This machine learning report is required reading
Maybe it's the nerdy thing you do when lounging with family this holiday season, or something you take in during a long walk or travel, but it's worth a look since it's one of very few attempts to collate a comprehensive look at the amalgamation that is the AI industry. See last year's newsletter on the annual report for a recap.
It doesn't hurt that leaders from the most advanced organizations in this space, including OpenAI, MIT, and SRI International, played a role in putting it together.
Some major takeaways worth considering:
– Strides in performance progress continue for benchmarks like GLUE for natural language understanding as well as improvements in the AI2 Reasoning Challenge to answer multiple-choice questions like a grade-school child.
– Growth in published papers in China has been driven in part by government-affiliated authors, whose work saw a 400 percent increase in 2017. Corporate AI papers saw a 73 percent increase. Conversely, the United States saw its biggest increase in published AI papers from corporate tech giants like Google, Nvidia, and Microsoft.
As the Index reports, Europe leads the world in total number of research papers produced, followed closely by China. Within less than five years, China could lead the world in total number of papers published, according to an Elsevier report released this week.
– AI is a global industry, with 83 percent of papers on Scopus published outside the United States
– Annual AI conferences NeurIPS (formerly NIPS), ICML, and CVPR saw thousands of attendees each.
– U.S. continues to lead in AI-related patents, and AI startup funding is up 4.5 times, compared to 2 times for other sectors receiving venture capital investment.
– More than half of Partnership on AI members are nonprofits now, like the ACLU and the United Nations Development Programme.
One of my favorite stats by far in this year's report, however, is the total number of mentions of AI and machine learning in earnings calls by companies listed on the New York Stock Exchange. It's a metric that points to how businesses are changing the way they talk about artificial intelligence.
It's true there are still companies selling magic beans and snake oil out there, but empty claims aren't enough anymore.
Earlier this week ahead of the release of the AI Transformation Playbook, I spoke with Andrew Ng. The former Baidu AI chief scientist and Google Brain cofounder said that as he was encouraged and a bit surprised that the irrational AI hype around AGI and killer robots did not seem as prevalent as it has been in the past. Understanding of what AI can and cannot do could help reduce these fears.
There may still be a fair deal of startups and businesses who want to call themselves AI companies now and sprinkle it all over the place to justify their value. But increasingly, it's not enough to call yourself an AI company — you've got to prove it, and demonstrate why that AI creates a virtuous cycle for your company to create a competitive advantage.
It's not entirely surprising there are more mentions of AI in earnings calls, as more companies are in fact looking to use AI. Tata Consulting reported this week that 46 percent of organizations have implemented some form of AI, but implement is not the same as successfully implemented, and smart companies aren't just talking about AI, they're looking for ways to successfully spread it throughout their organizations.
As time goes on and the luster of the first round of AI hype seems to go away, calling yourself an AI-first business doesn't seem to be enough anymore.
The smartest businesses seem to be building with trust, rapidly shifting consumer sentiment, and the value of diverse employees and perspectives in mind when building systems for intelligent machines.