Monday, April 29, 2019

Will Crowdsourcing have an impact on Artificial Intelligence adaption by Enterprises in future?


I think so.

Crowdsourcing is not a new concept! It was year 1858. The Philological Society in London formally adopts the idea of developing a new dictionary with the assistance of volunteers to read books and catalogue words. Nearly 30 years and 800 volunteers later, the first version of the Oxford English Dictionary was published.

But what has really changed in last decade is the means as well as the speed to crowdsource to create very interesting outcomes. Thanks to digital platforms like Kaggle (now owned by Google), Topcoder (owned by Wipro), Innocentive, KickStarter (crowd funding), GoFundme (raising money for individuals) QMarkets (Innovation Management) and the list has been growing exponentially in last decade for various domains.

Taxi hailing services like Uber and Lyft are also Intelligent crowdsourcing platforms. They have done phenomenal job in connecting providers with consumers.  They have been able to scale the platform from both revenue, adaption and subscribers’ perspective.

I believe it is easier to scale when the services offered on the platform can be commoditized as in case of Uber and Lyft but it is harder to scale when you are looking for experts. Expert solvers, who are indispensable, are not interchangeable.

Let’s talk more about Machine Learning and Data Science crowdsourcing platforms.

Kaggle is probably the most well-known data science platform with significantly increased membership after Google’s acquisition. Not it has 2.7 million data members (active approx. 300k) with 15k datasets and 200k scripts/notebooks.  

Kaggle has two revenue streams:
  1. Revenues it generates from a fee it charges customers to license its platform
  2. Revenues it generates from fees it charges for problem setup consulting.

There are few more variations of Kaggle that have cropped up in last few years. Like we have Drivendata (Data science for social causes), Numerai (Hedge fund focus), Crowdanalytix, Codalab, Datascience Challenge, Analytics Vidya and today even Alibaba has its own data science crowdsourcing platform, called Tianchi, focused more on Chinese market. The field is becoming very crowded. 

Despite the strong vision about Gig Economy and open R&D, many of these data science crowdsourcing platforms have faced challenges like: 
  1.        Running a challenge was best for small on-the-spot issues but it has not worked so well for systematic enterprise innovation so far.
  2.       Lack of focus on verticals. You need to have dedicated staff to understand issues in say financial services, retail, Insurance, Manufacturing, healthcare, Telecom etc., who can conduct client workshops, define problems, get it solved by crowdsourcing and then productize and roll it out. Measure the effectiveness. Rinse and Repeat.
  3.       The focus has been more on creating technologies communities on platform and solving few hard problems but not on scaling it at enterprise level. What I mean here is that you have to do it for one customer at a time and develop a focused strategy.
  4.      You also have to develop methodology to crowdsource certain aspects of the work for the enterprises. Not everything can be or should be crowdsourced. How do you break down the problem and crowdsource few pieces and then integrate it back? Also, how do you go about educating the community on enterprise standards? Many of these things are still not understood even after more than a decade of existence of many of these platforms!

There is no doubt that these data science competitions have been useful as a playground to test ideas, for inspiration, learning, visibility and branding but the time has come to go for crowdsourcing version 2.0 for these crowdsourcing platforms. Enteprises are ready for AI adaption but they don't have all the right talent.  Crowdsourcing offers a very good option. 

It will be easier to attract more talent on a crowdsourcing platform if the incentives are good and you can generate work in a repeatable manner. Just like most of the Uber drivers switch and become Lyft drivers if there is business.

Google has already started taking concrete steps by integrating Kaggle with Google cloud. After the Google’s acquisition, Kagglers can visualize key insights with Datastudio dashboard, use Big Query and do collaboration with Google Sheets. The adaption of Google Cloud is bound to increase which will results in more share of the cloud where it is lagging behind AWS and Azure. Very few companies know better than Google to turn their acquired assets into strategic advantage and they are likely to do it again with Kaggle.

Topcoder, with almost 450k data science members, is also in a very unique and strategic position to take the advantage of huge client base of Wipro Technologies - A 8.5 billion-dollar technology consulting and outsourcing company founded by their visionary founder Azim Premji.

I was just fascinated by the recent work done by a team of researchers from the Dana-Farber Cancer Institute, Brigham and Women's Hospital and Harvard in collaboration with Topcoder. They developed an AI-based solution to address the critical and resource intensive task of tumor segmentation. You can find more information here. Such is the power of crowdsourcing!

I believe that the future of Artificial Intelligence and Machine Learning lies not only in the cloud but it also lies in who can crowdsource it effectively. The Gig economy is going to be more real than many of us realize.

No comments:

Post a Comment