The best of both worlds: Human-machine complementarity

Making a good decision implies that we have some idea of what’s true. As Jill Lepore, professor of history at Harvard University, explores in her podcast series The Last Archive, the “unit of evidence” for discovering truth is data that is now only readable by a machine. The sheer scale of data available to us in the modern world is overwhelming. Humans are impressively efficient thinkers, especially given our limited resources. But we do not have infinite data inputs or processing capacity. We are limited by our lifetimes. We have somewhat unreliable connections to others. It can take us a long time to learn since we can only think in a few dimensions at a time.  

Making decisions in the age of machines is hard because our brains are not logic machines. Evolution doesn’t care about reason and logic, it only cares about what works. What works is staying alive long enough to have babies and raise children. What makes human intelligence fascinating is all the various ways that we have evolved to be one of the most prolific and dangerous species on the planet. Some of our success is due to logic and reason, but these are fragile components of our social existence.

Machines, by comparison, are logic machines. Machines also have almost unlimited capacity to gather, process, and share data in real time. Machines can think in almost infinite dimensions. As AI and biological sensors enable machines to subject more of our social and biological behavior to computation, we are increasingly competing with an intelligence that’s bigger, faster, and more precise than ours.

Humans and machines are also highly complementary in how we decide to classify certain information, such as images from medical diagnostic tests. Humans are very specific, with pathologists almost never mistakenly misidentifying cells as cancerous when they are not, while machines are impressive at their ability to spot even very tiny tumors.

When a doctor uses AI as a mind for her mind, she makes use of AI’s scale, speed and scope to find the slightest evidence of trouble while she makes the best treatment decision. The machine happily sorts through thousands of images looking for every lesion while the doctor decides which lesions are harmful. The machine might know that lesions are correlated with smoking but it’s the doctor who figures out why smoking causes cancer. Maybe her insight came from a hunch, an intuition founded in her expertise and experience. She is able to share the idea with the world using language. As fallible as our communication may be, knowledge transfer between humans is second nature.

Big data and AI now add to the complexity of using our intuition. AI uses training data to look for correlations that can then be used to make predictions. AI doesn’t necessarily need a human to guide it with a theory. Nor does it need a human to add any intuitions.

Powerful AI will find correlations that humans can’t, such as when a customer is likely to churn or how someone’s social media profile is predictive of their credit worthiness. Which is why it’s valuable and used by the most powerful companies in the world. But machines can’t always combine the correlative results. Meaningful combination is the domain of humans.