Incandescent lamp on a brick wall background
Khalid Amin

Khalid Amin

Ethical AI

Share this blog if you'd like :

Share on twitter
Share on linkedin

Ethical AI

One of the ethical concerns in AI is the bias that could exist in a deep-learning algorithm. There are some possible causes of bias in AI, such as inappropriate training data selected, and unintentional input of unethical values into systems that could be a known or an unknown existing prejudice. Some common prejudices include race, gender, sexual orientation, and socioeconomic status.  The ideal way to remove bias completely in the model is to provide sufficient and diverse data SELECTION to train the AI algorithm. Unfortunately, it is almost impossible to find a perfect fit as ‘unconscious’ biases exist in the free society. In the process of debasing datasets, many data are extracted out and the product may become less useful. Sometimes, you will not even be able to uncover the ‘unconscious’ bias until you get the AI algorithms in. Thus, researchers and analysts have to slowly reveal them, it is a process ongoing which takes time and effort to produce a fairer result.

The companies can reduce biases in their data and downstream use of datasets – algorithmic results by designing and adopting policies that create boundaries on what goes in the data storage systems and how it is utilized during data mining, and data analytics processes. Here are our backend and frontend recommendations.

Data Management: It all starts with Data Management and onus lies on corporations to ensures that data is collected, stored and managed in balanced and fair manners.  Every organization measures and policies could different, but here some broad guidelines that we recommend:

  • Data Privacy and security: Preserving privacy in a way that the general public will find acceptable. Developing a data-centric security approach that reduces reliance on the security of networks or servers. 
  • Data Design: Design and set up rules, policies, and standards that manage and organize the data collected and stored in the database system within an organization.
  • Training & change management: Develop a strategic plan for data fairness training program; employees are aware of data policies more likely to lower risk by implementing preventive measures.

Ethical Algorithms: With the rapid progress of complex AI algorithms, along with unprecedented economic benefits for the corporations investing in these technologies often leads to unconscious biases in data output. It is crucial for companies to enforce policies to ensure actions taken through these outputs are aligned with their core values. 

Here are some of the steps the organization can take to reduce biases:

  • Variable selections: When developing algorithms, data scientists have to carefully identify the variables associated with the outcomes and take every precaution to not include sensitive measures such as race, gender, ethnicity. Needless to say that these rule doesn’t apply to research scientists working toward the greater good for the communities.
  • Monitoring and feedback loop: Development should monitor and test the models to ensure bias is not creeping into algorithms, especially when new data are emerging each day. Create a feedback loop to the Data Management team to reduce biases from the data. 
  • Data Acquisition: Organizations need to make sure data acquisition is not compromising consumer privacy. Luckily, the government across are starting to take lead in this and setting standards for the companies to follow. 

Please reach out to us at to learn more.

More to explore

Modernize Data Ecosystem

To be able to compete in the digital age, organizations must take advantage of internal and external, and stay relevant to new business requirements such as AI, Analytics, Automation and Digital initiatives. To achieve this goal, the very first step for most companies is to modernize their data ecosystem and make data accessible for business exploration and analysis. The Data Transformation projects are extremely complex end to end undertaking, and stakes are often too high to fail. It is not a coincidence that we are seeing a rise of Chief Analytics and Chief Data officers roles to support CTO, CIO and Business executives.

In our opinion these are a few recommendations that organizations must consider before embarking on this journey:

Vision & Strategy: The organizations need to define a vision to modernize their data management system, along with a comprehensive implementation roadmap that doesn’t disrupt their operational needs and increase adoption. It is also important for businesses to not only evaluate the fit of various technologies within their Data Ecosystem but equally important is to assure it will maximize the return and minimize OPEX while implementing these solutions.

Organization Readiness: Data Transformation project cannot be seen as a technology project. If business leads own the data, which is the case in most advanced companies, and sponsor the technology upgrades. It also their responsibility to assure it has an appropriate coverage (during and post-project) to support technology team and actively participate in data modeling and governance activities.

Business Use Cases: The ideal implementation would include inventory of Data-driven use cases that looks beyond what companies currently have in their arsenal. Focus should not only be on improving current reporting performance and quality, but rather what next assets they can develop to enhance their decision making process. For an example: if organization fiscal priority is to reduce warehousing ops cost, that use case to drive optimize data-driven warehouse operations. It is not how quickly you can implement, but how quickly you can get the value out to the business folks.


An interesting analysis of a traditional marketing analytics technique vs. Deep learning, and what are the pitfalls, implications, considerations for organizations looking to modernize their front office operations by investing in AI.

Artificial Intelligence

A simple read on the Mathematical part of Neural networks and where some of the leading tech companies like Facebook, Google, and

Copyright 2019 © Acies Decisions - All rights Reserved.