One of the world's most famous computer scientists reveals his 'playbook' for bringing AI to every business
- Dr. Andrew Ng, one of the most famous computer scientists in the world, revealed his "playbook" for helping all businesses adopt AI.
- Ng is best-known for his time at Google and his stint as chief scientist at Chinese tech giant Baidu, though he's also affiliated with Stanford and cofounded Coursera.
- He says the best thing to do is to start with a small project before you even think about trying to form a grand strategy.
- He also says that a lot of companies make a crucial mistake - they hoard lots and lots of data, to be used in AI systems. But that data is often useless, Ng says, as it's impossible to separate signal from noise.
Dr. Andrew Ng has one of the most enviable resumes that a computer scientist could work for.
He's probably best-known in America for his time at Google, where he cofounded the Google Brain artificial intelligence project with famed engineer Jeff Dean. From there, he served as chief scientist at Chinese tech giant Baidu, where he grew its AI organization into the thousands. For a time, too, he was director of Stanford's AI Lab. Oh, and he cofounded online learning company Coursera.
Nowadays, Ng is the chief executive at Landing AI, a consultancy that helps the largest businesses pursue their own strategies for artificial intelligence, which stands to revolutionize everything from warehouse labor, to retail, to writing your own e-mails. Indeed, he says that the AI trend to watch nowadays is the adoption of the technology by non-tech companies, from fashion brands to factories to farms.
To that end, Ng released last week his "AI Transformation Playbook," a five-step plan for businesses to follow if they want to dip their toes into the water, all drawn from his experiences at Google and Baidu. We spoke to Ng earlier in December to get his perspective on the playbook, and his further advice for companies chasing AI.
First thing's first, Ng says, is not to worry so much about the hand-wringing in Silicon Valley about Facebook, Google, and other giants hoovering up all the AI talent, calling those concerns "overhyped." If you can pay well, he suggests, and you're working on interesting projects, you won't have any trouble finding AI engineers.
And besides, he says, there are lots of engineers out there doing work that might be considered AI (or, at least, AI-adjacent), insofar as they're parsing large amounts of data to help the system make smarter recommendations. Whatever they don't know, he says, they often pick up on the fly as the project demands.
"All education is self-education, because what's the alternative?" quips Ng.
"I see CEOs go big too often"
Once you're ready to get going, Ng says that it's a common mistake to start by forming a broad strategy. This is tempting, but it's usually a mistake, he says: You don't know what your company is capable of, AI-wise, let alone what it's good at. Rather than rush into timetables and grand designs you later have to alter, start small, he says.
The really crucial first step, he says, is picking a project that's not so big that you get discouraged, but also, not so small that "even if you succeed, no one cares."
"I see CEOs go big too often, then I see them go too small," Ng says. "Try to do something you can get done in a year."
In that sense, it's usually good to start with something core to the business - starting with something behind-the-scenes like HR or payroll could be doable, but it also might be hard to get the team jazzed about it, he says. This is also good, he says, because it forces you to think about how AI can actually be put into use in your business, rather than an abstract cure-all.
"AI doesn't magically solve your problems," says Ng.
From there, it's time to invest in people. For large companies, he says, it's worth the time and effort to formally train your people, even as you hire outside experts. The major cloud platforms, like Amazon Web Services and Microsoft Azure, offer a smorgasbord of AI-powered services for developers. Still, Ng says, those are pretty generalized tools; every company faces their own problems, and you want your own team building your own solutions.
"They're just fine, but you need to build on top of them," says Ng. "You want to be good at AI, you need your own team," he later added.
Don't be a hoarder
Then, once your team is on board and fully trained up on your own particular problems, then you can think about your AI strategy - a way to apply what you've learned to actually make a difference.
Finally, Ng says that there's one big mistake that he sees lots of companies make. Over the last several years, companies have been hoarding all of their data, from all parts of their business, with the expectation that it will some how, in some way, form the cornerstone of an AI or data analysis strategy.
Not so, says Ng, who routinely sees clients sitting on vast warehouses of irrelevant or otherwise useless data, where the signal is indistinguishable from the noise. Ultimately, Ng says that when you're training an AI model, it's better to have a few hundred high-quality data points than it is to try to use all of that extant data.
"To this day, I don't think their engineers know what to do with this supposedly valuable data," says Ng.