Caterina,

So what do you think of the state of the art in automated face detection and face recognition? Has it gotten good enough yet to be usable commercially? Or is it still something that isn't quite there yet...

Dave

Funny, I was thinking about exactly that tech as I was writing that post. I generally think the kinds of tech I mention here get you 75% of the way towards solving the problem, but the last 25% has to be done by humans. You'd never sign a legal document translated by OCR or by the various online language translators.

Similarly you wouldn't trust face-recognition tech for national security. You could use it to, say, have a lifeguard kick a kid out of the adult swim, but nothing more serious. So, not there yet.

So inspiring post!I was just passing by this site and came across this post.Its really interesting and the ideas are so innovative.You have done superb job.

Hi Caterina, interesting post.

I agree that many of the innovations touted by some startups are at best incremental, and that they frequently apply to areas that are very, very, hard (eg machine understanding in general).

That said, I think there is a lot of scope for what I would call "modest" technologies, which aim to add value using existing tech in a novel way. Hunch is a good example.

Yes, I'd hope that Hunch was a good example. Thanks David!

Totally agree with finding and honing in on the right balance... I find it very difficult to gauge that proper balance by catering to those who want to enjoy new "tech" but have to be handheld every step of the way and then those that are bothered by such hand-holding measures... Like the brain, it seems most only use 10% or less of the available features of a startups innovative technology:-)

Thanks for the post Caterina. I appreciate your observations.

Your comments on face recognition relate to a conversation a friend and I were having at lunch yesterday.

There are times when you are watching a movie or a TV show and want to know who an actor is. It would be great to be able to take a picture of the actor and find out who it is -- kind of like a photo-driven IMDB.

Does this capability exist yet? If not, maybe you or one of the people who read this post can get started. It's at least worth a $1.99 iPhone App for me ;-)

Couldn't agree more. We tend to think in algorithmic solutions because that's what we're used to. Using the Crowd as a global computer is so new a concept that startups still think that getting acknowledged requires something with super-advanced math.

I've been in the field of digital audio watermarking for years. That's pretty cutting-edge. But in the end nobody wanted our product. Least did the end users who's content it was supposed to be tracing.

So yeah, innovation doesn't necessarily mean rocket science. It's more like putting simple parts together in unusual ways.

Dan
@DanielStocker

I think there's a gap between the academic researchers and industry, where relatively few researchers write stuff that scales or release usable code. Being the first team to have write a usable implementation of some fancy algorithm from academia can be a significant boost, especially given that most of the people that read academic papers are going to be other researchers, not competing start-ups.

I think a distinction needs to be made between science and engineering here. I'd argue that startups shouldn't bother researching new science for many of the same reasons you mention here, plus the fact that researching new science can be extremely risky and better suited to universities and labs with the proper capital risk profile.

Engineering is taking the advancements in math and science and applying them to building a product. It's something that can only be done in industry and its great to aim for "hard" engineering problems that have a high payoff if you have the technical strength. Solving "hard" engineering problems is a huge barrier to entry because it can require understanding very complex science (and combinations of science that have never been attempted) to a level where you can use it to solve real problems.

Yeah, there's also a bunch of entrepreneurs who take university research and convert them into businesses -- Eric Paley, my confrere from Founder Collective took some MIT research and built it into Brontes Technologies -- dental technology. Another avenue.

I think you're spot on here. I've been in the technology world a long time and I've heard a lot of claims of "revolutionary innovations." For the most part, they're not revolutionary and, of the part that remains, few are feasible/defensible business propositions... meaning they're not really innovations.

My own personal example: I worked with a startup that was going to use web crawling and fancy Bayesian AI to create market analyst research reports. It failed for numerous reasons, not least of which was that you could hire humans to mine the same data... and humans are more trainable and cost fewer dollars in HW and ISP charges than a fancy Bayesian AI machine!

Great post. One comment, I have entered a few tech heavy industries like Speech Recognition and have been successful by using human horse power to overcome the initial need for crazy tech. Doing this gives you speed to market, ability to get real market feedback and breathing room to build your tech over time.

I agree with this post wholeheartedly. At Tellme we built a >$100 million/year business doing speech recognition, but we licensed all the tech off the shelf. Sure, we did a lot of things to make it better, but for the core recognizer, we licensed it from folks like Nuance or IBM who had massive R&D budgets and had sunk huge costs into building the core tech. The funny thing is, the core tech was not very usable out of the box -- all the art in making that stuff actually work in a real application was where most of our value was added.

A related item is how often "pure tech" startups end up becoming small (from a shareholder value standpoint) companies. If they don't flame out, they get gobbled up by a big Microsoft / Apple /etc. type of company for unimpressive amounts. Examples include Fingerworks, acquired by Apple to do things with touch, and Seadragon, acquired by Microsoft (and their tech is pretty amazing, one must admit). Neither was a large outcome from a shareholder standpoint, but both have technology that has the potential to impact millions of consumers in profound ways and create large commercial opportunities, if it can be harnessed properly by the new corporate parents. However, neither would have been a large successful company (from a revenue standpoint) on their own.

Caterina,

Very interesting post. I am curious specifically what you think of innovation in speech recognition and transcription technology. It seems every other week I read some article about how adoption is expected to rapidly increase in hospitals saving doctors time and hospitals money on transcription costs. This technology has been around for a long time though and adoption still seems very limited. This seems consistent with your comment that innovation needs to be an order of magnitude better as opposed to small improvements in accuracy before the technology will be accepted for widespread use. Do you think that sort of innovation is likely / possible given where the technology is today? Any thoughts?