This is what many of us assumed about AI and no one bothered to clarify, it is certainly memorable for a company, the day it decides to undertake its first Artificial Intelligence (AI) initiative.
However, the vast majority of these companies immediately move on to the one step that seems natural: finding, recruiting and hiring the team of experts who will deliver on all those AI-based transformation promises that have been months, or sometimes years, in discussion.
From this point on, the human talent area is entrusted with a mission that seems insurmountable, due to the large number of uncertainties involved in building an elite data team, also known as Data Analytics, or better Data Science, or perhaps Artificial Intelligence.
Or let's be honest, we're not sure what to call it either. The questions that arise range from the simplest to the most far-reaching:
- Are they engineers, mathematicians, physicists, economists, political scientists, anthropologists, designers or administrators?
- What is the name of the position offered, is it analyst, scientist or engineer, administrator or data architect? There are even at least 5 additional names you could name.
- How much professional experience should be taken into account for a profession that is not certain how long it has existed or when it will exist?
- Where is the best place to look for the most suitable candidates for the profile by defining data?
- What are the quasi-psycho-technical criteria to evaluate in order to select the best candidate to carry out the digital transformation initiative of my business?
Business Experts vs AI Experts
Developing AI vs Productivising AI
One of the greatest ironies I have had the opportunity to witness in multiple AI projects is the close dependency that is created between the data scientist and the data model.
This is because every model that supports a short or long-term initiative involves an initial design stage, followed by development, and finally support. The latter is the most misunderstood stage because 99.9% of models are created to be run more than once, in fact hundreds to thousands of times more.
However, the irony lies in the need for the data scientist to prepare, monitor, tune and finalise the execution. With the mitigating factor that the efficiencies in turnaround times and accuracy, measured in milliseconds and several nines after the point, respectively, are dwarfed by the couple of hours required by the scientist to get the AI model up and running.
This is equivalent to designing a car without an ignition system, where each journey requires the coordinated push of the driver and passengers to start the engine.
This situation stems from the fact that the roles and skills to develop an AI model differ from those profiles and knowledge to make the model a productive application with recurrent use. This gives rise to new concepts such as AutoML or MLOps, which will be the subject of an exclusive post.
Make the most of your data, save time and money in your supply chain using Artificial Intelligence with Datup.
An AI model is a one-time effort.
"AI modelling is a one-time effort". This is not only a common belief, but also a false one.
In an ideal world, AI teams have gained a deep empathy with the strategic business need, enabling them to develop a model that drives performance indicators into the green zones of dashboards.
Next, the model is put into production so that at each new analysis slice, it solves on its own and delivers green results, as it did the first time.
However, in this case, ideal is synonymous with unrealistic. The conditions under which a model performs are too highly variable to assume that the AI system will not require tuning as a continuous activity.
Leaving aside the most obvious reason which is the collection of new data with each run; the adaptation required by the model is mainly due to the ever-changing rules of the business game.
Here again, it is primarily the business experts who are responsible for noticing such adjustments to the AI model, and so the AI team is responsible for feeding the business readings into the model so that the results are consistent with the new context.
From here, it is easy to foresee that this dynamic, instead of tracing a straight line from point A to point B, describes a circular trajectory, a cycle.
If you have any questions or would like to know more about our Artificial Intelligence solutions, please do not hesitate to contact us!
