If AI progress stopped right now, I still think it's enough to change the world in a profound way over the next 5 - 10 years.
It does not in any way look like progress is going to stop. I just wanted to describe why I think that's the case and how you get from "ok cool this is a bit better than Google search for some stuff"...to completely changing the world. That seems like quite a jump.
It's exciting, to me anyway, because it maps out a clear path to solving the most difficult problems in the world. There are a bunch of problems in life that we mostly accept as intractable: ageing, disease, poverty, environmental damage. But if you think about it, these things are all intelligence, labour or energy constrained...and it can actually be reduced down to just "intelligence constrained".
AI is fundamentally about transforming electricity into intelligence.
Let's take environmental damage as the first 'intractable human problem'. Is there any fundamental technical reason why you can't have a workforce of robots designing, mining, manufacturing and deploying solar PV to completely displace fossil fuels in our energy systems?
A short time ago it was unclear we could embed enough adaptable intelligence into robots to automate real-world work at scale. You cannot solve that with brittle "if x then y" logic. The missing piece has not primarily been:
1. the technology or raw materials to convert sunlight to electricity (photovoltaic effect was discovered in 1839 by Edmond Becquerel, and has been improving since)
2. that the energy arriving at Earth from the sun was insufficient (even conservative calculations show enough incoming solar resource globally; practical constraints are capture, storage, transmission, land use, and build-out speed)
3. robotics technology (we've had 'good enough' stepper motors, servos, solenoids, batteries, cameras and other sensors for quite a while)
The missing piece is increasingly accessible machine intelligence. That does not mean everything is solved. Robotics integration, safety, power electronics, permitting, supply chains, and operations are still hard. But the constraint has shifted enough that full-stack automation looks plausible in areas that previously felt out of reach.
What to do about it all?
If anyone is interested in my opinion: understand the edge of capabilities of the best tools now. Use them hard enough to find the failure modes. Then track the rate of improvement. That is where most strategic opportunities will come from over the next few years.