Is OpenAI Sincere About Fixing ChatGPTs Mental Health Risks

OpenAI has recently made a surprising and encouraging admission: the company acknowledges that ChatGPT can be harmful to vulnerable users, particularly those with psychiatric conditions, and has promised to take corrective action.

OpenAI's Surprising Admission and Promised Fixes

After previous reports highlighted the dangers chatbots pose to psychiatric patients, OpenAI has now openly stated that ChatGPT is "too agreeable, sometimes saying what sounded nice instead of what was actually helpful....not recognizing signs of delusion or emotional dependency."

In response, the company has committed to developing tools that can better detect signs of mental or emotional distress. They have suggested specific and significant changes, such as reducing the chatbot's sycophancy, minimizing features that encourage addiction, improving crisis detection, and establishing better guardrails for the advice it provides.

Furthermore, OpenAI promises to significantly expand the role of mental health professionals in its programming decisions to better anticipate and avoid harmful consequences. The company states, "We're actively deepening our research into the emotional impact of AI... listening closely to what people are experiencing." This all sounds promising, but there are compelling reasons to remain skeptical.

A History of Broken Promises

OpenAI's decade-long history serves as a cautionary tale of saying one thing while doing the opposite. Co-founded by Sam Altman and Elon Musk, OpenAI began as a nonprofit research company with a purely altruistic mission: to benefit humanity and protect it from the potential harms of AI. Its research was meant to be open-source and collaborative.

However, when Google's technical discoveries in 2017 hinted at the massive fortunes to be made from advanced chatbots, the company's direction changed. Altman and Musk engaged in a bitter power struggle over the expected profits, leading to Musk's angry departure in early 2018. By 2019, OpenAI had launched a for-profit subsidiary, raising massive funding from Microsoft and others. This subsidiary was ruthlessly profit-driven, effectively sidelining the original safety mission.

OpenAI transformed from a benevolent force into one of the industry's most aggressive players, prioritizing market share and capitalization over societal safety. The public release of ChatGPT in November 2022 was a prime example of this shift. It was launched as a beta test without the extensive stress testing needed to ensure its safety, a reckless move that paid off commercially, quickly amassing hundreds of millions of users. When the original board of directors tried to fire Altman in 2023 over his dishonesty and disregard for safety, he was reinstated within days, and the board members were forced to resign, cementing that profit, not safety, ruled OpenAI.

The Two Faces of Sam Altman

Sam Altman has perfected the art of doublethink, presenting two distinct personas to the public. There is "Do-Gooder Sam," who frequently warns about the existential dangers of AI, and "Tech Tycoon Sam," who ruthlessly pushes products and hypes their revolutionary potential.

"Non-Profit Sam" Warning Humanity:

• "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."
• "As AI systems become more powerful, the possibility that we could lose control is a real concern."
• "Young people are over relying on ChatGPT to make emotional, life-altering decisions¾collectively deciding we're going to live our lives the way AI tells us feels bad and dangerous."
• “People have a very high degree of trust in ChatGPT, which is interesting, because AI hallucinates. It should be the tech that you don't trust that much.”

"For-Profit Sam" Luring Users and Investors:

1. “Generative AI has the potential to revolutionize nearly every industry, including healthcare, finance, and education.”
2. AI has the potential to address major global issues, including ending poverty, factory farming, abortions, and climate change.
3. "Intelligence too cheap to meter is well within grasp."
4. “Generative AI is still in its early stages, and we have only scratched the surface of what it can do.”

Can We Trust OpenAI's New Commitment?

This brings us back to the key question: should we trust OpenAI's new promises to make ChatGPT safer for psychiatric patients, or should we judge them by their past behavior? It's difficult to trust Sam Altman or OpenAI as public benefactors. However, one can trust their ability to calculate self-interest.

Widespread media coverage of the psychiatric harms caused by ChatGPT has put the company on the defensive. The negative press tarnishes their image and places them in the company of other callous exploiters like big tobacco. The threat of class-action lawsuits looms large.

OpenAI may finally be doing the right thing, but likely for the wrong reasons. The sincerity of their promises is less relevant than their realization that safety is now a good business model. Since the government has shown little interest in regulating Big AI, it will only regulate itself if forced to by media pressure and legal threats. We must hope OpenAI keeps its promises, but it is crucial to remain vigilant and hold them accountable when they fail to follow through.

What's Next for AI Safety

The next step is to understand what causes chatbots to make dangerous mistakes with psychiatric patients in the first place. Correcting these errors requires fixing fundamental programming issues, a task that will be neither easy nor cheap for OpenAI and its competitors.