Gains with generative AI are still uncertain – 08/23/2023 – Market

Even by the astonishing standards of past technology cycles, generative artificial intelligence enthusiasts are overdoing it.

Trillion-dollar companies, including Alphabet and Microsoft, declare AI to be the new electricity or fire, and are restructuring all of their businesses around it. Never knowingly outdone, venture capitalists are also pumping money into the sector. Fifty of the most promising generative AI startups identified by CB Insights have raised more than $19 billion in funding since 2019. Of these, 11 now count as unicorns, with valuations in excess of $1 billion.

Even sober McKinsey executives estimate the technology could add between $2.6 trillion and $4.4 trillion in economic value annually across 63 analyzed use cases, ranging from banking to life sciences. In other words, in very rough terms, generative AI could create a new UK economy each year (the country’s gross domestic product was $3.1 trillion in 2021).

But what if they are wrong? In a series of provocative posts, technologist Gary Marcus explores the possibility that we will see a “massive and harrowing correction” in valuations as investors realize that generative AI doesn’t work very well and lacks “killer” business applications. “Revenue has not yet arrived, and may never arrive,” he writes.

Marcus, co-founder of the Center for Advancing Trustworthy AI, who testified before the US Congress this year, had long been suspicious of the intelligence of the neural network models that preceded newer chatbots like OpenAI’s ChatGPT. But it also raises some new truths about generative AI. Consider the unreliability of the models themselves. As is now clear to millions of users, one of the biggest disadvantages of technology is that it hallucinates – or confabulates – facts.

In his previous book, “Rebooting AI” [Reiniciando a IA], Marcus provides a good example of how this can happen. Some AI models operate like probabilistic machines, predicting responses from data patterns rather than displaying reasoning. A French speaker would instinctively understand “Je mange un avocat pour le déjeuner” to mean “I have an avocado for lunch”. But in its first iterations, Google Translate translated it as “I’m going to eat a lawyer for lunch.” In French, the word “avocat” means “avocado” and “lawyer”. Google Translate chose the statistically most likely translation rather than one that made sense.

Tech companies say they are reducing errors by improving their systems’ contextual understanding (Google Translate translates that phrase into French accurately now). But Marcus argues that hallucinations will remain a feature, not a defect, of generative AI models, which cannot be fixed using his current methodology. “There’s a fantasy that if you add more data it will work. But you can’t solve the problem with data,” he tells me.

For some users, this inherent insecurity is a hindrance. Craig Martell, director of AI at the US Department of Defense, said last week that he would require an accuracy level of “five 9s”. [99,999%] before using an AI system. “I can’t have a hallucination that says ‘Oh yeah, put widget A connected to widget B’ — and it explodes,” he said. Many generative AI systems put too high a “cognitive load” on the user to determine what is right or wrong, he added.

Even more troubling is the idea that content produced by generative AI is polluting the datasets that future systems will be trained on, threatening what some call “model collapse”. By adding more imperfect information and deliberate disinformation to our knowledge base, generative AI systems are producing yet another “emerdification” of the internet, to use Cory Doctorow’s suggestive term. This means that training sets will produce more nonsense instead of less.

Undaunted, investors typically come up with three arguments for how to make money from generative AI. Even with its imperfections, they say, it can still be a valuable productivity tool, accelerating the industrialization of efficiency. There are also many uses, from copywriting to call center operations, where a “two 9s” level of accuracy is acceptable.

Second, investors are banking on the fact that some companies can deploy generative AI models to solve narrow real-world problems. The latest advances in AI allow data to be analyzed in real time, says Zuzanna Stamirowska, executive director of French startup Pathway, helping to optimize maritime trade or the performance of aeronautical engines, for example. “We really focus on enterprise use cases,” she says.

Third, generative AI models will enable the creation of new services and business models not yet imagined. During the mass electrification of the economy in the late 19th century, companies profited from generating and distributing electricity. But the big fortunes were made later, using electricity to transform ways of making things, like steel, or invent entirely new products and services, including home appliances.

Right now, it’s only cloud computing providers and chipmakers that are really making money in the generative AI boom. No doubt Marcus will also be right that much of the corporate money invested in technology will go to waste and most start-ups will fail. But who knows what new things will be invented and endure? That’s why God invented bubbles.

Translated by Luiz Roberto M. Gonçalves