Every Thursday, we choose a podcast about AI and summarize it into something you can read in about 10 minutes or less. This week features Jack Dorsey, CEO of Twitter and Square, who spoke with MIT research scientist Lex Fridman for his AI Podcast. [Note: Questions and answers were edited for clarity.]
Question: Seventy years ago, Alan Turing developed the Turing Test to measure the intelligence of a machine compared to humans. How hard do you think is it to pass the test in the space of natural language?
Dorsey: From a practical standpoint, for now and years out, AI and ML models can bubble up interestingness very quickly. You can pair that with human discretion around severity, depth, nuance, and meaning. For Dorsey, the chasm we face to general intelligence is to explain why and the meaning behind a decision or sets of data. One of the biggest risks of AI is building a lot of black boxes that can't explain why, or the criteria they used to make a decision. We're trusting AIs more, from lending to content recommendation, driving, and health. It's a hard problem since sometimes humans can't even explain why we make decisions.
Question: Do you think we'll ever be able to build a system like what was in the movie "Her" (aka, a deeper connection with AI)?
Dorsey: Hasn't that already happened? he asks. It's less a function of the AI, and more a function of the individual, and how and why they find meaning, he argues. Dorsey doesn't think it's a negative, but it's constantly going to evolve. Meaning is something that's entirely subjective. He doesn't think it's going to be a function of finding the magic algorithm that enables everyone to love.
Question: Is it possible to tell the difference between a bot and a human? Can we have fulfilling relationships with each?
Dorsey: It's useful in certain problem domains to be able to tell the difference, but in others, it might not be, Dorsey says. The technology to create is moving much faster than the technology to detect. It's a race, so the detection has to be two or even ten steps ahead of the creation. This is a problem the financial industry will face more, as a lot of its risk models are built around identity...