The Atos blueprint for responsible AI
Download the magazine overview AI is a broad topic encompassing many different families of algorithms and techniques. However, we currently live in a narrow AI era where AI is used for very specific tasks. Some AI techniques such as machine learning have proven to be more efficient than humans in areas such as computer vision, automated translation, predictions and anomaly detection.
Despite the progress, a number of technical challenges remain. There is a huge need for diverse and qualitative data and — in some cases — large volumes of historical data. Furthermore, we need more efficient computing, algorithms and DevOps scalability (e.g. parallelization for algorithms and MLOps tooling for DevOps).
Bringing the best out of AI
Atos defines responsible AI with four key dimensions: fair and ethical, robust and secure, industrialized and eco-sustainable. Atos enables you to boost your business by keeping these foundational elements of responsible AI in mind.
The proposed way in which Atos can boost your business are:
- Enable AI – design and deliver cost-efficient and secured infrastructure for your AI needs
- Augment with AI – leverage AI to optimize existing business processes and operations
- Grow the business with AI – leverage AI to create new business models
The four horsemen of the apocalyptic AI
A failure to follow the four dimensions of responsible AI can have a serious business impact. Below are some of the problems and consequences of (non)responsible AI:
(Un)fair and (un)ethical — Compliance fines, reputation damage, reduced talent attraction, negative corporate social responsibility impacts
(Non)robust and (un)secure — Reputation risk, user acceptability, revenue impacts, compliance fines
(Non)industrialized — Impact on gross margin, lower revenue, productization and scalability challenges
(Non)sustainable — Ambiguous decarbonization goals, reputation damage, higher cost for customers, reduced talent attraction, lower revenue, compliance fines
Ethical considerations are important in AI solutions. There has been a shift from algorithms written by humans to algorithms that learn their behaviour from data. This implies that humans need to be in the loop to control outputs of the AI algorithms, which can be biased by input data or potentially compromised. Full delegation of human-controlled tasks to AI solutions implies greater responsibility in the way AI solutions are designed and maintained in a secure and explainable way.
Putting customer’s purposes at the heart
Finally, it’s important to position AI ethics within a customer relationship. Ethics is relative to every purpose and company. Our goal is to support our customers in building responsible AI. We do this by first defining Atos ethics, which explain how we want to drive our AI projects from an ethical point of view. In addition, they must be aligned with the Atos’s enterprise purpose. Then, we define the AI ethics specificity driven by a particular project that requires a custom view on the situation (e.g. a self-driving car that needs to adapt to the principles and cultures of a particular country).
We establish processes and provide rules and tools to make sure that AI solutions take ethical concerns into account up-front, from their creation to their retirement. In a nutshell, we advise our clients to include AI concerns in their ethics.
Only by taking this broad array of factors into account and designing solutions that align with our priorities, client priorities and the specific business, cultural and legal parameters can we develop AI solutions that are a true win-win for all parties.
About the author
Tomas Pinjušić
Associate Cybersecurity Consulting Group, Atos
Tomas is an Associate Consultant at Atos. As such, he’s been working closely with senior cybersecurity consultants and assisting them in their initiatives. The main goal of his activities is to help global companies to have a secure ecosystem and go much further in their security aspirations than mere compliance.
He has developed the AI Business & Cybersecurity Maturity Assessment offer which exists to help companies discover how proficient are they with securing their AI models and utilizing cybersecurity solutions with advanced capabilities. In addition, he’s been working on the Partners in the Spotlight webinar initiative and coordinating activites for The Forrester Wave Q3 2021.
Interested in next publications?
Register to our newsletter and receive a notification when there are new articles.