Database Reference
In-Depth Information
Submitting yourself to the scrutiny of the scientific community is a much bet-
ter way of evaluating the quality of a piece of work than internal evaluation. It's
very difficult for a small number of people who may not be specialists to criti-
cally evaluate a piece of work within a company. This type of evaluation essen-
tially leads to internal hype. You see this in some types of industries that tend
to be more secretive, such as the defense industry. There are a lot of things
going on there that are not so great but, because they can't publish, they really
don't know. So it's useful for a company to have its scientists actually publish
what they do. It keeps them honest. It's also very good for the prestige of the
company. Publishing allows them to attract high-quality people, who generally
want to brag about what they do, and need to talk about what they do if they
want to be an integral part of the research community. So that's the first type
of particular industry impact—intellectually strong impact.
The second type of impact is focused on research projects that are deemed
important internally.These types of projects are ones where we say, “We know
we want to work on this project.” We don't know what ultimate application it
will have, but we know it's important. A bunch of people will work on it and,
if you have an impact on it, internally we can evaluate whether that impact is
important.
The third type of impact, of course, is impact on things that are deployed.
So if you work on something that ends up being deployed or you improve
something that's already deployed in some measurable way, then that's a very
direct way of measuring impact. There are organizations that claim to be
research labs, but they only measure the third kind of impact and so they're not
really research labs. They're just focusing on the short-term opportunities.
Gutierrez:
How do you get ideas for things to study or analyze?
LeCun:
That's actually a very important question that determines how a
research lab should be organized. There are several kinds of research scien-
tists. As a first step, you need people with vision. These are research scientists
who have a long-term vision. They may or may not be that good at actually
implementing it. They might associate with other people who do the imple-
mentation. Then there are people who are good at keeping their eyes on a
long-term goal that will have a long-term impact and are good at ignoring
fashions. Then there are people who are excellent problem solvers, who may
not necessarily have the long-term vision, but they do have an ability to solve
complex problems that other people just can't. And finally, you have people
who can actually implement things and get them to work. In a research lab,
you need all of those people.
Yes, you need all those people, but it is very essential that the people in
the management of the research have vision, which means that they have to
be respected in their field. The idea that somehow you can put a bunch of
research scientists together and then put some random manager who's not
