Opinions Technology

Pokemon Go and Google’s inception-ism

Folks, it is the Pokemon Go season ūüôā are you excited?

Can someone tell me where does the Pokèmon Go?

Perhaps not?!¬†Ok, tell me one thing honestly. Do you get strange¬†looks from your¬†millennial friends when they find out you¬†haven’t played Pokemon Go or that you don’t know how to use Snapchat?¬†

Don’t play Pokemon Go? good news, you are not alone.

I will give you in on one more secret. I have never watched the “Game of Thrones”. For whatever its worth, I don’t feel like watching it either.

In the hyperconnected ADHD world, virulence is an everyday thing. New fads¬†catch the masses and they spread like wildfire. Profound¬†technology¬†waves however, rarely do so. They start relatively unannounced. Grow like strong forces of nature, and are very hard to stop. The growth is rarely abrupt. Think –¬†iPhone, social, e-commerce etc. the tweet below is a nice reminder.

Artificial intelligence, I believe is one such profound technology wave.

It is growing like a strong force of nature devouring information, getting smarter and better everyday. We are still in its early days but are beginning to see some really really cool applications. 

A London based company, and fully owned by Google is called DeepMind. We recently¬†learnt that it improved the power usage efficiency / PUE of Google’s data centres by 15%. Can you imagine how much money that saves for the company that runs one of the largest data centre ops on the planet? Not too long ago, DeepMind AI¬†became the first to beat a world champion at the infamously difficult game Alpha Go. These are not small developments. On a much larger scale they signify a tectonic shift in the maturity of machine learning. DeepMind algorithms use convoluted learning, more specifically reinforcement learning. In simple words, they learn without specific programming given a goal, they “try to figure the best way” to solve a problem.

These relatively specific developments when multiplied by the opportunity scale promises nothing short of a revolution. 

Abhi Shah

This one¬†I believe is going to be an “Inside-out” one.

Take an industry e.g. aircraft manufacturing, apply deep learning and you have a 15% or even better efficiency in outcomes of a process for instance predictability in ordering of wing parts. Algorithms will let us learn causality like never before by letting these neural networks figure out hidden patterns. There is a good blog post by Google about machine vision visualisation; click the link in the caption below. 


How a machine sees a painting (credit Google Research)

We are not too far away from an applications of computer vision that once seemed far fetched, for instance looking at your CCTV, your computer might tell you who is on the door before you find out! (Edit: Already solved). Ultimately more exciting and probably scary applications will come too.¬†I believe that will take some time, and to my earlier point, highly potent AI has just made an unassuming entrance, now just wait and watch as it turns into an unstoppable force. I will leave you with an image created by convolutional neural networks – this sort of gives us a flavour of how computers “see” or some would say “dream” things. Credit Google Research.

Only if Google’s AI could teach me¬†how to use Pokemon Go or Snapchat.¬†Nope, I like being stubborn, I don’t think so.

Abhi Shah

So the difference between Pokemon Go and Google’s inception-ism is that one seems like a fad, and the other seems like a unstoppable tech wave, let’s see where they both lead us.¬†


Chatbot | Are Bots without chat the future?

Computers or your phones aren’t actually all that smart, not yet. A¬†few¬†real world machine learning applications have become commonplace though. For instance, your photos app¬†now recognises that you were¬†with a dog at the¬†park. Another example is your AI based keyboard. I vouch for SwiftKey. It uses machine learning to vastly improve auto-completion, and next word prediction. So much so, that¬†many times it¬†completes literally whole sentences while¬†typing. Now, that’s pretty great and machine learning has come a long way.

So whats next? we love asking this question, and rightly so!

Different tech companies have slightly different bets to answer this question. However, there seem to be some underlying themes. For example, chatbot has begun gaining traction. Especially since Facebook launched their chatbot service.

Chatbot search trend
Notice the spike in Google Trends for Bots (in RED) in April 2016 when Facebook announced chat bots for developers

Google followed with Duo and Allo, and we know that WeChat has been doing chatbots for a while too. Conversational UI, assistant, chat based commerce and all that is apparently the “next big thing”. But is it? What does experience tell us? ¬†well as Ben Evans nicely puts it in his blog. see original blog post here.

Is Chatbot just a bunch of “if-else-then statements”?

From my experience of using a Facebook messenger chatbot, it would appear so, it is NOT intelligent. Period. Perhaps one day, it will get there. Oh yes, and there is the Uber integration. Have you heard of a newly launched platform that does not have a sexy means to call you an Uber? Alexa, call me an Uber, Google Home Call me an Uber, Facebook bot, yada yada… Enough with Uber already. I quite like the concept of chatbot, don’t get me wrong but as Ben Evan’s blog nicely highlights, the magic dissipates as soon as the algorithm starts asking you too many questions. In the short term though, the fact that my Google photos can figure out my Christo Redeemer photos without me labelling them is definitely magic. It happens without me having to chat with anyone, or without providing any significant user input.

Christo Redeemer
Google magically finds my Christo Redeemer photos!

So, lets recap. Machine learning is great, but its not so great that it can converse with us without asking too many questions and create magic, just yet. Edit: GPT class might change this completely. See GPT2, and GPT3.

I believe the real magic¬†is somewhere else and we should not get too distracted by the user interface such as chat. We should rather take a real world problem, let’s say “making a shopping¬†list” and apply machine learning till the algorithm¬†matches or surpasses human abilities at solving for the task. For instance, looking through your regular shopping¬†lists and when you want stuff, can the algorithm automatically predict what you need every other week? That would be awesome.

Let us understand the first principle on this one.

The path of least resistance always wins. Images are better than text. Video is better than images. Automated curation of responses than asking the user to type in a full sentence. Chat requires too much user input and my bet will be on applications that make the input invisible real fast. Imagine that our machine learning shopping list app just gently notifies the user his or her auto-populated shopping list. Yes, you can then send it straight to Amazon fresh and order stuff at the push of a button. Simple, right? I know!!

Appreciate that it is a very hard problem to solve Рbut in my book it would definitely be magic. I am not the only one to say this, a lot of Silicon Valley pundits say that the next wave of startups would take a problem, and add AI to make things simpler. Thats what I find exciting about the future!

I am happy to be proven wrong, but hopefully we can do away with chatbots until they can truly become magical.


Ps. for any VC’s willing to fund my shopping list idea, do reach out at @abhinandanshah or the comments section ūüôā

For now, bot out…