The experts are divided on AI’s future. For now, that’s good
On Tuesday, I attended TED AI, a special edition of the TED conference devoted entirely to the topic that is presently consuming most of the tech world’s attention: the promise and pitfalls of artificial intelligence. It was held in San Francisco’s stately Herbst Theatre—but emotionally, it felt more like a roller-coaster ride.
In just the first hour or so, a procession of AI experts took the stage, mostly presenting in the classic TED format: a single person explaining something compelling in dramatic fashion. First up: Google Brain cofounder Andrew Ng, who addressed widespread concerns that AI could have catastrophic implications for humanity. “I’m here to tell you that anxiety is misplaced,” he said, pointing out that electricity also has its risks, but we humans have learned to manage them. What a relief!
And then, three speakers later, MIT professor and Future of Life Institute founder Max Tegmark said that he’d warned, in a 2018 TED talk, of the dangers of superintelligent computers. But his expectations had been wrong, he admitted: “It went even worse than I thought.” How alarming!
With so many manifestly intelligent and well-informed people assessing the situation in starkly different terms, TED AI was anything but a fount of consensus and clarity. Ultimately, the most unassailable takes came from the speakers who took pains not to have a take, such as DeepMind cofounder Shane Legg. “I don’t believe the people who are sure it’s going to go very well, and I don’t believe the people who are sure it’s going to go very, very badly,” he said.
In a way, I found the lack of agreement to be a relief. The one prediction about AI I’m willing to make is that an awful lot we can’t foresee is going to happen very quickly. I’d be worried if the experts were uniformly too blithe or reflexively bleak. And understanding that we don’t know precisely where the technology will lead is the first step toward figuring out what to do about it.
Ilya Sutskever, OpenAI’s chief scientist, ended his TED AI talk by saying he expects that the potential for AI to go badly wrong would result in “people [starting] to work in unprecedented collaborative ways out of their own self-interest.” I can’t quite tell whether that sentiment is utopian or dystopian. For one thing, with all due respect to Sutskever, people have never been all that great at pooling their interests even when the stakes are extraordinarily high. For another, I hope that we never have to confront the possibility of any of the truly cataclysmic prognostications turning into real problems. But I’m heartened to see early conversations happening, including the sometimes clashing perspectives at TED AI.
Meanwhile, AI’s here and now provides plenty to talk about, too. Some of TED AI’s best moments involved current research efforts, such as Genentech EVP Aviv Regev’s look at how machine learning is transforming drug development. I won’t try to convince you to stop worrying about what could be ahead—some of it, I worry about myself—but any analysis that fails to consider the ways AI is already a boon to humanity is incomplete.
More stuff about annoying printers
On September 20, I wrote about my frustrations with my Epson inkjet printer—and the separate set of problems I was having with the HP OfficeJet Pro I bought to replace it. I invited you to share your printer stories, and got scads of email, mostly about your own many and varied hassles with an array of different printer models.
I feel your pain. Did I mention that my new HP is finally printing reliably—but other features that once worked, like scanning directly to Dropbox, have conked out? Anyhow, a sampling of your feedback:
Kelly Fitzgerald: You hit the nail on the printhead! I bought an Epson XP-410 because my sister’s has always worked fine for her, but for my 3x/month print habit, it’s a disaster. I might get half of what should be printed actually appearing on the page. I have to do a “nozzle check” when this happens, which brings my ink levels down at lease a third. Due to this scam, I get the cheaper off-market cartridges that the printer complains about when I install them, but I just keep pressing buttons until it relents.
Scott Johnson: I was in the same place this month as you. I needed a new printer. HP and Epson printers can brick after getting old. I have been in the IT area since 1985, so I have owned many printers over the years. After many hours of research, I just purchased a Canon Maxify GX4020. It fits the bill for color and mono printing.
Louise Caulfield: My Canon Pixma G 6020 seemed like a revolution when I bought it two years ago. But alas, at the equivalent age of a toddler, it’s lost its usefulness. The issue is the printheads: clogging, and a disheartening lack of Canon replacement parts. (Surprised?) Aftermarket parts are out there, with questionable consumer reports about quality. This time, I’m trying an Epson, with an extended warranty package.
Rich Brinker: After years of inkjet printer breakdowns, replacements, and the otherworldly prices and inconveniences of inks, I decided to go laser and have never looked back. I bought my first Brother HL-5250DN in 2007. It only prints in black. After having the original toner cartridge for about 10 years of near-constant use, the image started to fade a bit. I was resigned to replacing the cartridge and bought a new one. I saw online someone mentioned about taking the cartridge out and giving it a shake for some additional usage. Five years or so later, it finally did give out. Now you tell me—exactly how many inkjet refills and printer replacements would I have gone through in all that time?
Peter Altschuler: If you need heart surgery, do you book time with an ophthalmologist? If you have an EV, do you take it in for a tune-up? No. It’s the same with peripherals. You have to make an investment in gear that does one thing well. I have a dedicated flat scanner, high-speed document scanner (both Epson), black-and-white laser printer (Canon), large-format inkjet printer (Canon), color laser printer (HP), and copier (Brother). If one dies, it doesn’t affect the performance of the others. If a multifunction goes out, it takes its other functions down with it.
(24)
