The AGI Debate Is a Distraction from Real Progress

Who cares when we are getting to AGI. Even in its current state, AI is already useful enough to do things that we weren’t capable of doing before.

In other words, it does not matter when we are going to get to AGI.

My hunch is that we will NEVER see AGI. We might see something asymptotically approaching AGI, but not quite there. There will always be some skills that fall on the wrong side of the jagged frontier, allowing humans to quickly tell it’s an AI: messed-up hands, getting tricked into being malicious when a prompter mentions a grandma with cancer etc. But us not reaching AGI is another discussion for another time.

So, let’s get back to this. If we are never getting to AGI, what does that mean for us? I’m going to argue that the current iteration of AI is already powerful enough to shift how we work dramatically. We are severely underusing this new technology we already have. And it will take DECADES before AI becomes a natural part of our workflow.

The current iteration of AI is capable enough and it already enables us to do many wonderful things. AWS claimed that they have saved 4,500 years of development time. Perplexity, powered by LLMs, is challenging Google’s hegemony in information search. BCG consultants are faster, better, and cheaper. Anyone who has spent a bare minimal amount of time learning how to prompt well has gained significant benefits in work and life improvements. Even if the entire AI field stops developing further today, what we have is already fantastic. Incorporating AI into work will indubitably increase your work/life efficiency.

However, looking around, you should notice very quickly that nothing around you has changed all that dramatically. Unless you are the kind of techies / early-adopters of technology (and possibly working at a company that does not think too much about you dumping data into LLMs), your life has been… roughly the same. The only shitty part is that your social media is probably now spammed by AI-generated content.

That’s not the fault of the technology. That’s the fault of the people using it. Do you know how long it took for some “key technologies” that we are using right now to actually get adopted widely?

• HTTPS, the guardian of your credit card information when you shop online? Started in 1994. Formally defined in 2000. Real adoption? Not until 2014 after Google said they would penalize non-HTTPS pages.

• A smartphone is undoubtedly “better” in the feature-space compared to a dumb phone. The first iPhone came out in 2007. Smartphones didn’t reach 80% market share until almost 7 years later.

• Google Maps came out in 2005. Most regions keep printing maps until 2012.

It takes a surprisingly long time for an “obvious” technology to catch on. For generative AI, we’re also looking at cost-benefit considerations, data privacy, and hallucinations. It will take time before companies become comfortable with the technology and integrate it into their day-to-day.

Bottom line: if your company is “behind” in AI adoption, you have a LOT more time than you think. Keep going at it. The improvement is real. But it will take work to gain from the technology.