October 22nd, 2014

The Story of Dark Blue Labs, and Google

It was only a month and a bit ago that linkedin suggested I congratulate my colleague Nando de Freitas on his new job: co-founder at Dark Blue Labs. This is exciting - I thought: Nando is a well known figure in machine learning, and one of the last few remaining high profile academics that did not already work for Google, Facebook, Amazon or Baidu. I was very keen to learn what he was up to.

I had a meeting with the co-founders and talked about their story so far. Dark Blue Labs was the stereotypical ex-academic stealth AI startup, their website looked (and still looks) very much like that of DeepMind before they got acquired by Google. This is what the website says about the company:

Learning deep structured and unstructured representations of data to make intelligent products, including natural language understanding, a reality.

If you are like me, and find this too vague, you can take a look at the founders' recent publications. The group were working on cutting-edge deep learning-based technology, which promises to understand natural language much better than existing approaches. Way better than the largely rule-based systems behind Siri or Google Voice Search.

The team were really just starting as a company and the founders still had their academic positions. According to DueDil, the company was incorporated in March this year. But when we met in September, they already were considering an informal acquisition offer from "a high-profile tech company in the US". It was kind of obvious they meant Google.

Last week, before we had an opportunity follow up with another meeting, the team sent us an update, saying they had indeed sold the company to Google. The news has just been made public on FT today. Today, linkedin suggested I congratulate Nando once again: this time for his new title as Senior Staff Research Scientist at Google.

Of course I'm happy for these guys. They have worked hard in their academic careers, and it's nice to see them being rewarded for their talent. I would like to use this post to congratulate the team of four for their unusual success.

But deep inside I am also a bit disappointed by the outcome.

Firstly, I believe the technology behind DBL is very powerful and would have loved to see a smart team figure out how to change the world with it. I don't think this high-impact innovation is going to come from inside Google. Why? Because they have enough applications that deep learning can improve and they are not under pressure to find another hard problem to solve. Google is one of the most innovative established tech companies out there, but they will rarely look beyond consumer technology. Why would google ever apply AI to law? Or waste management? Or to solve problems for the public sector?

Furthermore, I would argue that the presence of Google and other aggressive aquirors in the space creates a very unhealthy environment for startups. As a founder of a startup packed with top machine learning talent you don't really have to worry about product. You don't have to worry about a vision either. You always have the easy option of selling your team to Google, the question almost becomes when and for how much. Premature acquisitions are becoming the norm in this space. In other words,

Google increases the opportunity cost of innovation in the AI space

This is further reminder of how Google are monopolising machine learning. It has already been very hard to find top machine learning talent for startups. The available talent pool is largely owned by Google already. What's more, it appears today they announced not only one, but two deep learning acquisitions out of Oxford.

I can only hope that in a few years time we will see these then-ex-Googlers released to the market, who will eventually find awesome applications for their expertise. Once again, congrats to Dark Blue Labs and Vision Factory for their quick success stories.