Lets first look at AI's goal - to do what humans do, more efficiently and faster.
What do humans do:
We analyse data by looking at multiple factors across multiple facets of our world, related to but not limited to, the task at hand. Humans have the ability to call on every aspect of their current surroundings, any outlying influencing factors including gained experience and use the data to make a decision. It isn't always the right decision, but that gained experience helps the next time.
Is AI really capable of doing this?
In the past technology "replaced" humans by applying human derived rules to technology in the form of computer code. If the program says do X under these circumstances, then it does X it if it's matched - There had to be a rule. Rules can get quite complex but there is always, always a case where no one rule fits and unless there is a default (in all other cases do Y), the code fails.
AI is different - it can make up the rules by learning through historical evidence and adapt rules, create rules... so no hard and fast rule exists. AI can makes those rules based on complex (although in human speak simplistic) decision making with input from multiple sources. What AI doesn't do is look outside the task at hand. AI only has the ability to call on a finite number of input sources to make a decision, contrary to humans, where we could change a decision based on a seemingly random unconnected occurrence, like the weather. Lets look at that... we may know of an impending weather condition (Hurricane Harvey or the next one) and know that there will be a dramatic increase in retail sales of grocery items like drinking water and canned goods in a certain region of the country. So apply more "processing power" to POS payments coming from there...or allow low income earners to go into overdraft for payments to grocery stores as they stock up and so creating new financial products... "Hurricane Harvey OD allowance"
That begins opening up the predictive world of AI too... if this happened as a result of an event, the likely hood of that happening again if the event happens is increased...like the next hurricane, so create the product and keep it on hold, release it only as needed with the any minor adjustments.
This shows that AI could potentially ingress into everyday society and banking in a very positive way. But that is just one aspect, if you take that already very fringe AI use case, and expand that taking into account TPP's that now exist, mobile wallets, FB payments etc etc, AI could potentially start to manage (based on the order and payment process) the supply chain. So going back to that use case, order more water for the corporate grocer, give additional credit lines, extend that into payables or receivables finance.... The banking world is forever changed, revolutionised by the harmonious intervention of AI.
But, and there is always a BUT...
AI today is not able to discern the difference between real criteria and false criteria. If we go back to said user case of the weather... what if the hurricane report came from a source that is false, or a source that exaggerated the effect of the hurricane. How does AI look into that and understand, take in feeds, verify and ratify its decision to create the right product or make the right decision.
So right now, AI in payments is a specific user case where it's limited to how efficient it can make a process happen. Route the payment through the correct payment rail, not based on rules, but based on the data in the payments (like we do in our Temenos Payments Hub today but with rules), or apply the customer specific but continually varying criteria a HUMAN customer may ask for without changing the configuration, or applying new rules which have to be tested and released to production... Improve the efficiency of current products, product changes and the related development.
AI for now
AI is a world that is not new, but what is new is the interconnectedness that we now have in the world. It's about making AI closer to human thinking by using inputs from all sources digital and verifying that thinking through self-derived algorithmic calculations and applying. The computing power is only now able to take that volume of information and make "thought derived decisions" without the computer itself costing the price of a small country's GDP. But are we as humans ready to allow the feeble, sometimes unreliable technology make life changing decisions...? That's a question only you in your own mind can answer... Until then, we'll stick to AI in payments being centred on a task to gain efficiencies and reduce costs...