6 minute read 24 Sep 2019
Artificial Intellegent

When and how will society start trusting AI?

By

EY Belgium

Multidisciplinary professional services organization

6 minute read 24 Sep 2019

Show resources

The potential of Artificial Intelligence (AI) is being discovered and deployed operationally at supersonic speed. But do people trust the machines?

"Not enough to turn AI into a success story yet,” EY partners Dr. Frank De Jonghe and Dr. Patrice Latinne say. “Trusted and Ethical AI is now key.”

Imitating the human brain

AI is deployed in two ways, Frank De Jonghe explains. “Firstly, to make completely autonomous decisions, like in self-driving cars, and secondly to increase people's cognitive capacity. An ‘enhanced human’ no longer has to decide on its own which patient takes precedence in an emergency ward. That is done with the support of a recommender system.”

Patrice Latinne continues: “AI automates decisions by imitating human thinking with complex algorithms. This not only brings about technological evolution, but also a major societal revolution. It is logical that people are suspicious. If we want AI to be successful, we have to ensure that people understand the computer's decisions before they can trust them.”

Frank De Jonghe: “Is an AI system robust enough to make the right decisions in all situations? To answer that question, you have to understand how the algorithm works and especially when it does not work and has to be deactivated.”

“The result of machine learning is largely determined by the input of historical experiences. If the system encounters things it does not know or interacts with disparate algorithms, there is a threat it will go off the rails, resulting in flash crashes during algorithm trading, for example, or traffic accidents with self-driving cars.

“Creating operational performance and transparency – opening up the black box, if you will – is a crucial job for technicians in order to achieve trusted AI.”

Trust in AI consists of two layers: can we trust the decisions of AI and are they ethical?
Dr. Frank De Jonghe
EY EMEIA Financial Services Quantitative & Analytics Leader

Ethics of AI is a societal issue

The second aspect – are the decisions of AI algorithms ethical? – is a question for us all. Frank De Jonghe: “When a person is driving and simultaneously has to avoid a child and a senior citizen, they react intuitively. With AI, we suddenly have to really think about the ethical question of whether we are to incorporate priorities, and what they are.

“Another example: VDAB [Flemish Service for Employment and Vocational Training] makes job proposals to jobseekers based on algorithms. If a person refuses these offers repeatedly, their unemployment benefits can be affected. But shouldn't we take the relevance of the recommendations into account? What if the algorithm makes recommendations that systematically channel certain groups of the population into certain jobs based on historical stereotypes that are embedded in the code?”

Patrice Latinne: “It is not only up to the data researcher to determine whether the AI system is ethically sound or not when it targets clients or assesses the risks of fraud, for example. Firstly, it is and remains the responsibility of each organization to ask themselves the question, ‘Are we using AI for the right purpose?’.”

“Before that, the question must be answered by those responsible for the business and communicated systematically to the entire development chain. From the initial design of the AI system to the time of its activation in production, and even while it is doing its work, it is crucial to know which data are being used, how the AI model functions and how it will ultimately update itself automatically, the client groups it impacts and whether an ethical problem may arise in respect of consumers and employees, for example. At EY, we think it’s equally important to advise our clients on the technical development of AI applications and these types of issues of trust and ethics.

No matter what, companies are exposed to the risk of reputational damage whether they use AI or not. Whether it's a person or an algorithm who is racially prejudiced when granting credit, the organization is the one who remains responsible. In any case, companies must remain vigilant when it comes to ethical questions. That is necessary, but not simple.

“Thirty academic teams were once asked to examine whether racial basis was a factor in red cards being given in soccer. Surprisingly, the teams arrived at completely opposite results. That is to say, it is particularly difficult to answer these types of questions. But as a company, you have to be able to demonstrate that you are at least identifying and monitoring potential sore spots,” Frank De Jonghe says.

Are we ready for AI?

The possibilities of AI are exploding and there are great benefits for the taking. Nevertheless, tech giants are the ones who are primarily reaping the benefits at this point.

When introducing AI, there are all types of considerations to make the pieces fit together. What issue does AI solve? What does it cost? Does the company have the proper in-house skills, technologies and infrastructure?

Patrice Latinne: “Also importantly, is AI ‘business- and ethically critical’? The algorithm a bank uses to recommend a certain financial product to a client is a bit more complex than the one Amazon uses to recommend a book to a customer. An error with the client’s money is more ethically sensitive than with their literary preference. Customer experience and satisfaction are more readily jeopardized with these types of banking and insurance issues."

Frank De Jonghe adds: “It is important for companies to focus on what they want to do with AI and what they want to achieve with it. I was recently having a sales conversation with my electricity provider and I told him I don’t care about smart meters. When asked why, I said I did not like my supplier knowing when I was there or gone so it can perform load forecasting. The salesman didn't even know that this was his company’s intention... If there isn’t any transparency about AI’s purpose and added value, it is difficult to introduce AI to the broad public.

AI is not a beauty contest to develop the most state-of-the-art algorithm. What you do with that algorithm is much more important.
Dr. Patrice Latinne
EY EMEIA Financial Services Data & Analytics Partner

Gold rush with lackluster results?

It is clear that the potential of AI is huge and that a lot of desktop development experiments are underway. Frank De Jonghe: “The challenge now is how to translate that into high-performing industrial applications while maintaining control. This is often a difficult trade-off. Despite the clear advantages in terms of workload and savings, it is generally a big leap to trust.”

“EY offers added value in this regard. Not only do we have technicians who build that trust, we also speak the language of governance and management. AI is just a tool. It does not change management's responsibilities. They must continue to apply the checks and balances.”

“EY helps businesses to not simply make a blind leap into the unknown – with the risk of accidents – but to understand what they are doing. In addition, we make them aware of some ethical aspects and we help them monitor them."


According to Frank De Jonghe and Patrice Latinne, it is also necessary to broaden the debate on AI. Finland, for example, is training its citizens on AI to properly conduct the discussion.

Patrice Latinne: “In Belgium, platforms like AI4Belgium and Data Science meetups are thinking about the possibilities and limits of AI, about awareness raising and training. In any case, companies will have to be able to properly explain what AI systems do, why they make certain decisions and what benefits consumers get in exchange for releasing their data. ”

EY Belgium newsletter

Stay up to date with our EY Belgium newsletter. 

Subscribe

Summary

A high level of trust is needed to make AI successful, Patrice Latinne and Frank De Jonghe say.

About this article

By

EY Belgium

Multidisciplinary professional services organization