By Michel André
Expontential growth is everywhere around us and has been for a while, but our human brain is not equipped with the cognition or power to predict or analyze consequences of exponential growth in any way shape or form. Some exponential facts powering the world and disruption happening around us today….
- Humanity produces 2.2 exabytes (2,300 million gigabytes) of data every day; 90% of all the world’s data has been created in the last 24 months
- iPhone 7 -> 120x faster than iPhone in less than 10 yrs, and as fast as worlds fastest supercomputer 1996 (Hitachi SR2201/1024)
- In 36 months, the training speeds for some types of AI/ML workloads have boosted by 60X (yes sixty times) comparing the Nvidia K40 GPU to the P100 GPU with a modern SDK applied.
Its very hard to fathom or predict what this will and have given rise to in terms of technology and evolution, I am not even sure everyone understands that the compute power we carry around in our pockets around 20 years ago was analog to that times fastest and most expensive computer.
And the thing with exponential growth is that it is absolutely the most blazingly fast in the end by its very nature as can be seen by the animated illustration (click on image) below its very sudden and the trajectory is there and hasnt been broken for a very long time.
So there is a chance that we right now are standing in front of a veritable wall of development and there are more and more signs around us that this is happening in our time. And its like the little guy on the drawing below from the waitbutwhy.com article series on artifical intelligence.
Wether the fact that compute capacity will become indefinetly abundant and cheap leads to superintelligence or general artificial intelligence surpassing humans I will let be unsaid here and there are good treatises on this subject on both sides of the fence and it might be a risk worth considering as a human race. I think Kevins Kelly discussion of this is an interesting counterweight and food for thought on the superintelligence explosion discussed in other places (linked here Myth of superhuman AI). This still doesnt remove the fact of exponential growth and a lot of the interesting problems we can solve with technology going ahead in different areas and potentially a lot of “high value” knowledge intensive work that will change or move to computers.
Also the area of AI has seen and is seeing exponential growth – just a brief history of AI will show u that.
- Shannon’s Computer Chess (1949) – Minimax Algorithm
- Deep Blue vs Garry Kasparov (1997) – Custom HW/SW
- IBM Watson Wins Jeopardy (2011) – Curated Data + Q&A
- AlphaGo vs Lee Sedol(2016) – Deep Learning + GPU
Advancements come quicker and quicker and we can also see a progression from rules based, custom built hw/sw to more generic/general algorithms and hardware to solve the challenges presented. Around 2009/2010 something interesting happened with the invention of deep learning, essentially the compute power allowed us to make multilayered neural networks to solve a new heap of problems, that can be seen below, and its from that era we see the AlphaGo victory.
Also with the introduction of deeplearning we can see that in some cognitive areas the computer actually surpasses humans and thats when it comes to error rates in image recognition and speech recognition as can been see from the two graphs below. This is also why we see all the siris/alexas/cortanas and other assistants showing up now at rapid pace, and the image recognition will also drive things like self driving/semi autonomous cars.
I think an interesting application of image recognition is shown in the yolo algorithmand as you can see in the linked video below it can do quite accurate object detection at framerates of 40-90 fps on a “standard” TitanX gpu card that sits in high end workstations. And I guess probably in your teenagers gaming computer there sits a Geforce 1080 card which gets the speed down to measily 20-35 fps on commodity hardware available for anyone and given exponential growth it might soon be able to fit your pocket which is quite impressive.
So the question is why is all this happening now and at the relative pace its happening, well once again exponential growth in a lot of areas bringning us to this nexus. Mainly growth in
- Abundance of data, storage and capability to process data
- Computing power becoming more and more powerful e.g GPU
- Cloud services making a lot of the above commoditized and easy to use and invest in
- Research in these areas leading to new improved powerful statistical algorithms and methods to utilize/categorize and make predictions on data given data
Another interesting aspect of this is that a lot of the knowledge, technologies, math and infrastructure is out there ready for your taking to lab with and get going, it doesnt require massive investments in hardware, software or anything to join the revolution it requires some standard commodity computers and open source software to get going in the space of machine learning and data science. It has never been so open and out there as it is now, from things like Hadoop/Spark, mllib, jupyter, python and related libraries, tensor flow, cognitive toolkit by Microsoft. I think this might be unpresedented as well in the development this very open access to both knowledge, technology, software and hardware. And a google, Microsoft and amazon is also helping companies by putting AI/ML as service offerings in a simplified way out there in their different cloud infrastructures.
This all leads to that we see a new breed of companies that is transforming into or is leading this revolution, Sundar Pichai recently said google will be an “ai first” company, and the likes of amazon, netflix, uber is following suit and this happens in the large and the small driving innovation. And we can also use our imagination to comprehend what happens behind closed doors of governments and secretive companies like palantir which in itself is a bit of a scary name with products like gotham and metropolis – you do the math :). And I venture this development will be very quick and a lot of “old” companies not being able to capitalize and optimize using this new breed of technology and knowledge will be passed in different ways shapes and forms.
So yes exponential growth, disruption and “ai” is the new normal and here to stay! So you better connect read up and start exploring the space and thanks for bearing with me so far!