AI for Fraud Detection: Beyond the Hype

Spanish scientists are on record for having built an AI model that they claim would be able to predict exact points where government projects become susceptible to exploitation and corruption. That’s very logical and clearly applicable. But, there is a hype about the AI technology in general that businesses need to evaluate before fully adopting certain forms of its inventions as the only mechanism for fraud detection.

In simple words, it pays to understand the difference between facts and technological dreams when it comes to real life application of these techs.

The gold rush is there, FinTech companies, world forums, as well as think tank experts can’t stop talking about how AI is the “super holy grail” in fraud prediction. And you know what, it’s like boardroom discussions now revolve around one thing – devising AI strategy.

Are We All Jumping Before The Beat?

Source: medium

Everybody is talking about AI revolution – but the question is, are we really there?…Because 90 percent of experts agree that the AI we are holding today is still an infant – with a brain leveling that of a five-years-old, compared to what the predicted fully matured AI (or general AI) would be.

Well, you’ve heard that AI has done these and that — most of that is very true and interesting, but the limits and hurdles still to be overcome are obvious and self-revealing. It’s hardly 6 months since a section of lawyers sought to reprimand an AI model, which they believed was biased when helping a judge to decide a court case – which ended up sentencing the accused, to life imprisonment. Now imagine, depending on a potentially faulty model to man your organization against fraud. How sure can you be that the system won’t be fooled?

How Far are We to the True AI?

Okay, we cannot say that machine intelligence is the newest discovery on earth because ideally, practical implementation of the tech dates back to 1990s. The 21st century seems to own the technology because it is when computer became super-powerful in terms of processing speed, memory capacity and so on. It’s also the time when usable data has multiplied, and distribution systems like Hadoop have become accessible.

However, we’ve not yet reached the operational maturity because now is the time where AI is being used as a switch to spark inventions. There is a clear disconnect partitioning in how we see AI and the actual reality.

Researchers predict that deep learning and machine learning (both constituents of this tech) would take another five years to meet the basic market expectation. The so-called strong AI or general artificial intelligence is also not anticipated soon until the elapse of roughly 10 or so years.

The Fast Growing Segment of the Tech


Other sources predict that certain areas of the technology could move at a faster pace. The latest report by IDC FutureScape explains how and why cognitive computing and machine learning software would experience faster growth and that before early 2019 most businesses will have adapted tools driven by similar software.

The report is ideally supported by the fact that governments have joined the club of ensuring AI matures faster. President Macron of France recently pledged 1.5 billion euros to further his AI-strategy. The Pentagon Officials have also been in discussion about the topic, where one senior official explained how machine intelligence will define future warfare and the need for the US to lead related innovations.

Extra Vigilance for Fraud Detection

Despite the fact that machine learning and cognitive intelligence (both components for making super fraud detection systems,) command a measure of integrity, it would be safer to think of developing a hybrid technique to counter the currently unavoidable threat of data related allegations.

As in, most inconsistencies, like ethical issues and other AI failures seem linked to unbalanced or bad data. And to ensure accuracy in fraud sorting and detection — to ferret out blame-games, we need to come up with systems that can as well master people’s past and present history in relation to fraud attempts.

The hybrid fraud detection concept can also be built around individual’s lifestyle, peer group scrutiny and so on. That way a model can be regarded watertight to some degrees.