3/21/19

Head in the clouds, feet on the ground

The problem with AI is that it’s too big. AI has the potential to change everything, but it’s not quite there yet. There’s no general purpose AI, what we have now is very good, specific use case machine learning. Getting those models deployed to to actual work is hugely difficult.


It’s like NASA. The space agency has created inventions and advancements that have made their way into everyday life, but rickets are built by hand by people with PHDs. And making commercial products viable needs a bit of tuning, automation, repeatability. Essentially the ground-breaking work done at the cutting edge of science needs to filter into daily life.


Cutting edge science has it’s head in the clouds, what’s possible? What can be? How can we push the boundaries of what’s known and how we understand the worlds around us. But product is grounded in reality. One foot on the earth and the other reaching for the next thing.


Actually deploying AI in practical products requires a bit of work, thought, tooling. And the way to get there is to start at the bottom, start at common-sense, real-world applications and workflows. Start at real people doing work and see how their work can be improved by applying a smart machine to the task. And then tune for efficiency, tune for repeatability. The best AI will be hybrid AI systems where a human provides common sense, motivation, inspiration, and AI does some heavy lifting of information processing and organization.


Who knows maybe this model will become the inspiration for general purpose AI. Because we’ve learned that if a human can do it, a machine can learn it. What if we begin measuring inspiration, creativity, motivation and training a module to work in that space too? Maybe someday the cutting edge of AI will look like the efforts we’re pursuing today as we try to make AI useful in the world.


Maybe this is the secret we’ve been seeking all along, to get our heads out of the clouds for a minute and focus on the concrete, the real, the immediately applicable and useful. But even if it’s not the future of General Purpose AI, it’s certainly the short term future of productizing AI.


Subscribe to our newsletter