Previous Submission

Machine learning and Tencent

Next Submission

Data analytics has changed the world in many ways and it is becoming more important in the digital era. However, with the increased size of data available, human race starts to produce more mistakes and incorporates more biases. Machine learning became a very important topic that will reshape the world going forward in many practices ranging from spotting Parkinson syndrome [1] to city brain [2]. Machine learning is especially important for the internet giant because Tencent is an innovation company, its valuation, and capability to generate profit will depend heavily on the next step of innovation rather than improving the current operating efficiency. However, other players are also on the move. For example, Google’s Deepmind has developed AlphaGo which became very famous because the AI is able to beat the world champions[3]. In China market, other tech giants also pushing towards machine learning. Two notable players are Alibaba and Baidu[4].

 

There are two main problems towards the development of machine learning, the first problem would be the adoption of the technology and second would be the talent pool for development. On the first issue, Tencent has invested broadly in the industries ranging from healthcare [1] into entertainment and gaming []. Inspired by Deepmind’s AlphaGo, Tencent is developing its own AI to play Starcraft, a very popular strategy game developed by Blizzard. This might seem irrelevant to the market, but it is a very strong example to spearhead the development of AI and machine learning. The second issue on hiring and talent development, Tencent did put a vast resource into attracting and securing talents to enhance ensuring the company technological prowess in terms of machine learning and artificial intelligence by opening dedicated labs to develop the technology [6]. However, it is also acknowledged by the company that the biggest bottleneck for machine learning technology development is the human resource and there is a massive gap between demand and supply[7]. MIT tried to address this issue by launching a new school focusing on computing data and AI [8]. This is a very good progress for machine learning but still very US-focused initiatives and will not make a significant change in 5 years. Fortunately, Tencent also focuses on training and develop the talent by themselves through a various project in their pipelines [9].

 

Another massive bottleneck for machine learning technology would be the availability of labeled data to train the model, Tencent should start early on labeling the data they have through their eco-system. For example, Joox platform and the music recommendation. Currently the algorithm is matching song recommendations through similar genre but with machine learning in the game, they can move one step further to analyze the beats and pitch of the sound that customer tends to prefer and offer recommendations accordingly. Wechat will also play a big role in data collections, the app offers a wide range of services ranging from chat-app to matchmaking. While other competitors also have a strong platform for data collection, they do not have the same coverage as Tencent. This will allow them to have a competitive edge over the competitors on machine learning and AI development. The second recommendation I have for the management is to carefully analyze trends in the international market. One of the biggest barriers of Chinese mega corporation expansion is cultural and language. This can be a hindrance on the development and progress of the technology. However, to successfully progress on machine learning, Tencent should adopt a global mindset to serve global customers and not only greater China area. Globalization will be a painful but necessary process and the company is in process of making the necessary changes.

 

The remaining questions I have for my classmate is that, is there really any predictive capability from machine learning? because all the data we used to train are from past behaviors and it can be argued that with only past data, machine learning will only project the past into the future without real value-add. Another question would be that how many hobbies will machine learning destroyed. Some people no longer have interests on Go or Starcraft because you cannot be better than AI. Personally, I believed there are more unanswered questions going forward that can really use more ideas.

[1] https://www.forbes.com/sites/samshead/2018/10/08/tencent-aims-to-train-ai-to-spot-parkinsons-in-3-minutes/#259353e86f36

[2] https://www.cbinsights.com/research/china-baidu-alibaba-tencent-artificial-intelligence-dominance/

[3] https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/

[4] https://www.forbes.com/sites/bernardmarr/2018/06/04/artificial-intelligence-ai-in-china-the-amazing-ways-tencent-is-driving-its-adoption/#11a6537a479a

[5] https://www.technologyreview.com/the-download/612188/tencents-ai-programs-defeat-starcrafts-own-ai/

[6] https://techcrunch.com/2017/03/27/tencent-ai/

[7] https://www.theverge.com/2017/12/5/16737224/global-ai-talent-shortfall-tencent-report

[8] https://www.technologyreview.com/the-download/612293/mit-has-just-announced-a-1-billion-plan-to-create-a-new-college-for-ai/

[9] https://www.bloomberg.com/news/features/2017-06-28/tencent-rules-china-the-problem-is-the-rest-of-the-world

6 thoughts on “Machine learning and Tencent

  1. Thank you for the perspectives you are presenting. One your second question, I do think those games will still be relevant and interesting to humans, as we will soon simply accept the fact that we are worse in them than machines and the competition is between each other. Probably the same thing happened at some point with other skills like running, shooting, calculation, memory etc.

  2. To your point on the education gap, I wonder if Tencent can help solve this by partnering with local Chinese universities and research facilities that are focused on Machine Learning. With regards to innovation, I am also struggling to understand how Tencent can use machine learning to drive large leaps in innovation vs. just marginal efficiency and customer experience improvements for its existing and future tools.

  3. Thank you for interesting opinion on machine learning, especially on gaming industry. To your first question, I think still historical/past data is one of the most important sources for prediction on future. More serious issue is, as you mentioned, lack of valuable primary data to be used as the source of prediction. In most cases, machine learning technology needs something exceptional or failed cases on a certain situation, but most of data are focused on normal cases. Tencent should invest on gathering data with more diverse process and results.

  4. It’s interesting to think that machine learning could potentially destroy hobbies because human players cannot beat AI. Hobbies are important for people to re-energize from their day-to-day challenges and so many people enjoy playing online games. However, I believe AI’s performance in games will not discourage most fans from playing, because there is a socialization component to playing games. Many people play to interact with their physical-world or online friends, and they are not discourage by the fact that there are players much better than they are (whether those players are real-people or in the future, AI players). I would be more concerned about game companies using AI not focused on performance, but focused on sociability, to entertain users. I wondered whether AI applied to entertain users may bring they far from their “real-people” friends, potentially having bad psychological effects.

  5. great post! I agree that people need to be careful about assuming too many benefits of machine learning, as it can only ever be as good as the data it is given

  6. Great post on ML application in social media space. To your question, there are predictive capability from machine learning by detecting the pattern and training historical datasets. Through the automatic training algorithm, the modeling all are going through the optimization process overtime, which cannot be replaced by human being. The massive info consumed, analyzed and optimized in parallel are the true beauties of machine learning.

Leave a comment