AI no-code NLP platforms change the paradigm
In many professions controls are carried out on a random basis.
It is a generally accepted practice that an auditor reviewing agreements or an analyst screening litterature proceeds with random checks simply because it is humanly impossible and much too costly to process each individual document.
At the same time society puts an increasing negative value to errors. A small sentence in a contract that was not reviewed or a key information missed may have very significant consequences.
At Kairntech, we have faced this exhaustivity question two times in a row in two completely different domains:
- The life sciences industry
- The audit
Detect side effects thanks to AI
Today, in the life sciences industry, any pharmaceutical company has to prove among many other things that it has not heard of any side effects on anyone in any country where its drugs are authorized for sale.
To comply, those companies have to monitor some specific databases but also the local newspapers (local literature) specialized in the pharmaceutical domain.
The good practice is controlled by National Medical Agencies in coordination in the case of Europe with supra-national organizations like the European Medicines Agency (EMA).
It is easy to understand that if a side effect is not known immediately, the consequences could be disastrous for many patients.
Professionals must do their best to monitor local literature, social networks… in order to show their good faith but the scope is not clearly defined because in the past, it was always considered impossible to read everything.
Analyse all contracts for due diligence
Today, auditors have two main closely related roles:
- They must confirm that financial statements are sincere which means that they give an accurate picture of the reality of the company that produced them ;
- They can also audit a company for a private customer or other third parties in the case of an acquisition. The global purpose of the auditing is to confirm, deny or negotiate the acquisition price given by the purchaser.
For the auditing activity, we have again professional organizations that have defined good practice which are often monitored by supra-national professional organizations and sometimes government organizations like in France.
It is also common practice to proceed with random checks because of the sheer number of operations you have today even in a medium company, not to mention big companies.
Exhaustivity for controls
What is interesting is that both domains have been hit by scandals which could have been avoided by more effective controls:
- Lack of control for Enron in the USA, Parmalat in Italy, Wirecard in Germany, Credit Lyonnais or Société Générale in France ;
- Health fraud with the Opioid epidemic in the USA, the Servier case in France, the Thalidomide epidemic fifty years ago in Germany.
Progress in NLP AI technology
The good news is that technology, and in particular Natural Language Processing (NLP) technology, is changing the game, and in particular by automating the processing of large amounts of documents (all the way from extraction to the classification and detection of words, phrases or paragraphs).
And it is not just the technology itself, but also how the technology is applied. Traditionally it has always been particularly labor intensive to annotate documents in order to create the dataset on which a trained model produces reliable results.
Recent progress in annotation tooling combines a great user experience for analysts, active learning to speed up annotations, model experimentation to allow for end-to-end and easy to create annotation pipelines, ready for industrialization.
In other words, technology and tooling are now capable of analyzing all relevant documents.
A bright future for AI NLP platforms
It is not clear if recent progress in technology and especially NLP is known by authorities responsable for controls. The main reason being the topic is brand new.
Random checks are still accepted for auditing while an exhaustive scrutiny of local literature or social networks is not required by the FDA or the EMA.
One day end-users will push for better services. Non-exhaustive due diligence may become very expensive indeed.
It is unlikely that citizens will influence agencies like FDA or EMA but politicians should urge them to take into account progress made by technology in order to reduce public health risks.
Sooner or later, auditors, analysts, experts… will dedicate their time on elements requiring particular attention and by doing so the end-result becomes exhaustive, the number of potential errors dropping to almost zero.
We, at Kairntech, believe that exhaustivity will have a profound impact on the industry, well beyond the two use cases mentioned in this article.