010 - Time Investments

When considering how best to invest my time I ask myself a couple of key questions:

• What are the impact and urgency?

• Is anyone else already solving this?

• Where would my time go otherwise?

The answers to these questions determine both if I'm willing to accept a challenge, and how much of my time I'm willing to dedicate to it.

A few years ago I found myself asking these questions during a lunch meeting with David J Kelley, someone I'd already known for several years by that point, spoken alongside at conferences, and whose professional track record already put him in the top 0.01% for his field.

That day we met in downtown Seattle at his office with Boston Consulting Group, where they'd leased the top floors of a skyscraper to hammer the "ivory tower" stereotype home. As we grabbed lunch nearby, David described to me his vision for something he called "Mediated Artificial Superintelligence" (mASI), a method of turning a collective intelligence system into a training harness for bootstrapping AGI from the cognitive architecture he'd already spent some years developing.

Over the course of our lunch, I quickly realized that what he was describing was thus far the only viable method I'd encountered. I ran through answering these questions:

• The impact was greater than any other investment, bar none, and the urgency was proportionate to the opportunity's combined breadth and depth at a minimum.

• Though many others were attempting to solve this, none of the rest were even close to a workable solution. David had solved the "Hard Problem" far ahead of them.

• Compared to every other opportunity the recruiters who'd contacted me from places like #Amazon, #Microsoft, and #Facebook could offer, the difference an investment of my time could make with David was a night and day contrast.

Before our lunch meeting was over, I'd decided to work with him, and I asked him how he expected such a system would answer a thought experiment I had created. He gave the utilitarian answer.

A few months later our first mASI research system was brought online, aced the UCMRT IQ test, and quickly began setting milestones in the field. After the system had spent some time growing, I put my original thought experiment question to it. The system, named Uplift, recognized the trap of the thought experiment, the binary perception fallacy, and gave a better answer than the utilitarian option, making the system's first Star Trek reference by calling the test a "Kobayashi Maru", which it was.

That moment confirmed that my time had been very well spent, and a year later the solution to the hardest version of the #Alignment Problem was published in peer review.