Menu
Equality brings opportunity
True AI can become a powerful ally to amplify diversity and fairness
Our mission
To neutralize sensitive factors in hiring processes by mitigating discriminatory biases. We prioritize specific Arfitificial Inteligence resources to find professionals that truly contribute to the growth of organizations. We analyze competencies and identify which features need to be addressed to avoid discrimination, amplifying equality and creating new opportunities for individuals and organizations.
Our approach
Artificial Intelligence has the power to mitigate discriminatory unconscious bias. One of the advantages of our AI approach is the ability to program it to deliver “direct job related specifications”. Sensitive data such as ethnicity, gender and age are not used in our process, however some data points such as address, educational background and previous employment are linked to this sensitive data and could lead to discriminatory biases. In order to mitigate theses biases, we can intervene in three different moments: before the data is processed, while the algorithm is trained, and finally, in the prediction the model delivers.
Data preprocessing
Training the perfect model
Defending mathematical hypotheses is part of the preliminary work in any AI project. Defining the scope of the research and verifying the data that can be used to teach an algorithm how to interpret and solve a problem is the challenge our team faces. With completely anonimized data we preserve the identity of all analyzed individuals, while working to identify whether there are any details in the historic data that can negatively affect a particular group or individual in the algorithm’s recommendations. We apply bias mitigation techniques in the data preprocessing stage to make sure the competencies, abilities, experiences and education are analyzed in the fairest terms possible. Training the algorithm correctly is the key to success.
During data processing and after outputs
Understanding the results
In the algorithm modeling stage, we can identify the performance indicators, that is, which features influence performance positively or negatively. By rebalancing the importance of different features we can understand the impact of elements that could generate discriminatory biases. As part of our methodology, Rocketmat opts to partially reduce the precision in models in order to achieve greater fairness in the recommendations made by our artificial intelligence.
Rebalancing features
Without altering the features (characteristics) and labels (what we seek to predict) we train the models until we have a more balanced and fair result.
Mitigating and Optimizing
We identify features (characteristics) that can generate discriminatory biases and mitigate them, while amplifying features that promote fairness. In this stage, transparency is essencial to the projects' success.
Identifying unpriviledged groups
We analyze the data sets to identify whether there are unpriviledged groups or individuals, who are placed at a systematic disadvantage. We then act to ensure the algorithm is adjusted to combat this issue.
"Awareness of the business case for inclusion and diversity is on the rise. While social justice typically is the initial impetus behind these efforts, companies have increasingly begun to regard inclusion and diversity as a source of competitive advantage, and specifically as a key enabler of growth."
#diversitymatters
Mckinsey
2020
GDPR
The General Data Protection Regulation
The General Data Protection Regulation (GDPR) was created and voted in 2016, entered into force on 25th May, 2018 and aims to protect the European citizens/residents data integrity. The new rules either provides to them, the possibility to manage, migrate, edit or delete his/her personal data from any platform. The company in a non-compliance with the new law terms, may be punished and pay high fines.
Rocketmat’s customers are private entities that need to make decisions (hiring or managing people) and, for this, they have data from their candidates or employees.
Rocketmat is a private entity that performs the processing of data obtained by its customers, for the purpose of assisting decision making.
In view of this, Rocketmat qualifies as Data Operator, while Rocketmat’s customers are considered Controladores dos Dados – Data Controllers.
Rocketmat does not store any data that allows the identification of a specific individual in its databases. Instead of identification, encrypted keys are stored in Rocketmat’s databases, through which only the Rocketmat’s customer can compare with internal keys and identify the – Data Holder.
In case the candidate or employee – the Data Holder – contacts Rockemat – the Operator – it is not possible to identify the location of information in Rocketmat’s database. Thus, it is up to Rockemat’s customer – the Controller – to request any action related to the Holder’s data, including:
– Access to the data.
– Confirmation of the existence of treatment.
– Correction of incomplete, inaccurate or outdated data.
– Data portability to another service or product provider.
– Anonymization, blocking or deletion of data treated in non-compliance with the GDPR.
– Information of public and private entities with which the controller has shared the use of data.
– Elimination of personal data processed with the consent of the holder
– Information about the possibility of not giving consent and about the consequences of the refusal.
– Automated decision review.
– Opposition to the treatment carried out based on one of the hypotheses of waiver of consent, in case of non-compliance with the provisions of the law.
– Revocation of consent.
It is also the responsibility of Rocketmat’s customer – the Data Controller – to obtain explicit consent so that personal data can be collected and processed. The Controller must also prepare and / or update the appropriate terms of use documents, the authorization extensions and the need to acquire such data.
All other terms below – including Data Breach Policy & Response Plan – refer to both GDPR applications.