The AI Con
Emily Bender and Alex Hanna (UK); (US))
An alternative history of Silicon Valley can be told not through its successes 鈥� products like the iPhone, Facebook or Google 鈥� but through its scams, like Theranos鈥檚 faulty blood tests or FTX鈥檚 dodgy crypto accounts. From a bird鈥檚 eye view, these companies鈥� stories share a formula: invent a world-changing idea, generate hype, convince investors of potential, and then expand as far and fast as possible. But under closer scrutiny, the hype can falter 鈥� and quickly unravel.
It is that scrutiny that today鈥檚 AI industry desperately needs, argue Emily Bender and Alex Hanna in their new book The AI Con: How to fight big tech鈥檚 hype and create the future we want. In their telling, we have been massively oversold on the capabilities of these systems, and the hype generated by large AI companies is so vast and all-consuming that it threatens genuine societal harm in its wake.
Advertisement
If you have paid any attention to the many downsides of AI in recent years, then much of the evidence presented in the book won鈥檛 seem new: systemic biases, intellectual property theft and overhyped claims that fail to be replicated. Yet, Bender and Hanna write, the AI bubble continues to inflate. It is, then, their 鈥減rimary goal to inhibit the next tech bubble鈥�.
With that goal in mind, the contours of the book make sense. The authors are building the strongest case possible against thoughtlessly integrating AI systems into our lives without considering the wide-ranging societal and personal consequences. But in building such a case, it can sometimes feel like they are constructing an AI straw man, and genuine academic inquiry into what these systems, at scale, might be doing is left out.
They spend a chapter looking at what these tools are, both more traditional AI systems and newer large language models (LLMs) that power services like ChatGPT, but their details are scant. They conclude that ChatGPT is nothing more than 鈥渟ouped-up autocomplete鈥� and a 鈥渟ynthetic text extruding machine鈥�, and that it definitely isn鈥檛 sentient.
But the argument over sentience, which many scientists would rightly scoff at, misses a more interesting discussion. It is still an outstanding research problem as to how LLMs can solve problems that would appear to require a form of reasoning, like genuine mathematical solutions, or control. In framing the debate around whether or not these machines are thinking like humans, they omit scientific research into how the scale of large systems can produce new, emergent capabilities as complexity multiplies.
This may not justify the current AI hype bubble, but it seems odd to ignore the reasons that many sensible people are excited about these models, and instead argue that AI should be restricted to narrow uses, like processing images for radiologists or checking spelling.
Despite this frustrating oversight, many of the arguments that Bender and Hanna make about AI鈥檚 detrimental effects on society are compelling 鈥� and correct 鈥� regardless of whether you think these systems might be more powerful than they suggest. They rightly point out that the history of AI is littered with both overpromises and prophecies of doom, and that venture capitalists do see modern AI systems as a way to reduce jobs and increase profits for a select few.
These are all issues of genuine concern that are glaringly missing from discussions in governments and big tech boardrooms. But the increasing divide between AI evangelists and self-proclaimed neo-Luddites will be bridged only when both sides are honest about the true picture.
When presented in such a one-sided way, it isn鈥檛 clear The AI Con will convince anyone who hasn鈥檛 already made up their mind.
快猫短视频 book club
Love reading? Come and join our friendly group of fellow book lovers. Every six weeks, we delve into an exciting new title, with members given free access to extracts from our books, articles from our authors and video interviews.
