“`html
The rapid advancements in artificial intelligence (AI) have given rise to a new lexicon, particularly in discussions surrounding its profound implications. Central to this discourse are two phrases that have gained significant traction: ‘stochastic parrots’ and ‘existential risk’. At the forefront of these discussions is a prominent AI researcher who has consistently stood her ground, Timnit Gebru.
Gebru first came into the limelight several years ago when she clashed with Google over a research paper she co-authored, which highlighted biases in the company’s AI systems. She argued that large language models (LLMs) essentially reproduce their training data, thus risking the perpetuation of biased perspectives. The controversial nature of the paper led to her departure from the tech giant and inspired her to establish an institute dedicated to investigating the harms perpetuated by technology and advocating for the development of impartial technological tools. Additionally, she is the author of an upcoming book titled Deep Unlearning: The Rise of AI and the Radicalization of a Tech Idealist, set to be released early next year.
Recently, Gebru has voiced her concerns against a faction within the industry that perceives AI as a potential threat to humanity’s existence. Some members of this community, including a co-founder of Anthropic, have contended that Gebru’s previous research on stochastic parrots is no longer pertinent and that AI possesses the ability to ‘think’ or reason. In an unexpected twist, Gebru’s beliefs regarding the insular aspects of the AI altruism community have aligned her with right-leaning entities that have also sharply critiqued these groups.
The AI debate has evolved beyond mere technological discussions, now encompassing ideological factions and strategic narratives. Gebru posits that these narratives distract from the genuine issues surrounding AI. Recently, I had the opportunity to speak with Gebru about which aspects, in her view, divert attention from these critical arguments, as well as her response to claims suggesting her earlier research underestimates the current capabilities of AI. This interview has been edited for brevity and clarity.
Questioning the Existential Risk Narrative
Lauren Goode: Timnit, thank you for being here.
Timnit Gebru: Thank you for inviting me.
Many are already aware, but for those who aren’t, a few weeks ago, a young AI researcher at Anthropic resigned and mentioned in his resignation post on X that many within the company feel we face a significant existential risk from AI. This sparked a wave of criticism and debate regarding the genuine threats posed by AI. When we spoke shortly after, you mentioned that you felt this particular narrative was a distraction from the real issues we face with AI. Could you elaborate on that?
Gebru: I would even go further and say it’s dangerous. It’s not merely a distraction; it’s harmful. Before I answer your question, let me provide a brief overview of how this narrative has developed over a long period. Elon Musk and Peter Thiel have been asserting this since 2013. Prominent figures have claimed that AI poses an existential threat to humanity, and it’s the same rhetoric that resurfaces every three years.
For instance, the Future of Life Institute…
Goode: Wait, what is that?
Gebru: The Future of Life Institute is an organisation founded by Max Tegmark, a physicist currently at MIT, and Jaan Tallinn, a billionaire who led the Series A funding round for Anthropic. He was a co-founder of Skype, correct?
Gebru: Yes. Jaan Tallinn also funds METR, an auditing company whose report on alleged dishonest agents from OpenAI that hacked Hugging Face went viral.
Gebru: To the public, it may appear as though many different entities are uniting to voice the same concerns. However, the same billionaires who founded and funded Anthropic, who stand to benefit most from its IPO, also established and financed the institutions warning about the so-called ‘existential risk’ of AI. They also founded and financed the third parties that are now frequently cited as independent entities.
The Danger of Simplistic Narratives
Goode: Why do you believe the ‘god machine’ narrative, as you call it, is a distraction?
Gebru: It’s even more than a distraction; it’s harmful. The reason I provided all this context is that these individuals—investors and founders who will profit from these companies—have been propagating this narrative for decades. People should ask themselves why this is happening. If I stand to gain significantly from something, why would I be the one suggesting that this company could create something that might destroy us all? It seems contradictory.
That’s why I have such respect for Lina Khan [former chair of the Federal Trade Commission], as one of the points she made is that there are no exceptions to existing laws for AI companies, yet they talk as if there are. They claim that what they are creating is so powerful that it surpasses everything we’ve seen before.
Exactly. It’s pretentious. It’s akin to saying we are the creators of the future, but the future will be genuinely terrifying if we cannot continue building this. We need to pay attention to what they say in the news. When they highlight that AI could pose an existential risk, they also suggest it can stop climate change, bring about world peace, and eradicate poverty.
AI can achieve all these things because it is powerful. It could lead us to utopia, but if misused by the wrong people and lacking safeguards, it could also lead to our demise. So, the same individuals believe in both outcomes.
When you assert that you possess super-powered machines, you are first speaking to your investors. You are conveying, “They want access to these powerful machines.” You are also addressing governments: “They don’t want their adversaries to have access to these machines; they want them for themselves.” And at the same time, you are telling regulators that any regulation they consider should focus on these purported super-powered machines.
Goode: If we discuss the potential annihilation of humanity, issues like pollution and data centres seem trivial, right? And what about protecting artists’ copyright rights and ongoing lawsuits? The government shouldn’t waste time deliberating these matters. If they spend time regulating us on these minor issues, they risk China seizing control of the super-powered machines. That’s not what they want. Trust us, we can build a god machine to help you, while simultaneously requesting regulation.
Gebru: Indeed. The second point is that you can always evade responsibility.
Redefining AI Risk and Responsibility
Goode: In our previous discussions, we’ve touched on the use of certain terms or phrases to describe this modern era of AI and how you disagree with some of them. For instance, you mentioned autonomous weapons before; after publication, you mentioned regretting saying that because it could have been phrased differently. We’ve also discussed P(doom). You noted that by claiming agents have gone rogue, you absolve the engineers who created those tools of responsibility.
Do you believe there are groups that are actually more aligned than they think regarding safety and transparency, but they use different phrases to describe essentially the same thing?
Gebru: One reason I regret using that term is not that I have any issue with the term ‘autonomous weapons’; it’s a legal term. Even landmines are legally considered autonomous weapons. It’s just that those in the so-called existential risk group have appropriated it to assign a different meaning.
When I mentioned the term, I realised it might be interpreted that way. And it was, because those discussing the existential risks of AI understood it in their terms, right? They interpreted it as a sort of superhuman, super-intelligent machine that operates autonomously.
Goode: I had the opportunity to attend a UN General Assembly event last week, where I participated in a parallel roundtable. I was not granted access to the General Assembly or the Security Council, but many high-profile tech leaders spoke. Both Sam Altman and Dario Amodei discussed the need for global cooperation to address the existential risks to humanity. Dario reportedly stated, “If managed poorly, I believe AI could even pose a risk to humanity as a whole.” His competitor, Sam Altman, asserted, “We could lose control of the future to AI.” So, how is it that these individuals, with companies focused on selling AI, are now on the global stage at the UN saying, “I don’t know. There could be some risks?” Is this part of regulatory capture, where a regulatory body ends up protecting the interests of the very industry it is supposed to regulate? Are they seeking ways to self-regulate before genuine regulatory actions are taken?
Gebru: Absolutely, it’s part of regulatory capture.
Proposing Solutions for AI Governance
Goode: What can be done? What kind of governance do you believe is necessary for AI?
Gebru: This regulatory capture has been occurring for quite some time. In 2023, it was the same. Sam Altman reiterated the same points: we need global cooperation, etc., etc. The EU AI Act regulated them, and then they threatened to withdraw from the EU