Rule-Based AI Systems, outdated with GenAI?
Level Five Asia
At Level Five Asia, our goal is to provide customers with the right tools to build LEVEL FIVE cultured companies
The Energy Issue
I would like to present my thoughts on how fraud systems are increasingly adopting AI/ML technologies, and why I think rule-based systems aren’t going away anytime soon.
A few posts back we mentioned how there exists as of now an AI energy crisis. Let me elaborate.?
Each ChatGPT query uses about 2.9 Wh of electricity, 10 times more than a typical Google search. While this doesn’t seem like much, millions of people now rely on AI tools for work and productivity, with many making 10 to 20 queries daily.
Training large models like GPT-3 is even more energy-intensive, consuming around 1,300 MWh, or the yearly energy use of 130 homes. For perspective, 1 MWh is the energy 330 homes would use in an hour, or the equivalent of 1.6 million hours of Netflix streaming.
By 2027, AI's annual energy use could match that of the entire Netherlands, posing significant sustainability challenges.
An Increasingly Digital World
GenAI systems are known to consume a lot of energy, and with AI models evolving rapidly, regulations can barely keep up. Issues like deepfake blackmail and GenAI-based attacks highlight the gaps in current oversight.
When I was a sophomore, neural networks were just emerging, and I've seen how they've grown from simple models to powerful GenAI systems. The COVID-19 pandemic accelerated our move toward GenAI, as interconnected systems became essential.
领英推荐
Consider Rule-Based Systems
Some argue that rule-based systems, which use straightforward if-then logic, are outdated. However, they remain relevant, especially in finance. For example, if someone tries to transfer RM500 from an account with only RM800, a rule-based system would instantly block the transaction and alert the user.
Rule-based AI is simpler, less energy-demanding, and easier for domain experts to reprogram than GenAI. Its low energy needs also makes it more affordable.
So until research yields a solution to sustain GenAI’s energy requirements, the industry should not overlook rule-based systems. In practice, this could be employing rule-based systems as a first and last line of defense, spotting more obvious mistakes fraudsters make that would not be efficient to leave it to GenAI systems. GenAI systems can then swoop in to handle large datasets to spot deviations in user or device behaviors.
Links to resources: