What is transparent artificial intelligence, and why is it important? Interview on the subject with Giovanni Mangani, West Europe Lead AI and Marketing Solutions at Pegasystems.
L’Artificial intelligence is increasingly used by businesses from all sectors. This technology offers new possibilities, and many experts see it as a new industrial revolution as important as the steam engine or the appearance of the first computers.
However, this revolution cannot take place properly if the use of AI is not subject to a certain ethic.. To find out more about this topic, we met Giovanni Mangani, West Europe Lead AI and Marketing Solutions at Pegasystems. Founded in 1983 in Cambridge, Massachusetts, Pegasystems is a company that develops software for customer relationship management, including AI.
What are the dangers of an opaque artificial intelligence?
According to Giovanni Mangani, an “opaque” artificial intelligence presents multiple risks. First of all, it should be remembered that to many RNs are “biased” by prejudice inculcated (voluntarily or involuntarily) by humans. A well-known example is Microsoft’s Tay chatbot, which quickly became racist after spending a few hours on Twitter in 2016. For good reason, this AI had been “fed” by the racist Tweets of social network users.
However, when AI is opaque, it is impossible to detect these prejudices and their origins. On the contrary, Transparent AI helps to better identify the problem and therefore to remedy it promptly.
Besides.., Artificial intelligence is a tool as powerful as it is dangerous.. A few years ago, in a study entitled “Private traits and attributes are predictable from digital records of human behavior,” an AI was used to profile Facebook users based on their “likes. Depending on the content on which users had put the famous “blue thumb”, artificial intelligence was able to predict certain very intimate details of their personality.
For example, she has been able to determine whether they use drugs, whether they are in a relationship, what their sexual preferences are, their level of intelligence or their degree of emotional stability. At the end of the experiment, the researchers had drawn some worrying conclusions: with a dozen likes, an AI knows you better than your employer. With 150 likes, she knows you better than your parents. With 300 likes, the AI knows you better than your own spouse.t…
So we understand that he is it is essential that the use of AI be firmly regulatedand subject to rules of transparency and ethics. Otherwise, the door is open to all possible and imaginable excesses.
That is why to many companies make it a priority that their AI systems be transparent and explainable. According to a study conducted by PWC, this is the case for 61% of executives. This is one of the major trends in AI for 2019. Similarly, 55% want to create ethical, legal and understandable AI systems. Indeed, in Europe, the use of AI is also subject to the GDMP.
How does the RGPD impact the use of AI in Europe?
For artificial intelligence algorithms, the data could be compared to fuel. Machine Learning algorithms need data to train and become more efficient. Without data, there is no artificial intelligence..
In fact, the use of AI in Europe is strongly subject to the General Data Protection Regulations. In order to be legal and compliant with the GDPP, the use of the data and therefore of the AI must be transparent.
In the area of customer relations, that of Pegasystems, the limitation on the use of customer data induced by the DPMR also hinders the use of AI. The customer consent is required so that their data can be exploited.
This is why more and more companies are looking to convince their customers to sell their data directly to them. Although it is an investment, the benefits of using the data for AI make it very profitable.
Pegasystems’ techniques for transparent AI
For an artificial intelligence to be transparent, it must be must imperatively be possible to explain the path that led him to make a decision. It also requires be able to understand what data have been exploited and in what way.
In order to ensure the transparency of its artificial intelligence models, Pegasystems uses several strategies. First of all, the company sets rules to define the scope of action of an AI and limit its impact precisely.
In addition, Pega has developed a scoring system allowing customers to know the level of transparency the different artificial intelligence models proposed for customer relationship management. A system with a high score will be transparent, while a low score indicates some opacity.
For example, decision trees are models that can be explained and understood by humans. On the contrary, Deep Learning models are generally comparable to complex “black boxes” whose “reasoning” is impossible to explain. For this reason Pegasystems advises customers not to use Deep Learning for decision making, even if they are given the opportunity to do so.
Towards a more transparent future for artificial intelligence?
According to Giovanni Mangani, to meet these different ethical and legal requirements, artificial intelligence will evolve and adapt in the years to come. Some already existing AI models will be optimized to become more transparent.
This has already begun, as Pegasystems is, for example, using “adaptive” models created from existing models that have been modified to comply with the DGR. Likewise, new, more transparent models will most certainly be created. in the future.