"We are very far from what is possible in AI": Fei-Fei Li
 |
| Google Cloud chief scientist and Stanford prof Fei-Fei Li at a TED talk |
Image: Google
The
field of artificial intelligence (AI) has evolved from academic research to one
of the biggest drivers of business today, said Fei-Fei Li, chief scientist of
Google Cloud and an associate professor in the computer science department at
Stanford University. So big, in fact, that she called AI “one of the biggest
technologies of the twenty-first century”, while acknowledging her possible
“bias as an AI technologist”.
Li
should know: one of the top AI experts in the world, especially in the AI
sub-domain of computer vision, she and a few other researchers at Stanford are
credited with giving the trigger to the “big explosion of AI today.” (http://bit.ly/2uwyb8R) Among other things, she and her
associates worked on ImageNet, a huge dataset that now contains over 13 million
images of animals, people, products and other ‘objects’ all over the world. The
collective effort put up by AI researchers in developing and improving deep
learning algorithms over the years has recently made it possible for an
AI-powered system to correctly identify and label images with an accuracies of
97% and above--better than the human eye. (Deep learning is a specialty in AI
which is based loosely on how neural networks in the brain work.)
“But
we are very, very far from all that is possible,” said Li when asked about the
problems that still need to be cracked in computer vision. While machines have
used the brute computing power and large datasets to solve the issue of “naming
objects”, she said that AI would be no match to someone getting up early in the
morning and trying to find their car keys, for instance. Describing the properties
of objects and figuring out their relationships with each other are still “very
difficult” for AI to handle. Elaborating it further, she said that while AI
systems have to be trained thousands of times on the same data to “learn something”,
children can learn to do that same task easily with just one or two
demonstrations—for instance, looking at a tool, they would know where to hold
it and can grab it properly whereas AI-based robots would struggle for long.
Nevertheless,
Li believes that AI is putting pressure on a lot of repetitive tasks and
impacting “the landscape of labour”—something that is only going to accelerate
in the next few years as AI capabilities and access to AI tools improve.
“The
smart CEOs should be thinking about AI and its impact on their respective
business,” she said. They should also be looking at how AI can be adopted or
adapted for their industries.
Li
was speaking to mediapersons in India through a webcast at Google offices in
Delhi, Mumbai and Bengaluru.
She
likened the current and future job losses feared from wider applications of AI
across industries to similar technological advancements in the past. For
instance, she said that while a tool such as spreadsheets made a lot of book
keepers redundant (while making them available for other analytical and
accounting work), it was “curious” that the human tellers in banks continued to
grow in number despite automated teller machines (ATMs). Li said that there has
always been interplay between technology and the labour market and it needs the
“whole society” to ponder over the question of job shifts in an AI-powered
world. “As for me, I’m an optimist,” she said.
Google—like
its peers such as International Business Machines Corp. (IBM) and Microsoft
Corp.—has of late been opening up application programming interfaces (APIs)
that allow coders to use the AI capabilities of the company for developing
commercial applications in healthcare, travel, entertainment and media, and
several other industry segments.
Li
shared a few examples of how the AI interfaces are being used in the real
world. For instance, Shazam, a UK-based music service with over 40 million
songs in its database uses AI for “audio fingerprinting” which enables it to
update its index of popular songs every hour instead of once a day earlier.
Some Indian examples include publisher DB Corp, which uses Google Cloud for
real-time publishing and uses the speech APIs to capture and document interviews;
real estate portal NoBroker.in, which uses Google Vision API to identify rooms
and figure out the condition of properties listed on the site; and software
solution provider Searce, which has developed an intelligent software called HappierHR
for converting paper-based processes into paperless ones.
While
the AI technology is improving, Li believes that more and more companies and
people need to have access to the AI tools and datasets to be able to get their
own projects off the ground—something that Google is evangelizing through what
is often termed “the democratization of AI”. Essentially, it involves opening
up datasets and machine learning libraries (such as Google’s TensorFlow) to
more programmers so that AI-based applications can be built and implemented as
quickly as businesses need.
(Note: an edited version of this post first appeared on Livemint.com.)
No comments:
Post a Comment