Are We Overestimating GenAI?

We could be overestimating what GenAI is capable of, according to MIT robotics pioneer Rodney Brooks. In an interview with TechCrunch, Brooks noted that while GenAI is perfectly capable of performing a certain set of tasks, it can’t do everything a human can. Yet we keep overestimating its capabilities.

“When a human sees an AI system perform a task, they immediately generalize it to things that are similar and make an estimate of the competence of the AI system; not just the performance on that, but the competence around that,” he said.

In a nutshell, the reason we are overly optimistic is because we subconsciously apply a model of a person’s performance on a given task and extrapolate from there. Yet GenAI isn’t human, and it’s a flawed notion to assign human capabilities to it.

There’s an LLM for that

Moreover, for its profound capabilities, there are plenty of scenarios where GenAI won’t be a good fit, says Brooks. That’s an interesting remark coming from the co-founder of multiple startups, including his latest – Robust.ai, which deploys robots with “human-centric design and AI-powered workflows.”

According to Brooks, someone had recently suggested that he build an LLM to tell his warehouse robots where to go. He says this won’t work.

“When you have 10,000 orders that just came in that you have to ship in two hours, you have to optimize for that. Language is not gonna help; it’s just going to slow things down,” Brooks said.

He does concede that GenAI could lead to robots with useful language interfaces for people in care situations. Just not in the warehouse. There’s traditional AI and other solutions that utilize data science.

“We have massive data processing and massive AI optimization techniques and planning. And that’s how we get the orders completed fast.”

To find out more about what Singapore companies are doing with GenAI, join the 2nd Annual DataScience&AITrends Summit on 29 August, 2024. Register here.

Limits are inevitable

I recently wrote that LLMs show no sign of plateauing and pointed to remarks from OpenAI that it intends to double down on even larger GenAI models.

“[We] will continue to push the scaling paradigm so models get more powerful as we put more compute, high-quality data, and more data into them. There is no evidence of the scaling paradigm stopping anytime soon; we should expect AI models to become much more powerful [just by scaling up],” Mira Murati, CTO of OpenAI had said.

However, Brooks thinks it’s a mistaken belief that there will always be exponential growth in technology. Specifically, that GPT-5, GPT-6, and GPT-7 will each exhibit the same kind of leap ahead that GPT-4 demonstrated.

Things don’t work quite as well in the real world, says Brooks. He pointed to the iPod and how its onboard memory grew quickly from its initial 10GB, doubling in capacity over each subsequent generation. But this growth tapered off at 256GB by the final model two decades later, instead of the 160TB that initial growth suggested.

The reason? Nobody actually needed more than that. Could this be the case with GenAI?

Ultimately, organizations should always look at the business case and the return on investment, says Brooks. And accept that there is a long list of hard-to-solve outlier cases that AI cannot solve today.

Image credit: iStock/BeritK