WiDS 2021: Data is the New Code
40-odd years ago, the software began to eat into the paper-and-ink economy that had ruled the business world for generations. Today that takeover is nearly complete. The world of commerce runs on digital data and software algorithms. But now a similar takeover is brewing. This time it is software that is getting eaten by AI. There will be a point in the future (we can argue about when, but it is inevitable) where software’s role is a small fraction of the contribution that AI makes to how work gets done.
Make no mistake: AI is not software. AI is trained and not programmed. AI isn’t governed by an inflexible algorithm. Its response to its environment and the input it receives from humans is not deterministic. The AI we release onto the world isn’t dependent on human programmers updating it; AI can grow and evolve without direct human modification. And yet the processes we use to build, deploy and monitor AI are based on our experience writing and deploying software. But these processes are doomed to fail. We can’t test, audit, control, or secure AI using the processes we use to do all those things for software. We must rethink everything we know about the development process.
Watch Daniela Braga discuss how DefinedCrowd approaches the process of developing AI beginning early in the data collection phase. Daniela will make the argument that we cannot safely build AI with the processes used to build software. To get to trustworthy AI, we must build it differently, because once it is released it will be very hard to control.