Crystal Nights (2008)

More caviar? Daniel Cliff gestured at the serving dish and the cover irised from opaque to transparent. It s fresh, I promise you. My chef had it flown in from Iran this morning. No thank you. Julie Dehghani touched a napkin to her lips then laid it on her plate with a gesture of finality. The dining room overlooked the Golden Gate Bridge, and most people Daniel invited here were content to spend an hour or two simply enjoying the view, but he could see that she was growing impatient with his small talk. Daniel said, I d like to show you something. He led her into the adjoining conference room. On the table was a wireless keyboard; the wall screen showed a Linux command line interface. Take a seat, he suggested. Julie complied. If this is some kind of audition, you might have warned me, she said. Not at all, Daniel replied. I m not going to ask you to jump through any hoops. I d just like you to tell me what you think of this machine s performance. She frowned slightly, but she was willing to play along. She ran some standard benchmarks. Daniel saw her squinting at the screen, one hand almost reaching up to where a desktop display would be, so she could double-check the number of digits in the FLOPS rating by counting them off with one finger. There were a lot more than she d been expecting, but she wasn t seeing double. That s extraordinary, she said. Is this whole building packed with networked processors, with only the penthouse for humans? Daniel said, You tell me. Is it a cluster? Hmm. So much for not making her jump through hoops, but it wasn t really much of a challenge. She ran some different benchmarks, based on algorithms that were provably impossible to parallelise; however smart the compiler was, the steps these programs required would have to be carried out strictly in sequence. Julie said, All right, it s a single processor. Now you ve got my attention. Where is it? There was a charcoal-grey module, five centimetres square and five millimetres thick, plugged into an inset docking bay. Julie examined it, but it bore no manufacturer s logo or other identifying marks. This connects to the processor? she asked. You re joking. She tugged it free of the dock, and the wall screen went blank. She held it up and turned it around, though Daniel wasn t sure what she was looking for. Somewhere to slip in a screwdriver and take the thing apart, probably. He said, If you break it, you own it, so I hope you ve got a few hundred spare. Her face flushed. Of course. If it was a few hundred grand, everyone would have one. She put it down on the table, then as an afterthought slid it a little further from the edge. As I said, you ve got my attention. Daniel smiled. I m sorry about the theatrics. No, this deserved the build-up. What is it, exactly? A single, three-dimensional photonic crystal. No electronics to slow it down; every last component is optical. The architecture was nanofabricated with a method that I d prefer not to describe in detail. Fair enough. She thought for a while. I take it you don t expect me to buy one. My research budget for the next thousand years would barely cover it. In your present position. But you re not joined to the university at the hip. Julie couldn t help herself; she picked up the crystal and examined it again, as if there might yet be some feature that a human eye could discern. Can you give me a job description? I believe you re the best AI researcher of your generation, he said. I want you to work for me. He reached over and took the crystal from her. With this as your platform, imagine what you could do. Julie said, What exactly would you want me to do? For the last fifteen years, Daniel said, you ve stated that the ultimate goal of your research is to create conscious, human-level, artificial intelligence. Then we want the same thing. What I want is for you to succeed. She ran a hand over her face; whatever else she was thinking, there was no denying that she was tempted. It s gratifying that you have so much confidence in my abilities, she said. But we need to be clear about some things. This prototype is amazing, and if you ever get the production costs down I m sure it will have some extraordinary applications. It would eat up climate forecasting, lattice QCD, astrophysical modelling, proteomics ... Of course. Actually, Daniel had no intention of marketing the device. He d bought out the inventor of the fabrication process with his own private funds; there were no other shareholders or directors to dictate his use of the technology. But AI, Julie said, is different. We re in a maze, not a highway; there s nowhere that speed alone can take us. However many exaflops I have to play with, they won t spontaneously combust into consciousness. I m not being held back by the university s computers; I have access to SHARCNET anytime I need it. I m being held back by my own lack of insight into the problems I m addressing. Daniel said, A maze is not a

Crystal Nights (2008)

Key takeaways

More in computer-science