Data transparency: Why is reversibility essential for AI development?
The declining confidence rate in the ecosystem
A Capgemini study published last month takes stock of the immense impact that AI could have on retail companies. Major retailers have become aware of the opportunities offered by AI but are not yet well prepared for its deployment throughout their supply chain. The good news is that companies are developing a better understanding of AI issues. Consequently, they are better assessing their own ability to implement AI, as well as the difficulties to do so.
But with awareness comes uncertainty: in 2017, more than eight out of ten distributors trusted their data ecosystem to implement AI. Today, this percentage has fallen to 55%, and it can be assumed that this is not only true for distributors but also for many business sectors. The growing doubt among companies about their ability to implement AI on a large scale must absolutely be removed. In fact, these are projects with a strong strategic scope that represent a considerable financial gain as well as an opportunity for job creation. Indeed, Capgemini’s study estimates that retail companies could save $300 billion with AI if it were deployed on a large scale, and 71% of companies surveyed say that AI has led to job creation.
Reversibility, a potential solution
In order to remedy this situation of uncertainty, there is a key concept: reversibility.
Reversibility is, in IT contracts, the ability for clients to recover their data upon termination of the contract with their supplier, or more generally the ability to resume, at the end of the contract, the use of the data or software as part of a migration to another IT infrastructure. Many contracts today do not contain reversibility clauses, nor any clause stipulating the fate of the client, his information system and data, at the end of the contract. The client is then “captive” to a service provider who is often reluctant to let him leave because he benefits from appropriating as much data as possible, particularly in the field of artificial intelligence. Indeed, by aggregating the data of several clients, the AI provider can develop more efficient neural networks that he can then sell to other clients, thus considerably increasing his turnover.
However, it is essential for companies to be able to retain ownership of their data, whether for strategic reasons of competitive advantage (they do not want other companies to have access to the same services as they do, which, moreover, will be better in part because of their data) or for reasons of freedom and flexibility in building their own IT solution: clients must be able to build their project as they please, with a supplier or on site. Also, as part of an externally developed project, they must have the freedom to be able to take over this project internally and update the AIs, even after the project is completed. Only the right to reversibility can guarantee such protection to clients.
Reversibility and AI
So what is reversibility in the specific context of artificial intelligence?
Through what is called “supervised learning“, companies’ neural networks are trained on large datasets. This form of training is, for the moment, the most effective way to provide companies with solutions for training industrial AIs. Since the strength of AI lies in its ability to continuously consolidate data sets, and to select better neural networks in order to reconcile performance and implementation capacity, it seems essential to us that companies can own their data sets. It is only under these conditions, and in the name of reversibility, that they will be able to develop their AIs over time with the freedom to change methods or subcontractors as they wish.