CEO DailyCFO DailyBroadsheetData SheetTerm Sheet

How Amazon is tackling the A.I. talent crunch

June 1, 2021, 7:38 PM UTC

This is the web version of Eye on A.I., Fortune’s weekly newsletter covering artificial intelligence and business. To get it delivered weekly to your in-box, sign up here.

Amazon, like other tech giants, is desperately hunting for workers who have an expertise in artificial intelligence. 

The online retailer has many businesses—its core e-commerce division, the Alexa voice-activated digital service, and the AWS cloud computing unit—that depend on machine learning. But there are relatively few computer scientists who know the technology, and those who do are in high demand.

One way Amazon has adapted to the tight labor market is to require potential new programming hires to take classes in machine learning, said Bratin Saha, a vice president and general manager of machine learning services at Amazon. The company’s executives believe they can teach these developers machine learning basics over a few weeks so that they can work on more cutting-edge projects after they’re hired.

It’s a strategy that many companies can emulate—and many have. Online education company Udacity, for instance, offers courses that companies can use to train managers in A.I. basics.

Some of Amazon’s coursework involves teaching developers Python, a programming language used widely by machine learning experts. The courses also teach rudimentary machine learning concepts including statistical regression methods that are used for tasks like predicting product prices over time. Another area of focus is deep learning, in which researchers train neural networks—or software that learns—to automatically translate languages.

These are not formal college courses, and Saha said the recruits aren’t graded like they would be in school. Instead, the courses are intended to give new developers a foundation in machine learning and statistics so they can understand the theoretical underpinnings.

If coders don’t understand how machine learning works, they are less likely to create usable A.I. products or troubleshoot related problems.

Eventually, to show they understand machine learning concepts, Saha said new coding recruits are asked to create a recommendation system or a forecasting model. He didn’t say what happens if they fail the challenge, but he said that Amazon has “a pretty good interview process,” implying that the company weeds out underachievers early in the interview process—before they take the classes.

As for other companies recruiting A.I. talent, Saha said he recommends managers tell prospective candidates will have a big impact on the company’s products. Making prospective hires feel valued at work instead of unnoticed is key.

And despite the shortage of A.I. talent, Saha said that college students seem to be increasingly enrolling in A.I. courses. So there is some hope that companies will eventually have a bigger pool of A.I. recruits to choose from.

Jonathan Vanian 


Big data goes private. KKR and Clayton Dubilier & Rice will buy data analytics firm Cloudera and take the company private, Bloomberg News reported. Cloudera was once a leader in the so-called big data movement, driven in part by the open-source data crunching Hadoop technology. But the rise of cloud computing and rival data analytics services from Amazon Web Services and Microsoft proved difficult for Cloudera to compete against.

Killer drones. Turkey deployed a drone that attacked Libya’s military, making it the first time an autonomous drone was used for warfare, according to a report from the UN Security Council’s Panel of Experts. From the UN report: “The lethal autonomous weapons systems were programmed to attack targets without requiring data connectivity between the operator and the munition: in effect, a true ‘fire, forget and find’ capability.”

Meet the new A.I. supercomputer. Nvidia and the National Energy Research Scientific Computing Center has debuted Perlmutter, which is said to be the fastest supercomputer specifically tailored for A.I.-related tasks to study climate and astrophysics. In a blog post about the supercomputer, Nvidia said traditional supercomputers “can barely handle the math” required for some computationally heavy tasks like generating “simulations of a few atoms over a few nanoseconds.” Scientists are now using machine learning to create more accurate and complicated simulations, which require specialized A.I. chips, like the kinds of GPUs, or graphics processing units, that Nvidia sells. The new Perlmutter supercomputer uses Nvidia GPUs, specifically tailored for deep learning.

A new A.I. company hits the market. The A.I. startup Anthropic has raised $124 million and plans to create “generally applicable AI technology” that can be used in many different industries, The Financial Times reported. The startup is led by Dario Amodei, a former head of safety at the high-profile A.I. research firm OpenAI. Several OpenAI researchers who worked on the firm’s popular GPT-3 language model joined Anthropic. Former Google chief Eric Schmidt, and Facebook co-founder Dustin Moskovitz also invested in the company, the report said.


Former United States chief technology officer Michael Kratsios has joined the startup Scale AI as managing director and head of strategy, The Wall Street Journal reported. Scale AI, which specializes labelling data used to train A.I. systems, recently raised $325 million and is privately valued at $7.3 billion.  

Splunk hired Shawn Bice to be the data monitoring firm’s president of products and technology. Bice was previously the vice president of databases for Amazon Web Services and a general manager at Microsoft.


Let’s all admit that A.I. is hard. A Santa Fe Institute researcher published a paper supported by the National Science Foundation about common misconceptions people have about A.I. and why the field has boom times known as A.I. springs and downtimes known as A.I. winters. The author Melanie Mitchell, a  Santa Fe Institute professor, explains that people underestimate the difficulty of modeling human intelligence through the prism of computer science.

From the paper: Even with today’s seemingly fast pace of AI breakthroughs, the development of long-promised technologies such as self-driving cars, housekeeping robots, and conversational companions has turned out to be much harder than many people expected.

Mitchell notes that enthusiasm over machine learning’s ability to do specific tasks, like OpenAI’s GPT-3 language system that can automatically generate text when given prompts, leads people to mistakenly believe that researchers are about to create general A.I., the holy grail of computing in which A.I. systems are as capable as humans on multiple tasks. “The philosopher Hubert Dreyfus (using a term coined by Yehoshua Bar-Hillel) called this a “first-step fallacy,” the researcher wrote.


Germany, birthplace of the automobile, just gave the green light to robotaxis—By Christiaan Hetzner

European privacy activists launch international assault on Clearview AI’s facial recognition service—By David Meyer

Can A.I. help Hollywood dub The Rock into another language? This startup thinks so—By—Eric J. Lyman

Insurance firm Lemonade backtracks on claims it uses A.I. to scan customer faces for hints of fraud—By Jonathan Vanian

JD Logistics soars in Hong Kong debut, extending the empire of ‘China’s Amazon’—By  Yvonne Lau

A new way to close Asia’s digital skills gap—By Mona Mourshed and Michael Fung


Deep learning comes to the guitar world. Keith Bloemer, a guitar effects pedal creator, has created a new effects pedal that uses neural networks—the software used for deep learning—to simulate the unique sounds of an overdriven tube amp, the gadget site Tom’s Hardware reported. Guitar enthusiasts tend to love the sounds of vintage tube amps and typically scoff at effects pedals that attempt to digitally simulate the hard-rocking sounds of old-school amps.

Bloemer’s new pedal, based on a Raspberry Pi computer circuit, can allegedly create more realistic and distorted sounds than more conventional digital pedals, because the pedal’s software learns to emulate the actual sounds of guitar amps. Researchers at Queen Mary University of London have previously published an academic paper about using neural networks to create more realistic digital effects.

Bloemer said via his GitHub software coding page that users will be able feed the pedal recordings of other “real amps and pedals” so that the pedal learns to emulate their sounds.


Our mission to make business better is fueled by readers like you. To enjoy unlimited access to our journalism, subscribe today.