AI myth or reality

Expert opinion

Jean-Baptiste Guignard, co-founder of Clay, Fellow Researcher at the Ecole des Mines de Paris and Associate Professor at UTC-Sorbonne Universities
Published on:
Updated on:

AI is neither a unified issue nor a discipline. It refers both to the current fantasy of “superseding humans by machines” and to its scientific and technological reality – statistics and categorization.

We also call these “AI” classifiers. Their role is to distribute information chosen as input into categories, also chosen, called “output”. For example, we will try to determine if a shape in an image (a mass of pixels) is a person, a dog, a laptop, a truck, etc. To which we will add a degree of probable identification: this shape is a dog at 88%, a truck at 22%, etc. In short, classifiers classify pre-given data into predetermined classes. Humans are present everywhere and AIs are probability classification operators.

Its implementation, variably complex, in the industry and supply chain segments in particular is recent and massive – 1.3 $TRN of annual value creation would result. When it comes to AI, however, we are never sure whether we are talking about robotics, virtual reality, optimized trajectories, automation, calculation for games (go, chess), etc. The general discourse is ambiguous.

However, AI is not a recent issue: the Markov chains that we often discuss appeared in 1906, the (very poorly named) computer “neural networks” – which have nothing in common with neurons physiological than the vague understanding we had 55 years ago – experienced peaks of scientific interest in the 60s then 80s, and have recently re-emerged through the entrepreneurial world. Marvin Minsky, founder of the MIT AI laboratory, finally declared “AI has been brain dead since the 70s”, a resounding echo of the declaration of Luc Julia in 2017 [father of Siri on iOS/Mac OS] “AI n 'does not exist ".

Faced with both the naive (but enthusiastic and positive) buzz and the rigorous mistrust, clarification is essential and we can easily distinguish engineering sciences from cognitive sciences. In one case, we will attempt to model and make operational an action that would typically require the intervention of a human using computer and/or mechanical tools (what we call “weak AI”). In the other, we will be interested in the nature of human cognition – namely knowledge and memory, in particular.

In the history of scientific ideas, engineering sciences and cognitive sciences have separated rapidly. The first school/era of cognitive sciences misunderstood the “computational” nature (similar to a computer) of cognition and proposed alternative models in the 1960s: the most recent are embodied, distributed, situated, externalist models, which recognize the constitutive role of the body, the environment, the surrounding technical objects and generally reduce the centrality of the brain.

In this context, it is fairer to recognize both the operational power of the tool (mis)named “AI” and the legitimacy of cognitive sciences (not entirely naturalistic) in describing human attitude and thought. In short, let us continue to grasp the power of statistics for industrial optimization, while trying to distance it from ethical debates which often stem from a misunderstanding of its nature: there is no strong AI today. today, and it is not an increase in (computing) power that will change the situation but a variation in nature. In the meantime, let's talk about AI not as Artificial Intelligence (there is no less reflective than a neural network) but as (human) Intelligence Augmented by statistical tools.

Jean-Baptiste Guignard, co-founder of Clay, is a Fellow Researcher at the Ecole des Mines de Paris and Associate Professor at UTC-Sorbonne Universities.


Read all the « Expert Opinion » articles on the SprintProject blog

SPRINT PROJECT NEWSLETTER

Subscribe to the newsletter

Innovation, from reflection to implementation

You claim to have read our page privacy policy. You can unsubscribe at any time by using the unsubscribe links or by contacting us at contact@sprint-project.com

A question?
An opinion ?