Many people wonder what is actually going on with artificial intelligence (AI). There is a huge hype about it; this notwithstanding, entrepreneurs may feel that the “AI revolution” has not impacted their business yet.
The hype is based on facts, though. We have witnessed enormous progress in the last ten years, particularly in deep learning. This type of machine learning uses neural networks (i.e., software representations of neurons and synapses) to find patterns in huge amounts of data. Results are impressive, especially in computer vision and natural language processing. A few weeks ago, the news reported on the Google engineer who claimed their conversational system LaMDA is sentient, to the scepticism of all AI experts. Yet LaMDA conversations did appear impressive. This is a consequence of (i) the use of ever-bigger neural network models; (ii) of getting them trained on all the texts on the internet; (iii) and on smarter and smarter learning algorithms coupled with unprecedented computing power. In a phrase, a consequence of better engineering.
To the risk of trivialising the question, engineering can be understood as “how to” make things work. Science, instead, is concerned with “why do” things work, that is, on understanding the underlying mechanisms. That the recent explosion of AI is based on how more than why gives us a clue to interpreting the current state of affairs: that we can build artificial “savants” that make bright predictions while actually not understanding much of the underlying phenomenon. At the same time, we do not understand enough why they do what they do. Can this type of AI be useful to us?
Definitely so. This AI is already boosting progress in a great variety of fields, from developing new drugs to predicting weather and even developing complex mathematical theories, letting alone helping authors to write poetry, to create music, and allowing all of us to communicate in languages we do not know. The “how to” has given us a great tool to extend our capabilities to explore, understand, and be creative.
However, it has not given us an AI that autonomously solves problems and explains its way to the solution: it has not given us so-called artificial general intelligence (AGI). Despite the different views that researchers hold regarding the possibility of achieving AGI, it seems inescapable that AGI systems should embed a model of how the world works: knowledge of mechanisms; why-knowledge. A prominent scientist like Judea Pearl, co-author of “Book of Why”, would say knowledge of causes and effects. Why-knowledge is absent in mainstream AI.
The question is that many, if not most, of the pressing problems that small to big companies face to be competitive lie in the “understanding” range of the intelligence spectrum. It can take a good while to reach AGI and even the far simpler goal of an explainable AI. How are companies supposed to profit from AI in the meantime?
Given that humans are still the only guardians of why-knowledge, there appears to be just one way out: relying on clever data scientists coupled with AI power.
Data scientists are often misinterpreted as those technical guys who can run all the possible algorithms on all the available datasets in a data-crunching fashion. On the contrary, data scientists that deliver value to a company are those who deeply reason about a problem; they are scientists before being data scientists. They understand the business needs and processes, creatively link them to problems that AI can address, and eventually solve them — often with a great deal of discussion and thought. And then they convince the management that the solution can work and eventually help software engineers to industrialise it.
Getting together a team of scientists like this is a challenge. One has to recognise talent, be able to attract and retain it. This is complicated by the peculiarities of data scientists, who make them atypical workers, and by the competitiveness of the AI job market. It is further complicated by the rigidity of many companies’ processes and the actual (in)disposition to innovate of their employees. And yet, much more than relying on algorithms, relying on people is the core problem that companies must address to enable AI-based innovation.
I have been fortunate to work in a first-class research institute in AI for the last 25 years, which evenly pursues basic and applied research. The “Dalle Molle“ Institute for Artificial Intelligence was founded in Lugano in 1988 — which relates to Switzerland’s well-known foresight of innovation.
This has put me in a privileged position to see the dynamics of people and companies around AI. Recently I have gained some entrepreneurial perspective too by co-founding an AI company. My current views are as follows:
- The selection of excellent and experienced data scientists is central to the success of AI. Technical qualities are important as much as personal ones and motivation.
- Good AI requires a team. One data scientist alone won’t do much — AI is a very big field.
- Working groups must be actively composed of subject-matter experts. A strong data-science team won’t help much if its counterpart is not truly engaged.
- Data scientists value informal, academic-like environments where they can openly discuss, learn, and vary the problems they work on.
- Staying up-to-date with recent advances in the scientific literature is fundamental. Having tight connections with top-notch AI researchers in academia is a great plus.
I am aware that it sounds quite unrealistic that traditional companies can internally set up a data science team in as demanding a way as above. Indeed, I envisage the need to create private and public “AI-lab on demand” services so that traditional companies could rather flexibly tap into them.
The main obstacle to boosting competitiveness by AI is never technological; it is people. We need clever data scientists as well as open-minded management and experts working together to promote innovation. It is team play.