Why is Artificial Intelligence like a big Skyscraper

Maurizio Sanarico is Chief Data Scientist of SDG Group, he is a great data expert, an experimenter of new technologies and new design models.

Looking at the 2023 AI Week (from 17 to 21 April) we turned to him to understand the dimension and quality of the big change underway, which began when (it was the end of November last year) the he generative artificial intelligence, that of ChatGpt to understand us, has become within everyone’s reach.

We asked ourselves together about the way in which artificial intelligence changes data driven projects and consequently how the way companies do consulting changes; on a future made of graph analytics, continuous intelligence and quantum AI; on the new paradigms of research and development that are dictated by artificial intelligence, on the skills that are needed and, more importantly, on the sense of human progress.

In doing so, we discovered that the vision of a data scientist is by no means merely numerical, rather it can be very creative.

And then there is the comparison of generative intelligence with a skyscraper, as well as the enhancement of Mediterranean intelligence and the reference to the exposome.

How is the way of doing consulting and doing data driven projects with artificial intelligence changing?

The main change lies in the need to combine a set of heterogeneous skills, in supporting cultural change, in particular in the ability to explain artificial intelligence (explanable AI) to different actors, not only to make it accepted, but also to enhance the contribution that each of them can make.

As a side thought, I would like to demystify Foundation Models in general.
They are fantastic because they allow you to answer many questions, generate working code for limited problems, do many useful things but do not constitute a paradigm shift towards General Artificial Intelligence.

I compare GPT3 (and ChatGPT3), LlaMA, PaLM and derivatives to ingenious and great engineering works, like a large skyscraper.

In order to be built, the Petronas Towers had to wait for the invention of a new type of concrete, after which the construction principles were those of the construction sciences.

In this case we have the transformer (machine learning models specialized in processing and interpreting sequential data - ed.) playing the role of concrete.

From a consulting perspective we have tools to accelerate certain activities and, using well-constructed prompts, obtain quick and useful answers to business questions contemplated in the training data space.

Cloud, analytics, data management, dataops, metadata, are extremely topical for digital transformation, but today they almost seem like the past. At the gates are Graph analytics, continuous intelligence and quantum AI, there is the future. What are these technologies and what can still come?

In reality, it is the extension of existing paradigms to more complex data structures (networks, be they macromolecules, social networks, metabolic pathways, or other).
continuous intelligence is an extension of a principle of MLOps with the limited ability to modify learning rules.
Networks based on Neural Operators (by Anima Anandkumar) and Parametric Machines (by Bergomi and Vertechi) are more interesting advances and should be followed carefully, both for the theoretical innovation component and for the performance.

Quantum computing in general and Quantum AI in particular are currently restricted to a limited set of applications, interesting for example OpenFermion by Google, the first open source quantum computing environment.

The generalist applications are still at the window. From a computational physical point of view, both a quantum system and a classical system obey the Church-Turing law, only that the former could be much faster, allowing to solve problems that are today impossible from a computational point of view.

Do you also change the way you do research and development? How are the new R&D paradigms dictated by artificial intelligence?

In this case, artificial intelligence allows for an acceleration that can be defined as a qualitative leap as well as quantitative.
An example: in causal analysis the control of confounding factors in the classical paradigm is based on relatively rigid methods (parametric models, matching and propensity matching, often combined), while generative artificial intelligence allows the use of specialized GAN (Generative Adversarial Networks) for this reason, in which the relational structures between observables and treatments (the object of causality research) are extremely flexible and even dynamic.

Again: realistic synthetic data to perform training on reconstructable data sets, ability to combine multi-modal and multi-frequency data to derive the health status of monitored patients, taking into account environmental data and their impact on organisms, i.e. the exposome (concept defined in 2005 by the epidemiologist Christopher Wild, to indicate the globality of environmental exposure starting from the origins of life in determining the health conditions of an individual - ed.) are just some of the examples, so as not to have to repeat things that are often found in the media.

The so-called physics-informed neural networks are another important contribution of artificial intelligence to research.

What skills and resources does a company need today to carry out projects with artificial intelligence? How does he build them, or where does he find them?

Like all complex activities, skills and composite resources are required, from the expert who has in-depth knowledge of algorithms, theory and practice, up to a profile capable of applying the algorithms and to study the data, guided by more expert figures, without going into depth but aware of their meaning and scope of validity.
In between are experts covering the intermediate skills spectrum.

The greatness of artificial intelligence lies in its being a metaphor for age-old human progress, with the ability it has to bring together technological, economic, ethical and regulatory issues. Is there one that predominates? What is your thought on this?

Currently, as expected, the algorithmic aspect dominates, embodied in the software. In a more mature future and, I hope, in the making, we will talk about Humanist AI, in the direction of the Human-centered AI of Stanford.
However, I believe that a Mediterranean cultural richness must be recovered in contributing to this direction in an original way.

Will we succeed?

I hope so.

Original article published on Money.it Italy 2023-03-11 16:23:00. Original title: Perché l’intelligenza artificiale è come un grande grattacielo