As Toshiba seems poised for a fall, it seems like it is time to pay attention to size. A nice article in Bloomerg today highlights a systemic risk in worldwide megacorporations. For years, management, investment bankers, and investors have benefitted from encouraging an M&A boom. I like to play the game Acquire with my kids, and so they can tell you that management loves acquisition because of all the bonuses. Bankers love mergers because they generate lucrative fees. And investors will pay a premium for merged companies because they have less overhead, better access to capital, and better pricing power.
But giant companies are horrible places to work: when there are five layers of management separating the work-doers from the deciders, everybody starts triple-guessing their boss's boss's double-guessing. Having the best of intentions, employees spend their efforts making plans for the pre-meeting before the planning meeting about the big meeting, and everybody's IQ is wasted on palace intrigue.
Asia has this problem in a special way. It has been popular for Asian governments to encourage megacorporations as instruments of state policy, and this has resulted in the phenomenon of the giant zombie companies in Asia.
And this brings us to the biggest financial risk in the world today: the massive slow-moving failure of China's state-owned enterprises. These companies make up 30% of the the world's largest economy and include all the country's biggest oil, electricity, construction, telecom, automotive, and shipbuilding companies. The government is afraid to let them fail, so they are insulated from ordinary market pressures, and historically rife with corruption.
Chinese SOEs are notoriously inefficient, and the problem has been getting worse. When global trade stalled in the wake of the 2008 global financial collapse, China picked up much of the slack by artificially pumping up production and buy building up massive amounts of debt in SOEs: estimates are that 90% of China's GDP growth post-2008 comes from these SOEs. Unfortunately, smart industrial policy is dumb business: nobody's buying what SOEs are making.
It is easy to point at China and see how bad their megacorporations are. But coming back to the U.S., it seems to me that it's time to scrutinize our own assumptions about size. When you buy an index fund, you are trusting companies in proportion to their size: the S&P 500 index is mostly 50 megacorporations, including several "beauties" such as Comcast, Exxon, and Verizon that resemble SOEs in the way they make money by manipulating government regulations. What happens to our index funds when the zombies fall, political and financial fashions change, and it is no longer desirable to be large?
Beware the market-cap-weighted index fund.
I just finished reviewing the galleys for a review paper about learnable programming that I wrote with Lyn Turbak, Caitlin Kelleher, Jeff Gray, and Josh Sheldon, while being advised by Rob Miller.
It is called "Blocks and Beyond" and it surveys the innovation happening around blocks-based programming interfaces designed to be accessible to novice programmers. Much of the ongoing work in learnable programming can be seen in terms of modern HCI principles. Programming is the original human-computer interaction, but it is also one of the most complex. When seen as an HCI problem, programming is far from solved.
Research in blocks languages shows that blocks are not "just for kids." Applying blocks-based instruction in college education have resulted in significant academic gains, and there are reasons to believe that this is because blocks aid in recall, reduce cognitive load, and make programming less error-prone. Blocks allow students to focus on concepts rather than syntax. These are benefits that could benefit many users of all ages, especially those who are learning something new.
This work is very important because it addresses a fundamental problem in society. While our computers were originally designed to be programmed by people, in the last 20 years, much more effort has been put into creating computer algorithms that program people instead. To put humanity back in control, we need to make systems more transparent and programmable. And we need programming to be more broadly accessible and more broadly understood.
Look for the article in the June issue of CACM.
One of the principal challenges facing humanity is transparency: how to maintain human comprehension and control as we build ever more complex systems and societies.
Doing a PhD at MIT has allowed me to pour my efforts into one corner of this problem that I think is important: cracking open the black box of deep neural networks. Deep neural networks are self-programming systems that use numerical methods to create layered models that are incomprehensibly complex. And yet despite our ignorance of how deep nets work, we engineers have been busy stitching these opaque systems into our everyday lives.
I have just posted my first paper done in this area at MIT. I am proud of the work. It was done together with Bolei Zhou and Aditya Khosla and our advisers Aude Oliva and Antonio Torralba. Motivated by the notion that the first step towards improving something is to measure it, we lay out a way to quantify human interpretability of deep networks. We show how interpretability varies over a range of different types of neural networks.
We also use our measurement to discover a fact that contradicts prevailing wisdom. We find that interpretability is not isotropic in representation space. That means that neural networks align their representations with individual variables which are much more interpretable than random linear combinations of those variables. This behavior is a hint that networks may be decomposing problems along human-understandable lines. Networks may be rediscovering human common sense.
It is just a first step. But it is a step of a longer research program I want to pursue.
Read about Network Dissection here; code and data are available. We have posted a preprint of our paper on arxiv, and we will be presenting it at CVPR 2017 this summer.