Last Updated: March 11, 2016
Bayes’s Rule states that your understanding of the world should be updated every time you receive new data about it. That’s what I’m doing here. This is the set of what I believe right now, given what I know right now. I update these as I learn more.
Automation
Will robots steal our jobs?
Yes, and they will create new ones. The new ones will be in the industries that grow due to automation: science, creativity, entrepreneurship, and innovation-based industries. The real question is: over the long run, will automation create jobs faster than it replaces them?
A better way to think about this question is is: what parts of what jobs will be automated? In very few cases will a “robot” entirely replace a job. We should think more about a job’s tasks and skills and what proportion of those are completely automatable.
Jan 2018 Update: The jury is out of the net effect of job (creation - destruction). Very little work is going into the creation effect. More good research should be done, especially with a focus on what governments can do to create good, high-skill jobs that people can transition into.
Is automation overwhelmingly good or bad?
Overwhelmingly good. First, it is hard to stop it. Second, the long-term benefits will outweigh the short-term costs of unemployment and wage differentials. Automation will advance everything from manufacturing processes to disease diagnosis to government policy analysis. However, because it is inherently an unevenly distributed process that unfairly harms the poor and less-educated, governments will need to rectify its social costs.
The biggest problem I see right now isn’t unemployment, but inequality.
Will we be work-free and utopian?
Work-free no; utopian maybe. We will not stop working – humans have endless desires and a natural inclination to invent and create. We will, however, do more meaningful and satisfying work. “BS jobs” and “menial tasks” will have low returns. The opposite jobs and tasks will have high returns. Humans will be channeled, right from the beginning of education, into these jobs. That sounds pretty good to me.
Jan 2018 update: Redistributive mechanisms, like UBI, that make non-work life more compelling, will increase the return to working in some dissatisfying areas because you will need to pay workers more to trade their time for the job.
How will the economy transition?
Information, automation, and knowledge-based industries will grow. Highly automated economies will do things like develop high technology, write software, and produce knowledge.
Jan 2018 update: Currently, the fastest growing occupation sectors have to do with health (with an emphasis on personal health), education, management, and engineering/science (especially computer science.)
A better question is – what new industries will be created?
How unequal will our future be?
Highly, unless we do something about it. Structural capital forces – brilliantly illustrated by Piketty – are already exacerbating wealth divides. Automation will exacerbate this.
Automation means:
- Companies require fewer workers to produce the same output – meaning more money per worker and shareholder. (Walmart and Facebook are worth ~$200B. Facebook has 500x fewer employees.)
- Displacing low-wealth, low-skill employees, in favour of high-wealth, high-skill employees.
Artificial Intelligence
How close are we to (general, human-level) AI?
Much further away than sensational media implies. In order to have Artificial General Intelligence (AGI), I suspect we need to decode fundamental principles like “planning”, “concepts”, “generalisation”, “storytelling” and “abstraction”. We are not close to doing any of those very well.
Take what you read with a grain of salt. We have AlphaGo & Watson – but those are narrow AI; they’re great at a few specific things. They do not, as yet, verge on the fundamental theories of intelligence.
I believe that breakthroughs will come equally from neuroscience, computer science, and linguistics. Neither of these fields have had significant, functional breakthroughs. However, I expect our view of the world to be very, very different in 2036; we will consider AI, I think, as an “almost-human” part of our everyday reality.
Jan 2018 Update: I am more pessimistic now than I was when I wrote this. In any case, we are more likely to destroy ourself first with non-AGI. If we survive that, then we should worry a lot about AGI.
Is there something special about humans vs. really good computers? Or, are we just biological “wetware” with neurological “software”?
I don’t know. You hear AI and computational neuroscience experts suggest that humans are just complex reinforcement learners – with dopamine channels acting as “rewards” and the brain as a highly parallel processor. Behaviourally it seems true. Whether or not it’s the case, I’m unresolved.
What I believe is this: the right way to find the answer is to challenge our dogmas of being human. Rather than working upwards from a computer to an Artificial Intelligence, I say we work downwards from a human to a computer. There’s a lot of power in realizing that we are, largely, the end result of a 2.7 billion year process of molecular interactions resting on a 3.2 billion letter-long string made of 4 nucleotides. Viewed this way, our complexity seems to be explainable. We detach a little more from our insecurity-based dogma that “humans are special.” Who knows – maybe childhood is like training a model and we really are Bayesian machines that continually update and optimize hyperparameters.
Economics
How will economics, as a discipline, change in the future?
I foresee three big changes:
1. It will become highly empirical. Abundant data about our world will allow economics to take the pulse of the earth. Economics began as political philosophy; a set of frameworks, often anecdotal, that explained the world and attempted to predict the effect of various actions. The next big change was mathematical. Paul Samuelson’s famous textbook gave the philosophical topic the mathematical framework it has today. However, both operated under the constraint of little data and poor analysis tools. Today, advances in computing and data measurement will allow economics to derive new theories and insights from data itself. Math won’t disappear, but I suspect math, intuition, and data will have at least equal weight in creating new economic insights; skewing towards data.
2. It will borrow from Machine Learning, computer science, physics, biology, and others. An economics graduate student will code brilliantly; will know more than linear models; will be fluent in Bayesian statistics and Machine Learning statistics; will use massive data sets and streams; will be encouraged to take complexity and network sciences, like physics and biology, to inform ideas of the economy.
3. It will be rigorously open-access. This, I hope, will be the easiest. Two thirds of all economics papers cannot be immediately replicated if one downloads data and attempts the given source code. Through adopting a more computational mindset and ethos, the community will become more open-source and open-access, leading to higher quality data and experiment replication.
What are the key challenges in economics?
Jan 2018 update: I am just about to write a post on this. To the young economist, I think you should feel excited that economics will be one of the most exciting and powerful thinking toolkits of the 21st century. You will be tasked with figuring out the answers to:
- Making the Energy Transition happen as fast as possible, and making a lot of wealth out of it
- Figuring out if Post-scarcity is possible
- On a related note, should we focus on Redistribution or should we focus on growth
- Predicting the effects of Robots and Automation and identifying New Industries of the 21st Century and how to foster them
- Dealing with Inequality, in particular Inequality of Mobility/Opportunity
- Asking if the economic development model has changed
- Imagining the most extreme extent of AI in economic systems:
- The extent of automated manufacturing
- The effect of automated scientific discovery
- Removing transaction costs
- Eliminating information asymmetries
- Removing physical access asymmetries
- Figuring out a great multidimensional indicator of country wellbeing: GDP, inequality, environmental wellbeing
- How does innovation happen? This is something we’re not very clear about at a micro level. (See: automated scientific discovery)