Is artificial intelligence trustworthy?

These days, we hear more and more about AI or artificial intelligence. In all or almost all areas of activity, AI seems to be the magic bullet to meet today’s challenges and to face those that lie ahead. How do we solve labour shortages? AI. How do we better target an audience? AI. How do we reduce the administrative burden? AI. However, according to Jacques Gagnon, the CEO of Imagem, a high-tech firm specialized in the development of health technologies, there is still a fly in the ointment: we overestimate the power of AI.       

To start off, Gagnon, who is an engineer by training, says that any AI system needs to rely on valid and useful data worth exploiting. The exercise itself requires a great deal of rigour to ensure that the collected information is relevant and verifiable and that all parameters that are likely to influence the data to be generated are taken into account.   

Beyond AI, there are people who still today have the responsibility of asking the right questions and of collecting clear, quantifiable and verifiable answers to those questions before processing the information using various algorithms that have been developed for the purpose: this is what we call “deep learning” or DL. In the field of science, just as in the health sector, rigour is law.       

Gagnon adds that a simple device that is improperly calibrated can skew data completely just as an erroneous interpretation given by personnel can invalidate the information that was collected. When we talk about artificial intelligence on social media, it is of no real consequence if Facebook erred when it showed you a pair of pants you should have liked. However, when it comes to health information, to err is not an option. That is why, in order to conduct a rigorous analysis, we must first go out on the field and ensure that the data that is collected is reliable and that the source of the data can be determined.      

According to Gagnon, AI must have roots in reality if we are to give it some kind of merit. In that respect, some aspects of work will always require human intelligence and methods will always need improvements, but, needless to say, AI represents a major leap when it comes to using processes that were completed, up until now, in a subjective way.

At Imagem, we are constantly focusing on eliminating errors from the processes completed in our firm – to the extent of what is possible, and from those put in place in our clients’ establishments. To do so, we follow strict control measures, among other actions. Imagem is subject to the particularly strict MDSAP (Medical Device Single Audit Program) and ISO 13485 standards. All of our applications seamlessly generate comprehensive log files, and the millions of actions that have been recorded allow us to know what there is to know about the operation of our software and to recover what needs to be recovered when needed. This scientific rigour is what carries our guarantee of quality. Gagnon reminds us that we should keep in mind that AI roughly corresponds to data processing and that human intervention is required to validate the accuracy and the veracity of the data.