The future of AI is specialized artificial intelligence (ANI) not general intelligence (AGI) or Superintelligence (ASI); that is the stuff of myth and fantasy. Intelligence was never a general ability. Not even in humans.
So, let’s forget all about any kind of general intelligence and let’s also forget about any substantially human kind of intelligence. It’s not going to happen.
Soon, a new AI winter will come, as this is a highly “seasonal technology.” Probably within the next two to three years, all expectations of any kind of general artificial intelligence will be gone. Companies will have lost patience with the AI hype (and the bill), because it is not bringing what was promised.
But that does not mean AI will disappear, quite the contrary, it’s just going to be clear it will not magically jump out of the screen and start doing our jobs.
It will also be clear that AI is a highly useful technology for many specific applications.
Focus will instead move to applications of machine intelligence that have a documented effect, that create demonstrable value, and where the risk is manageable. This is artificial narrow intelligence (ANI).
Specialised forms of intelligence will appear and win over general LLM type intelligence because they are better performing, lower cost and have lower energy consumption. LLMs can be used to create a demo or POC, but to optimize, a switch to more specialized forms of AI will be done.
If we look at the current use cases using AI that has the highest documented impact we find the following:
Protein folding – The European Bioinformatics Institute houses the AlphaFold Protein Structure Database, which has made structural predictions for over 260 million proteins publicly available, transforming structural biology workflows globally, which accelerates research in rare diseases, antibiotic resistant bacteria and vaccines.
Fraud detection – AI and advanced analytics routinely reduce false positives in fraud detection by up to 50% while protecting billions in transaction volume. Federal Reserve and Bank for International Settlements working papers evaluate machine learning’s efficacy in lowering systemic fraud losses across payment rails.
Coding – Empirical academic and industry studies have measured the effect of using AI on developer productivity. Controlled experiments evaluating tasks completed with versus without AI assistants report completion rates increases around 26% and general productivity gains of 10-30%.
Predictive maintenance – Research published byThe World Economic Forum document that deploying AI driven predictive maintenance and advanced analytics provides multiple benefits like 47% reduction in defects, 24% reduction in conversion costs 32% lower power costs 15% lower material costs.
The pattern from previous AI winters repeat: Inflated expectations on the general applicability of AI capabilities give way to more productive but isolated applications. But AI winters do not make useful AI disappear. The useful applications survive, mature and become less dependent on hype. We should just adjust our expectations and focus.