Artificial intelligence is entering a new era with the emergence of so-called frugal AI technology. Economical in data, machine resources and energy, frugal artificial intelligence is particularly relevant when the entity wishing to deploy it does not have large volumes of qualified data. Applied to new generations of intelligent chatbots, frugal AI allows cutting-edge industries or sectors, even from very technical documents, to adopt high-performance conversational agents.
Artificial intelligence is everywhere … But very often, to deploy Machine Learning or Deep Learning technologies, it is essential to have a considerable volume of data (Big Data). Relatively accessible for generic domains, this data is on the other hand much rarer (and much more expensive) when it comes to very specific or very technical domains. We can thus think of industries such as aeronautics and space, defense, pharmacology, chemicals, banking and insurance… and more generally, of all cutting-edge sectors.
Indeed, the lack of abundant data is a brake on the use of today’s dominant artificial intelligence models. In this context, semantic analysis, theming or the deployment of an intelligent chatbot are then impossible for many sectors with these models. But now, the so-called frugal artificial intelligence technology lifts this brake and will revolutionize, among other things, the world of chatbots. All sectors, even the most specialized, will thus be able to acquire an intelligent conversational agent to facilitate access and the sharing of knowledge.
Frugal IA, how does it work?
Unlike the dominant model which is based on Deep Learning or Machine Learning, Frugal AI does not require large volumes of data to train its model. This is the main strength of this technology. Indeed, most companies that call on an artificial intelligence specialist to, for example, deploy a chatbot, do not have a sufficient volume of data. It is therefore necessary to consider using resources free of rights or third-party paying, not necessarily very representative of their sector of activity.
One possible solution to overcome this problem is to create your own data resources. This solution requires the mobilization of a large number of experts in a given subject, over the long term. DFor example, it is difficult for an aircraft manufacturer to mobilize pilots and mechanics for months to annotate documentary resources. The cost would be exorbitant and the work particularly tedious and far removed from their profession. In addition, the machine resource requirements would then be particularly high, as would the associated energy consumption.
Second solution, adopt models that circumvent the problem of a large volume of available data: frugal AI. This technology is able to learn from very few examples. The major challenge is to succeed in initiating a sufficiently powerful model at first, in order to gain the support of users, and then to learn continuously through human-machine interactions. Data acquisition is therefore no longer done only upstream of the project, but over a long period of time, without mobilizing significant expert human resources.
Frugal AI applied to smart chatbots
Some chatbots – or conversational agents – called intelligent, are able to read and understand documentation, thanks to technologies such as Machine Reading. Where frugal artificial intelligence goes even further is that it allows this type of technology to be efficient on very technical documentary corpus, with very few examples. Thus, it is quite possible for a pharmaceutical laboratory, an aerospace company or a company specializing in finance to deploy a chatbot in order to facilitate access and the sharing of knowledge within its teams.
Frugal AI, a big plus for more virtuous digital technology
Did you know ? In six years, the calculation needs related to Deep learning – currently dominant model of artificial intelligence – have been multiplied by 300,000, just that! Also note that the cost of training this type of technology doubles several times each year! Artificial intelligence is therefore far from the sobriety of human intelligence. When a brain uses around 20 watts to function, an AI based on Deep Learning requires almost half a million watts! Frugal AI clearly appears to be a considerable lever to accelerate the sobriety of technologies based on artificial intelligence.