Responsible AI: Hugging Face Calls for Rethinking the Role and Practice of Artificial Intelligence

For Hugging Face's climate leader, the difference between "responsible AI" and regular AI is frustrating.

Today, companies developing AI models are in fierce competition for market share and profits. Whether it's Google or Microsoft, the slogan seems to be: "Death to ethical standards." In 2023, both of these large tech companies, which have invested billions of dollars in the industry, have downsized their responsible and ethical AI teams, and there are no signs of pause or reorganization in other AI teams.

For Sasha Luccioni, an AI researcher and climate leader at Hugging Face, the distinction between "responsible AI" and regular AI is meaningless. "It's like the difference between a car with safety measures and a car without safety measures," she sarcastically remarks. Reflecting on the layoffs, she speculates that budget cuts and economic recession may be reasons, but the lack of alignment between teams may be the biggest factor for their dismissal.

"When you have such a distinction, there is too much friction because the work of responsible AI teams is essentially to hinder progress," Luccioni reflects. Recalling Google's dilemma two years ago when the company's renowned AI ethicist Timnit Gebru and her team were fired for warning about the risks of large language models, Luccioni says, "They did what they were supposed to do, but when conflicts arise, responsible AI researchers are the first to be pushed out because they conflict with the company's broader profit model."

In the list of companies trying to find positions for AI ethicists, there is Meta, the multinational conglomerate run by Mark Zuckerberg. The company disbanded its responsible AI department and transferred its resources to different generative AI teams. Luccioni sees this strategy as reasonable and calls it a rational move. "You shouldn't have an isolated responsible AI team; you should have an integrated responsible AI network," she insists.

Similarly, Hugging Face's responsible experts regularly meet with people working on different projects within the company, so there is no tension between the responsible team and other parts of the company. "This means that from the start of a project, we are thinking about responsibility, rather than just adding it as an appendix at the end," Luccioni proudly points out.

Generative AI: Interesting but not revolutionary

Luccioni expresses frustration that artificial intelligence has become so focused on marketing and who generates the most discussion upon release. She points out, "There is little focus on robustness, transparency, or responsibility. Even when models are launched, there are always some issues in the details," citing Google's recent Gemini model release as an example.

This is one of the strategies the company has used to gain an advantage in the field of generative AI. Last year, after ChatGPT became the most talked-about hot topic in the industry, Google went into panic mode and lost $100 billion in valuation after inaccurate answers were given during Bard's live demonstration.

When it comes to this technology, Luccioni's skepticism is evident. "Generative AI is interesting," she admits, "but I haven't seen the business advantage yet. Generating an image is great and may be useful, but it won't change the world. The same goes for ChatGPT or large language models; they are useful for different cases, but they are not the disruptive, revolutionary technologies people claim them to be," she adds.

Technically, Luccioni is very fond of the old-fashioned approach. "I spent some time working with the United Nations and realized that not all the large models we develop in the lab are as useful in the real world," she says.

Breaking the binary

In addition to her research at Hugging Face, the 33-year-old founding member is also a founding member of climate change AI. She has good reasons not to be impressed by this technology that has become a darling of Silicon Valley.

Luccioni is not very interested in the application of technology in climate change. Some researchers have trained a model that can answer questions about global warming and greenhouse effects based on IPCC reports. "But honestly, you can do the same thing with simple information retrieval; you don't need generative AI," Luccioni asserts. While acknowledging the potential use cases, she doubts whether it can change the game. "Especially considering the resources and impact on the planet that these models require, I don't think it's worth using them considering the cost," she firmly adds.

Luccioni suggests taking impactful actions, such as leveraging AI expertise to support existing climate change research communities. "But it's not starting from scratch. They have been doing this work before AI came along," she says. "This work is difficult to publish at conferences because it's usually not novel or cutting-edge. But it can really help the scientific community be more effective in their work."

Luccioni recalls, "When I started working in AI, I thought we were actually creating better, useful technology to help people." "I just read an article about Imagen 2, which is great research, but why are we doing this? What problem does it solve?" she ponders. "I feel like the values in this field have changed, and I find it hard to get excited about these things anymore."

Seven years ago, Luccioni was a researcher working on applied AI in the finance sector. She changed her true passion - climate and nature. She quit her job, took a significant pay cut, and decided to use her AI background to help fight climate change. After collaborating with computer scientist Yoshua Bengio for a while, she sought a space between academia and industry.

"I had several job offers from big tech companies, and then I chose Hugging Face because I believe in the mission and importance of making AI as accessible as possible," she says.

"That's why I continue to do what I do because I want to understand it better. It's not a binary thing for the Earth. AI is not a question of good or bad; it's a question of what areas it can be beneficial to. But to figure out whether it's good or bad, you need to understand its impact," Luccioni concludes.