An AI Ethics Framework

The conversation around ethical AI is getting louder by the day and we want to take a moment to share with you how we at DATAmundi, as a data services provider, approach the ethical side of artificial intelligence and data services. In part to give you a peek behind the curtain but also to reassure you that in everything we do, we consider data privacy, transparency and ethics.

In our plans we strive to:

  • Mitigate bias
  • Provide transparency
  • Ensure data privacy
  • Ensure inclusivity

 

Bias mitigation starts with ethical sourcing

It’s not just about making sure AI models don’t perpetuate or amplify social, racial, or gender biases. It’s about how we source the data, too, so that annotated data does not enter the model with biases.

We collect, annotate, and use data in ways that are legal, respectful of individual and collective rights, and transparent. The client is informed of the sources we use, and we obtain informed consent from human participants.

Bias, for example, is not just about a personal preference for coffee over tea, cookies over cake, or cats over dogs. It is more nuanced than that when it relates to data and becomes even more granular when we look at different languages. If we’re building a dataset that includes languages that label certain words with a gender, like you find in French, it means we need to make sure to tag the word but also offer a context and an alternate word without gender.

Nurse in French is a feminine word and without context an AI could easily assume then that all nurses are female. Since that is not the case in real life, that’s a bias we look for and correct in the data set before training the AI.

If we were to use voice recordings of nurses, we would make sure to source an equal number of male nurses.

Ethical data sourcing to comply with regulations

In much of the work we do as data service providers, we work with specialized data sets, often in finance, medical/pharmaceutical, or other heavily regulated industries. In addition, creating data sets that will be used in AI tools in the European Union, means meeting strict data privacy regulations as set out by GDPR.

Our data gathering and use is in compliance with GDPR, HIPAA, and CCPA, wherever these regulations are enforced, keeping our data legal, transparent and respectful of individual rights. This means we:

  • Obtain informed consent from data contributors
  • Compensate data contributors fairly
  • Avoid exploitative practices
  • Ensure compliance with regulations
  • Validate the authenticity of the data

 

Ethics in AI are just good business

We maintain a high ethical standard at every stage of the data lifecycle because we understand the value to our customers and our reputation. As AI data increases its reach across industries and individual consumers, it’s essential that companies maintain high standards.

We want to make sure that the data we provide you is as free of bias as we can make it, and not likely to be at risk of future litigation or penalties.

Recently we were asked to source accents and dialects for an audio-based project. We were aware of potential sensitivities in the different groups we were asked to source dialect sounds from. To make sure we didn’t offend any ethnic or racial groups in our dialect sourcing, we very carefully worded our advertisements to source participants and kept to a strict 1:1 ratio of male and female. We sourced along an age range of 6 – 65. Participants appreciated our sensitivity and cultural understanding, and our client was pleased with the resulting data set.

The social responsibility of ethics in AI

With more and more companies offering data services, we all play a role in shaping the future of technology. Regulation of AI is still in the early stages, so it falls to the companies involved – AI data providers – to set a standard of quality and ethics, and to offer an example of socially responsible business practices.

At DATAmundi we do just that. We are open to talking with our customers, and with the subject matter experts and linguists we hire, about any ethics concerns they might have. We work to address them at every step along the way in a project.

Contact us to talk about your data project needs!