Hacker News Reader

Comment by glangdale | Hacker News Reader

I have often found it baffling that tech giants, with near unlimited resources to fund all sorts of things, don't establish these kinds of labs. I'm aware that there are many things called something like "{Google, Facebook, Twitter, ...} Labs" but from what I understand they are significantly different from the corporate research labs of old.

View on HN

Google Research is a thing. It’s even cited in the article as a positive example.

https://research.google/

Microsoft Research is a notable exception

Also Mozilla Research

Why spend so much on product research when you can simply wait for some startup to create a new product and bring it to market while you can just acquire them or create a clone and drive them out of business abusing your dominant position.

I'm not sure what "corporate research labs of old" did differently from what Google Research, Microsoft Research, or FAIR do. They are all significant investments in personnel and equipment by these firms, and produce fundamental research and are arguably as good as the best universities in the world at it (if not better at some things owing to sheer scale). A lot of MSR's work isn't even directly applicable, people work on theoretical computer science and optimization theory while being permanent researchers there. Similar cases exist across the industry.

The Google model is better though, in my view. The problem with the older style "lab" approach, typified by Microsoft Research, is that it basically recreates academia inside a company. There's then often huge difficulties integrating resource output into actual products. MSR does OK but is still quite notorious for producing long term expensive research projects that never get integrated into any product (Singularity/Midori being perhaps the most famous), whereas the Google model doesn't have that problem. Google has an extraordinary high rate of incorporation of research ideas into products.

This is better for a lot of reasons, obviously, it makes it easier to justify funding the projects. But it also ensures that the researchers truly understand the nature of the problem they're solving. Seeing actual translation queries on Google Translate helps direct the next piece of MT research towards what kind of things people really use the tech for. Building actual big data systems helps train the researchers in how to work with huge data sets and puts them on research paths they otherwise would have been unable to explore.

Basically it recognises that a hard division between research and practice doesn't really make sense when working at the cutting edge of technology. You want the constant feedback from real world practice to guide the theory.

"Fail fast" is not conducive to good research. It takes persistence in the face of failure.

Facebook Research makes real strides in many areas.

Disclaimer: I work at FB, but my employment there does not change my opinion. I worked at Uber and my opinion of Uber's "research" was different ;)

Aren't Google's "other bets" basically just huge research labs?

Exactly. Calico Labs for example is under Alphabet and devoted to researching aging and longevity:

https://www.calicolabs.com/

Why would that be a better arrangement than, say, universities?

The main issue with corporate research is that it's not easy for them to capture the output; it very quickly leaks into the public domain, despite patents. So why bother?

This is one thing that I really have to credit MS for doing right.

Reply on news.ycombinator.com