Uncanny Examples of AI applications


We want to highlight a few fun and unexpected outcomes of AI technology. The first is a textbook example of how to manipulate public opinion, get celebrities to talk about ICOs and stocks publicly, and power propaganda bots. Just fake a video! In this example, actor Jordan Peele provides a vocal track of his impersonation of Barack Obama, saying various outlandish things. A deep fake AI does the rest.

What does that actually mean? First, hours upon hours of footage of Obama were fed into a machine learning algorithm, which is taught to recognize how sound data correlates with visual data. Next, the process is reversed to generate images instead, with the original algorithm acting as a gatekeeper to assess whether the generated image is good enough to pass for Obama. And with a bit of magic dust for interpolation, we have the uncanny outcome where we can make anyone say anything that we want.

Another odd example is this -- a dog was wired up with various sensors that used machine vision to its daily activity. From this data, researchers were able to isolate the dog's ability to choose on which surfaces to walk. This involves quite a bit of judgment, understanding whether the path is too high or uncomfortable. After this experiment, an AI has a statistical layer that represents a dog's pathfinding in the world based on visual data. Perhaps we'll see this making its way into a Boston dynamics death critter.

How do we translate this back to financial services? The short term answer is this -- anything that humans do in a rote manner, where the task is a result of human intuition of reasoning but has a fairly stable decision process, can be done by machine learning. Full stop.