ead Of Artific I Elli En Eand Machin Learning Dept

Everyone always say that we need to look into the past in order to plan for a good future.

Well, I think this is not always true, simply because the issues we face today were not present in the past (sorry for the clash of words) in a similar form; but one thing is true: looking into the past will at least give you an understanding of the velocity at which events are evolving. This concept is known to people working in agile contexts.

So, looking back to when I started working in AI&ML ecosystem (2017), things were *very* different in terms of both technology and readiness of the business stakeholders. My team had not only to explain every single project rationale but also deep dive into details of every feature used estimating the impact it would have to production systems; nowadays things are much more easy, in a way, thanx to the big impact caused by LLMs like GPT, Bard and similar entities. On one hand, business people are not anymore scaried by AI and are generally very open to adopting those solutions immediately just after viewing a simple MVP of their use case; on the other hand the level of their expectation reaised to the stars: "can ChatGPT do all the work ?".

This intro is probably oversimpliefied but it should give a fair picture of a drastic change in people's mentality and understanding of the AI related technology. In a previous article I tried to elaborate some concepts related to AI replacing humans and, at that time, I underestimated the velocity of technology progress; I couldn't even imagine that just after few years the maturity level of NLP and NLU could generate something like GPT-ish models.

  ​On the one hand, corporate employees are no longer scared of AI and are generally very open to adopting those solutions immediately after viewing a simple MVP of their use case. On the other hand, their expectation level was raised to the stars: "Can ChatGPT do all the work?   

To bring some reality touch, I can elaborate on some tasks that I've seen (in a business context and not just on the reasearch desk) evolving from pure manual to nearly full automated in a time shorter that I could forecast:

- data entry: many back office tasks were performed by humans copying text from website to website or from digital documents to applications or from paper documents to applications. The first two examples were automated simply using RPA; the latter, more complex, is now simply depending on the quality of OCR more that the underneath computer vision model (or NLU or other different ML technology). Document classification is a very similar (maybe easier) problem already solved by such alghorithms. This is the area covered by Back Office teams in which machines are reaching human level performances.

- understanding, extracting key concepts from human discussions (even code) and summarizing them in the most spoken languages

technical and copy writers; maybe we can extend to brokers and client facing people dealing with explaining contracts and product features. Any co-pilot technology has reached a very good level of maturity.

- dealing with end-user with tasks like information collection, subscription process, signature and document verification. This is an area covered by sales assistant, hotline support, post sales customer support and similar. Most of the above tasks can be performed by LLMs at the cost of providing a correct and detailed prompt.

- writing code in a widely used programming language simply providing a detailed description prompt. Again, this is a technology available and integrated in most of the coding UI.

Are the results good enough to be used in a production context ? Well... it depends on how busines critical is the task and which percentage of error someone can sustain before impacting revenued. As a matter of fact, all AIs are making mistakes: if your process is very sensible to errors then you still need human in the loop that can correct the machine (and allow it to learn from mistakes).

Machine need humans to learn and business needs humans to ensure all is moving smoothly but... what will happen in the future?

One of LLM stated that: "In summary, the future impact of AI on employment will be a complex interplay of job displacement, job transformation, skill evolution, and the emergence of new opportunities. How society manages this transition, through education, policy, and innovation, will play a crucial role in determining whether AI leads to a more prosperous and equitable job market."

On the same side, a human forecast states that by 2030 we'll have a modest socioeconomic impact; by 2035 those impact will be significant and will not stop. The evolution will not be limited to the Back Office tasks I described before but will extend even to more physical jobs involving movement, maintenance, 3D vision, medical and diagnostic. At a latter stage, machines will be able to fully himitate humans in physical movement and also in creativity opening issues like intellectual property and legislation.

We're getting to the point - humans must be prepared to evolve their roles leveraging on the machine capabilities instead of fearing and rejecting them. Progress (and business) cannot be stopped.