This article aims to raise the impact of the use of visual intelligence (a branch of artificial intelligence) regarding possible fraud, as well as reliability failures in its application to claims. After a study with more than 6 systems we have carried out an in-depth examination of current and future techniques in the coming years.
In the first place, we want to point out that this article does not want to be a treatise where it is exposed what AI is, as well as its different branches, but an article that shows the reality of the use of the technology that concerns us, and if it can have or not a real and safe technical application in the insurance sector.
As a first point, we would like to remind you of the abandonment of this type of technology that the big technology companies already carried out at the very beginning of Covid. In February 2020, they recognized that AI technology was unable to recognize people of color through cameras (https://computerhoy.com/noticias/tecnologia/microsoft-amazon-ibm--software-reconocimiento-facial-policia -658955). This data, by itself, indicates the shortcomings that exist today in this technology.
Can AI correctly detect real damage, can it replace touch, experience-based intuition, and the 3D view of a person?
Risk tables
This table has been calculated for ranges of assessment and sectoral average frequency in Spain. Thus, for example, we have the range from 0 to €300 with a frequency of occurrence of 28% and an average cost of damage of €180. After this data, the claims cost has been calculated for a base of one hundred thousand claims, where an estimate of the possible savings that this technology can introduce with respect to the current costs of the human process is made, and then four fraud hypotheses have been considered that You can easily perform the system, always from a point of view of defense of this technology, assuming fraud between 3% and 12% per section.
Some may think that this technology is not easy to fool. However, here we show a video in which it is seen that it is not difficult to do it, and given the world of social networks that we live in, these methods of fraud could quickly go viral.
This table is based on the reliability error of the AI system. In this case, we have taken into account the detection of false positives, that is, the system detects damage that really is not, data that we have obtained in the tests carried out in our study. For this, we have taken the most optimistic situations of the efficacy/reliability of the different systems tested.
Combining the result of the two tables, it is concluded that the use of this technology in the auto claims field is HIGH RISK, since this study concludes losses from the claims section of €0-300, losses of more than €200,000, more than 4%. loss, reaching up to 39% loss.
NOTE:
In the next article we will publish the analysis we are doing regarding its use to verify risks in contracting policies.