(Finance) – On theTO THE“when the first version of Chat GPT was launched two years ago we started a race to experiment. However, all this energy risked being dispersed without a process and without clear objectives, so we worked a lot on the organisation, giving clear objectives and clear, distinct responsibilities between those who have to invent the use cases and those who have to implement the prototypes“. He told Finance Andrea Bei, Head of Digital Services ReActiveon the occasion of the ninth edition of the Payments Fair.
“And then – he continued – we defined a process to accompany from the idea to the experimentation up to industrialization the use cases and we worked on this process which is built on three phases: a reconnaissance of the most promising use cases from a business point of view, a selection of a short list of those with a better business case and with a technical feasibility, to then arrive at experimentation on real data or synthetic data, and finally industrialization for those most promising prototypes worthy of an investment.
“A very promising use case, which in fact passed all these stages and ended up being financed by ReActive, is that ofadopting artificial intelligence to modernize banks’ outdated applications – said the EIB – Most banking operations, as we know, are still based on applications which, although in operation for many years, are obsolete, poorly documented and need to be managed; and this has meant that a significant technical debt has accumulated which the banks are now paying for in terms of poor understanding of the as is and therefore difficulty in dealing with change, difficulty in adopting the cloud and low time to market”.
“Our tool acts precisely on this, on the ability to recover the business logic of old applications of millions of lines of code, accelerating the work of a human analyst – said the Head of Digital Services ReActive – So our point of view is that AI-based solutions work, and in this case they work very well, when it is always there a human contribution that is synergistic with artificial intelligence. So think, rather than in terms of absolute automation, in terms of a synergistic contribution between human beings and artificial intelligence.”
On the lessons learned, he said: “First of all, work on the organizational model to distinguish and assign very accurately the responsibilities between those who have to conceive the use cases, the business, and those who have to implement them, research and development. The second the point is training: a lot of training, but not only for the more technological skills, but above all for business people, with training on the methods and criteria for using artificial intelligence to then ensure that in their specific business area they are able to identify those processes that can be accelerated through artificial intelligence”.
“Another very important lesson that we have learned in the field is to always consider solutions with a complementary contribution between human beings and artificial intelligence, thinking that it is It is necessary to pre-treat some problems to ensure that the dimension and context can then be handled by artificial intelligencetherefore a pre-processing of the problem to reduce it in terms of complexity”.
“Finally, a very important element is build an architecture with stable interfaces around solutions based on artificial intelligencebecause it is essential to be able to change – in the case of generative AI – the linguistic model. We know that it is constantly evolving, there are at least thirty of the most relevant at the moment, and therefore it is essential not to create lock-ins with these technologies. Also because by doing so it is possible in the future to also think about model brokers, to ensure that depending on the specific problem it is possible to choose the cheapest or best linguistic model in terms of accuracy”.