplexity, meaning how predictable the text is from an AI perspective, and for variation in word use, sentence length, and structure.Given that these detectors are not always accurate, again human intervention is key. Once the software detects a potential AI use, the content manager can review the work and monitor the content over time to decide whether further action is needed.In effect, AI detection tools are in a constant arms race against AI large language models. In order for companies to keep on top of AI use, detectors have to keep up with the compounding quality of AI-generated text.