We have ChatGPT because OpenAI CEO Sam Altman wanted to build a god.
For all the buzz, the chatbot is only a prototype along the way to a loftier goal of “artificial general intelligence” (AGI) that surpasses the cognitive abilities of humans. When Altman cofounded OpenAI in 2015, he made it the non-profit’s goal. Five years earlier, Demis Hassabis cofounded DeepMind Technologies Ltd, now Google’s core artificial intelligence (AI) division, with the same AGI objective. Their reasons were utopian: AGI would create financial abundance and be “broadly beneficial” to humanity, according to Altman.
It would cure cancer and solve climate change, according to Hassabis.
There are problems with these noble goals. First, the financial incentives of large tech firms are likely to skew AGI efforts toward benefiting their coffers first and foremost. Those early altruistic objectives of Altman and Hassabis have fallen by the wayside in the last few years as the generative AI boom has sparked a race to “win,” whatever that means. In the past few years, DeepMind’s Web site has removed content on health research or discovering new forms of energy creation to become more product-focused, spotlighting Google’s flagship AI platform Gemini. Altman still talks about benefiting humanity, but he is no longer a non-profit “free from financial obligations” per his 2015 founding statement, and more of a product arm of Microsoft Corp, which has since sunk about US$13 billion into his company.
The other issue is that even the people who are building AGI are fumbling in the dark, despite how sure they are of their timeline predictions. Anthropic CEO Dario Amodei has said the world would have AGI by 2027.
Altman has said it is “a few thousand days” away, and we would have the first AI agents joining the workforce this year.
Billionaire Masayoshi Son thinks we would have it in two or three years.
Ezra Klein, a New York Times podcaster who regularly has AI leaders on his show, recently wrote: “It’s really about to happen. We’re about to get to artificial general intelligence.”
However, ask the tech leaders what AGI actually means, and you would get a smorgasbord of answers. Hassabis described it as software that can perform “at human level.”
Altman said it would “outperform humans.”
Both, alongside Amodei, often take greater pains to talk about the complexity and challenges of defining AGI. Altman has also called it a “weakly defined term.” Microsoft CEO Satya Nadella has even derided the AGI effort as “nonsensical benchmark hacking.”
I get little comfort from seeing AI’s top leaders trash talking or dancing around the definition of their North Star, while simultaneously racing toward it. Not just because computer-science experts, along with Elon Musk, worry that the advent of AGI would also come with rather high stakes, existential risks to human civilization, but because tilting at a vaguely defined goal opens the door to unintended consequences.
A more worthwhile aspiration would be narrower and more concrete, such as building AI systems that reduce medical diagnostic errors by 30 percent. Or in education, improving math proficiency in students by 15 percent. Or systems that could enhance energy grids to reduce carbon emissions by 20 percent. Such goals not only have clear metrics for success, but serve concrete human needs, just as AI builders such as Altman and Hassabis originally envisioned.
There is no evidence that when a company such as Google or Microsoft — or China’s DeepSeek — claims to have finally built AGI, it would have the key to curing cancer, solving climate change or increasing the wealth of everyone on earth by “trillions” of dollars, per Altman. So intense has the arms race become that it seems more likely they would instead position themselves as having a competitive advantage in the market, raise prices and lock down information sharing. There would be questions about geopolitical ramifications. When he founded OpenAI, Altman said that if his team ever noticed another research lab was getting closer to AGI, they would down tools and collaborate. That looks like a pipe dream today.
Last month a large group of academic and corporate AI researchers published a paper that called on tech firms to stop making AGI the be-all and end-all of AI research. They argued that not only was the term too vague to measure properly, creating a recipe for bad science, it left key people out of the conversation — namely, all those whose lives would be changed. Technologists have far more societal power than they did at the turn of the millennium, able to reshape culture and individual habits with the social media products they have deployed, and now large swathes of jobs too with AI, with few checks and balances.
The researchers, including Google’s former AI ethics lead Margaret Mitchell, suggest not only that tech firms include more voices from different communities and fields of expertise in their AI work, but that they drop the vague shtick about AGI, which one scientist memorably defined for me as “the rapture for nerds.”
The obsession with “bigger is better” has gone on long enough in Silicon Valley, as has the jostling between people such as Musk and Altman to have the biggest AI model or the biggest cluster of Nvidia Corp’s AI chips. J. Robert Oppenheimer had much regret for his role as the father of the atomic bomb, and now epitomizes the notion that “just because you can, doesn’t mean you should.” Rather than look back in hindsight with remorse, tech leaders would do well to avoid the same trap of gunning for glory and trying to build gods, especially when the benefits are far from certain.
Parmy Olson is a Bloomberg Opinion columnist covering technology. A former reporter for the Wall Street Journal and Forbes, she is author of Supremacy: AI, ChatGPT and the Race That Will Change the World. This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
The first Donald Trump term was a boon for Taiwan. The administration regularized the arms sales process and enhanced bilateral ties. Taipei will not be so fortunate the second time around. Given recent events, Taiwan must proceed with the assumption that it cannot count on the United States to defend it — diplomatically or militarily — during the next four years. Early indications suggested otherwise. The nomination of Marco Rubio as US Secretary of State and the appointment of Mike Waltz as the national security advisor, both of whom have expressed full-throated support for Taiwan in the past, raised hopes that
There is nothing the Chinese Nationalist Party (KMT) could do to stop the tsunami-like mass recall campaign. KMT Chairman Eric Chu (朱立倫) reportedly said the party does not exclude the option of conditionally proposing a no-confidence vote against the premier, which the party later denied. Did an “actuary” like Chu finally come around to thinking it should get tough with the ruling party? The KMT says the Democratic Progressive Party (DPP) is leading a minority government with only a 40 percent share of the vote. It has said that the DPP is out of touch with the electorate, has proposed a bloated
Authorities last week revoked the residency permit of a Chinese social media influencer surnamed Liu (劉), better known by her online channel name Yaya in Taiwan (亞亞在台灣), who has more than 440,000 followers online and is living in Taiwan with a marriage-based residency permit, for her “reunification by force” comments. She was asked to leave the country in 10 days. The National Immigration Agency (NIA) on Tuesday last week announced the decision, citing the influencer’s several controversial public comments, including saying that “China does not need any other reason to reunify Taiwan with force” and “why is it [China] hesitant
A media report has suggested that Chinese Nationalist Party (KMT) Chairman Eric Chu (朱立倫) was considering initiating a vote of no confidence in Premier Cho Jung-tai (卓榮泰) in a bid to “bring down the Cabinet.” The KMT has denied that this topic was ever discussed. Why might such a move have even be considered? It would have been absurd if it had seen the light of day — potentially leading to a mass loss of legislative seats for the KMT even without the recall petitions already under way. Today the second phase of the recall movement is to begin — which has
RSS