Contact form
Your request has been sent.
Please enter your name
Please enter your email
Please enter a correct email address
Please enter your company name
Please enter your message

3 questions to Clément Delangue

← All resources
3 questions to Clément Delangue

By Preligens, 04/01/2022

To face the revolution of the democratisation of the use of artificial intelligence on a large scale, Clément Delangue and his associates have created Hugging Face. He answers our three questions for Roberta.

In your opinion, what are the challenges of the democratisation of artificial intelligence technologies for companies? 

The "democratisation" of machine learning is as much about the uses of AI as it is about the technologies themselves. 

Hugging Face provides over 10,000 companies with an open source platform where multiple AI resources are available to all individuals and organisations who want to easily use the latest AI innovations. That's over 30,000 machine learning models, 5,000 datasets, and 1,000 machine learning demos shared. 

Our models and datasets are very useful for data science teams who want to use them to improve their own applications. Our private hub service allows for safe experimentation and training of models. For less expert users, we offer no-code solutions to train, evaluate and deploy models for different uses. For example, the Spaces workspace makes it possible to test machine learning applications, and the Tasks initiative allows non-experts to seize the resources needed to automate a variety of tasks (summarisation, text generation, object detection). 

The democratisation of machine learning also and above all involves the community.  As an open source company, we are automatically led to animate a community of contributors, from the organisations that make their models and datasets available on our platform, to our partners with whom we work. Following this same philosophy, we have initiated the BigScience partnership project, which aims to make available the largest multilingual model ever created! The training of the BigScience model started in mid-March, and we expect results within 3 to 4 months*. 

Finally, we want the use of machine learning to follow ethical principles and respect regulations. This is why we are investing heavily in research into what can be called "responsible AI". Our ethics teams, led by Dr Margaret Mitchell, who co-founded the AI ethics team at Google, are developing bias assessment and measurement blocks for models and datasets, and we are conducting research into the environmental impact of machine learning models.

2. How can we ensure that the technologies developed, particularly in artificial intelligence, are ethical? 

This is an issue that requires a maximum number of multidisciplinary players to be brought to the table. 

Today, many regulations are being discussed, particularly at European Union level with the AI Act and the Data Governance Act. It is exciting to see that more and more attention is being paid to the opportunities and limits of AI, while allowing innovations to develop in a safe environment. The introduction of concepts such as "data altruism" or "responsible AI" also helps to shed light on alternative approaches with a high social impact.

On our scale, to ensure that technologies evolve in a direction that benefits the greatest number of people, we want to make their potential and their limits understandable to all. To do this, we describe the uses of our models, which allows everyone to try them out online without any particular technical skills, and we warn against the harmful biases identified in certain systems (such as GPT-2 for example).

In the context of data-driven technologies, understanding systems requires understanding their training data. 

3. According to a recent study, 85% of the jobs of 2030 do not yet exist. What is the impact on recruitment and training for a company like Hugging Face? 

Attracting talent is a real issue for technology companies today! At Hugging Face, we have implemented a recruitment approach guided by two motivations: diversity and decentralisation. This is essential for us as we are doubling our workforce this year!

The more multidisciplinary and diverse the teams, the greater our impact. 

Our ambition to develop open source machine learning tools, aligned with ethical principles, is also a real lever: today, more and more talents in the tech industry want to put their skills at the service of technologies that are similar to ours. It also happens that researchers who use our libraries or contribute to our projects from outside integrate our teams.

It is in this sense that we are keen to involve our current teams in recruitment. Most often, the people we want to recruit are not necessarily looking for a job. It is therefore essential to find them in an original way. For example, via Twitter, we run campaigns that look like us: we ask our community what they are interested in at Hugging Face and the conversation can open up into a recruitment discussion. 

At the moment, we are experimenting with a global workshop tour (ML Demo.cratization tour) aimed at students and in collaboration with universities. We want to diversify this content to reach even more people, whether they are already experts in machine learning or not!