Trust in Artificial Intelligence Comes From Both Sides
Do you rely on Artificial Intelligence (AI) to do what you intend to do, and do your users and customers trust you to use AI responsibly?
“To ensure AI can be trusted both internally and externally, organisations must demonstrate that ethics and accountability are embedded throughout the entire lifecycle, from design to use,” blogs Anne Laure Thieullent of Capgemini. I can wholeheartedly endorse this. This is a necessary condition. But is this also a sufficient condition to gain confidence in the application of Artificial Intelligence in an organisation or company?
Reliability is also a virtue
“Trustworthiness”, reliability, is one of the values that an artificially intelligent system must honour. A reliable algorithm does what it promises. Not only now, but also in the future. The English word “trustworthiness” is better: the system is worthy of being trusted. Perhaps the translation “confidential” is better. That is how the EU also defines reliability. “Failure to demonstrate that AI is worthy of trust could have subversive consequences and hinder its adoption by citizens and consumers, thereby undermining the realization of AI’s vast economic and social benefits,” the European Commission writes.
We must realize that reliability does not only arise from the fulfilment of ethical values but is also an ethical value in itself. It is thus not only an instrumental value for achieving economic and social benefit but also a guideline for good behaviour.
“Trustworthiness is a moral value considered to be a virtue. […] A person can prove their trustworthiness by fulfilling an assigned responsibility — and as an extension of that, not to let down expectations,” reads a description of trustworthiness.
Trust in Artificial Intelligence and the reliability of Artificial Intelligence go hand in hand. Let me put it this way: we can only trust Artificial Intelligence when the value reliability is guaranteed at all times. That the algorithm does what it promises — in a responsible way — and does not let us down or cheat us.
Reliability is more than meeting a list
In recent years, many ethical principles have been formulated around trust in Artificial Intelligence. Algorithms that meet the requirements of these lists are reliable. Capgemini uses the following principles:
- AI with carefully delimited impact
- Sustainable AI
- Fair AI
- Transparent and explainable AI
- Controllable AI with clear accountability
- Robust and safe AI
- AI respectful of privacy and data protection
These are excellent ethical principles for Artificial Intelligence. And clients, designers and developers must do their utmost to realize these principles. There is no discussion about that.
But are we there? Sometimes the suggestion is made that it is. If I meet the conditions, then I am reliable. But reliability is a value. And ethical values cannot be captured in checklists, as much as we would like to. We need to do more. Reliability must be internalized. We as individuals, on organisations, want to be reliable. Not just because we have to, but because we want to be.
Jan Tullberg, an expert in the field of Business Ethics, elaborates on it. He divides trust into two aspects: being trusting and giving trust. It’s very simple: your artificial intelligent system can be reliable, but if your users or customers don’t see it that way, it’s of no use. Your customers have to give or trust you. Trust comes from both sides.
Making the algorithm comply with all kinds of ethical principles to make it trust-building is therefore not enough. Users and customers must also find you, your organisation or company, reliable. But then the question remains, is having ethical AI applications enough to be seen as confidence-building as an organisation or company?
Trust in AI does not come on its own
I am afraid not. I think more is needed for that. Ethical AI applications can lead customers to gain trust in the company that makes or uses these applications. But customers have more confidence in AI applications created or used by companies that act ethically.
“Consistently ranked as one of the most ethical companies in the world and by using AI as a critical, transformative service for its customers, Capgemini feels an obligation to express its commitment to the ethical application of AI,” writes Zhiwei Jiang from Capgemini.
That is why Capgemini follows two tracks. Capgemini wants to create ethical algorithms and AI applications for its customers. But Capgemini also wants to be an ethical company across the board. The ethics of the AI applications Capgemini creates are also a result of that pursuit. And ethical AI applications are proof that Capgemini wants to be an ethical company. Those two things are intertwined.
“According to data from the Edelman Trust Barometer 2019, up to 75% of consumers outright refuse to buy from unethical companies, while 86% say they are more loyal to ethical companies. Trust has now become a competitive advantage, with companies rushing to adopt AI to use it ethically in ways that inspire trust from customers, partners and the public,” writes Carly Read on CX Today.
So, this is about credibility. Trust also requires credibility. Without credibility, your AI algorithms cannot be trust inspiring. Even if they meet all requirements and principles, if you don’t believe that, you won’t get very far. “Reliability must come from the core of an organisation’s way of working. All links must be reliable, approachable, transparent and relationship-oriented,” writes Marieke Karssen.
Ethical applications are trusted when they come from credible organisations. Ethical AI applications can help you gain credibility or remain credible. Trust in AI comes from both sides.
The Dutch version of this article has been previously published on the Capgemini website.
