Skip to main content

What Artificial Intelligence Can and Can’t Do Right Now

https://hbr.org/2016/11/what-artificial-intelligence-can-and-cant-do-right-now


Many executives ask me what artificial intelligence can do. They want to know how it will disrupt their industry and how they can use it to reinvent their own companies. But lately the media has sometimes painted an unrealistic picture of the powers of AI. (Perhaps soon it will take over the world!) AI is already transforming web search, advertising, e-commerce, finance, logistics, media, and more. As the founding lead of the Google Brain team, former director of the Stanford Artificial Intelligence Laboratory, and now overall lead of Baidu’s AI team of some 1,200 people, I’ve been privileged to nurture many of the world’s leading AI groups and have built many AI products that are used by hundreds of millions of people. Having seen AI’s impact, I can say: AI will transform many industries. But it’s not magic. To understand the implications for your business, let’s cut through the hype and see what AI really is doing today.
Surprisingly, despite AI’s breadth of impact, the types of it being deployed are still extremely limited. Almost all of AI’s recent progress is through one type, in which some input data (A) is used to quickly generate some simple response (B). For example:
W161026_NG_WHATMACHINEv2
Being able to input A and output B will transform many industries. The technical term for building this  A→B software is supervised learning. A→B is far from the sentient robots that science fiction has promised us. Human intelligence also does much more than A→B. These A→B systems have been improving rapidly, and the best ones today are built with a technology called deep learning or deep neural networks, which were loosely inspired by the brain. But these systems still fall far short of science fiction. Many researchers are exploring other forms of AI, some of which have proved useful in limited contexts; there may well be a breakthrough that makes higher levels of intelligence possible, but there is still no clear path yet to this goal.
Today’s supervised learning software has an Achilles’ heel: It requires a huge amount of data. You need to show the system a lot of examples of both A and B. For instance, building a photo tagger requires anywhere from tens to hundreds of thousands of pictures (A) as well as labels or tags telling you if there are people in them (B). Building a speech recognition system requires tens of thousands of hours of audio (A) together with the transcripts (B).
So what can A→B do? Here’s one rule of thumb that speaks to its disruptiveness:
If a typical person can do a mental task with less than one second of thought, we can probably automate it using AI either now or in the near future.

A lot of valuable work currently done by humans — examining security video to detect suspicious behaviors, deciding if a car is about to hit a pedestrian, finding and eliminating abusive online posts — can be done in less than one second. These tasks are ripe for automation. However, they often fit into a larger context or business process; figuring out these linkages to the rest of your business is also important.
AI work requires carefully choosing A and B and providing the necessary data to help the AI figure out the A→B relationship. Choosing A and B creatively has already revolutionized many industries. It is poised to revolutionize many more.
After understanding what AI can and can’t do, the next step for executives is incorporating it into their strategies. That means understanding where value is created and what’s hard to copy. The AI community is remarkably open, with most top researchers publishing and sharing ideas and even open-source code. In this world of open source, the scarce resources are therefore:
  • Data. Among leading AI teams, many can likely replicate others’ software in, at most, 1–2 years. But it is exceedingly difficult to get access to someone else’s data. Thus data, rather than software, is the defensible barrier for many businesses.
  • Talent. Simply downloading and “applying” open-source software to your data won’t work. AI needs to be customized to your business context and data. This is why there is currently a war for the scarce AI talent that can do this work.
Much has been written about AI’s potential to reflect both the best and the worst of humanity. For example, we have seen AI providing conversation and comfort to the lonely; we have also seen AI engaging in racial discrimination. Yet the biggest harm that AI is likely to do to individuals in the short term is job displacement, as the amount of work we can automate with AI is vastly bigger than before. As leaders, it is incumbent on all of us to make sure we are building a world in which every individual has an opportunity to thrive. Understanding what AI can do and how it fits into your strategy is the beginning, not the end, of that process.

Andrew Ng is VP & Chief Scientist of Baidu, Co-Chairman and Co-Founder of Coursera, and an Adjunct Professor at Stanford University.

Comments

Popular posts from this blog

The Difference Between LEGO MINDSTORMS EV3 Home Edition (#31313) and LEGO MINDSTORMS Education EV3 (#45544)

http://robotsquare.com/2013/11/25/difference-between-ev3-home-edition-and-education-ev3/ This article covers the difference between the LEGO MINDSTORMS EV3 Home Edition and LEGO MINDSTORMS Education EV3 products. Other articles in the ‘difference between’ series: * The difference and compatibility between EV3 and NXT ( link ) * The difference between NXT Home Edition and NXT Education products ( link ) One robotics platform, two targets The LEGO MINDSTORMS EV3 robotics platform has been developed for two different target audiences. We have home users (children and hobbyists) and educational users (students and teachers). LEGO has designed a base set for each group, as well as several add on sets. There isn’t a clear line between home users and educational users, though. It’s fine to use the Education set at home, and it’s fine to use the Home Edition set at school. This article aims to clarify the differences between the two product lines so you can decide which

Let’s ban PowerPoint in lectures – it makes students more stupid and professors more boring

https://theconversation.com/lets-ban-powerpoint-in-lectures-it-makes-students-more-stupid-and-professors-more-boring-36183 Reading bullet points off a screen doesn't teach anyone anything. Author Bent Meier Sørensen Professor in Philosophy and Business at Copenhagen Business School Disclosure Statement Bent Meier Sørensen does not work for, consult to, own shares in or receive funding from any company or organisation that would benefit from this article, and has no relevant affiliations. The Conversation is funded by CSIRO, Melbourne, Monash, RMIT, UTS, UWA, ACU, ANU, ASB, Baker IDI, Canberra, CDU, Curtin, Deakin, ECU, Flinders, Griffith, the Harry Perkins Institute, JCU, La Trobe, Massey, Murdoch, Newcastle, UQ, QUT, SAHMRI, Swinburne, Sydney, UNDA, UNE, UniSA, UNSW, USC, USQ, UTAS, UWS, VU and Wollongong.

Logic Analyzer with STM32 Boards

https://sysprogs.com/w/how-we-turned-8-popular-stm32-boards-into-powerful-logic-analyzers/ How We Turned 8 Popular STM32 Boards into Powerful Logic Analyzers March 23, 2017 Ivan Shcherbakov The idea of making a “soft logic analyzer” that will run on top of popular prototyping boards has been crossing my mind since we first got acquainted with the STM32 Discovery and Nucleo boards. The STM32 GPIO is blazingly fast and the built-in DMA controller looks powerful enough to handle high bandwidths. So having that in mind, we spent several months perfecting both software and firmware side and here is what we got in the end. Capturing the signals The main challenge when using a microcontroller like STM32 as a core of a logic analyzer is dealing with sampling irregularities. Unlike FPGA-based analyzers, the microcontroller has to share the same resources to load instructions from memory, read/write the program state and capture the external inputs from the G