AI, Big data, Automation,
Bill Gates and now Mark Zuckerberg have used public forums to talk about the coming seismic shift in jobs that learning machines and automation will have. But I worry what they are actually doing is softening the message for their audiences. Zuckerberg focused in his Harvard commencement address on the impact on low end jobs. Gates sounded his trademark optimistic tone about the future. While I agree that the stats don't lie and we live in an unprecedented period of peace and declining poverty I worry that both of these tech leaders may be sparing their audience a painful message.
Be wary of any conversation about AI and automation that focuses exclusively on self driving cars, warehouse workers and the like. We can deal with those because it implies there will be plenty of opportunity for high paying white collar workers. You can write the whole thing off as an overall positive element of development. From subsistence farmers to information workers in a few centuries. Affected workers can hope to get enough time out of their careers to put their kids through graduate school and look forward to their family moving up.
Instead, what if Zuckerberg stood up at Harvard and said the real risk is to high paying jobs including architects, engineers, lawyers, accountants, managers, doctors and even computer programmers. Presumably, there were few long-haul truck drivers, current or future, in the audience at the Harvard commencement. What if the pair came out and said that programming jobs are at risk? How differently would their audiences respond?
Our economy and society arguably depend upon a few agreed upon premises; social mobility, progress destroys jobs but creates better ones. People need work and will lower standards if forced to. If graduates with JDs cannot find jobs as lawyers and begin competing with business graduates for management or consulting jobs and those people compete with sales people for those jobs where does it end? You cannot run our economy on taxing billionaires alone, in spite of that tiny group's value in political campaigns. The top 1/8th of US households have incomes well above $100k a year. The majority of these people rely on wages, not capital gains or investment income. They are wage earners from dual income sales couples and programmers to dentists, lawyers, doctors and CEOs. Decimate this group and you will have real problems. Imagine the best educated parts of society competing for shrinking job opportunities. What if we create too few robotics engineering and automation design jobs to make up for this. This group represents the upwardly mobile hopes of millions of working class people. Drivers may accept that their job wont always be around so long as their kids can go to medical school. Without that hope and the critical tax base this population represents how can we hope to invest and innovate our way through the job disruptions?
The Verge and others are reporting that Apple's big bets on AI is leading to product innovations sooner than you may expect. This includes their plans to add dedicated AI processing in future versions of the iPhone. The initial use of this is an important one. A lot of needed developments in smart phones are code intensive. This includes augmented reality, biometrics, simulations and advanced encryption technologies. This is not to mention the potential for sci-fi inspired leaps in automated assistants.
Tasks that are very code intensive are therefore processor intensive and that makes them power and therefore battery intensive. Add AI chips to iPhones and you reduce the processing power that needs to be dedicated to these tasks and make smaller batteries last longer.
This is important because productizing AI technologies will drive massive increases in innovations. Profits, manufacturing scale, increased job creation and attraction of investment dollars are the fastest way to accelerate any field. Academic, government and venture investment are important too but the drive they provide is very different. When the field gets pushed forward by the economic tsunamis that products like the iPhone create it will lead to more rapid advances on the other end of the computing spectrum.
Today it can take super computers up to a year to run certain simulations. Quantum chromodynamics is the theory of strong interactions and relates to model of nuclear forces. Astrophysical and cosmological dark matter simulations can take months. So when you buy future iPhones you may be powering the forces that will unlock our understanding of everything from atomic processes to the very structure of the universe.
Last November Microsoft and the University of Cambridge published a paper about how its AI was writing code. The process includes scanning bits of code and considering lots of possibilities for how bits of code could fit together to create the right solution. This was falsely reported as Microsoft AI plagiarizing existing programs. That says something interesting about us and about what learning really is.
We all accept that technological change destroys some jobs and creates new ones. We remain excited about the process because, in general, the newly created jobs tend to be good ones and the innovation lead to positive things for society. As bad as factories were, the small-scale farming people left behind had its issues too. The personal computer revolution created entirely new categories of white collar jobs in the tens of millions. Because of advances in tools and training, it became easier to become a professional computer programmer even without a college degree in computer science. My grandfather turned wrenches, my father was the first in our family to work in an office and I work with computers. One worry about the AI revolution is what if this one doesn’t create new jobs like the past revolutions did?
When you conduct media interviews you get a peak into the process of creating stories. I have seen how journalists find new ideas for new stories in academic and other original sources of news and report on them in industry press. If they are interesting enough then reporters in mainstream press will pick it up. It is often a lot of steps from medical journal to the table in your dentist’s waiting room. It is a complicated process whereby journalists learn about a topic and then simplify it and make it accessible and interesting to their audience. Some reporters do it very well but others get parts wrong. Much like a game of telegraph (or telephone?) reporters will find story ideas from other reporters, often without reading the original source materials. When a reporter misinterprets some coverage and drives an engaging conclusion that can be picked up by another journalist in another story, and so on and so on.
The process of how stories are generated reflects a lot about how humans learn and how original works are created. Even the most novel innovation contains some derivative parts. Solving new problems means building upon what is done before. This is natural and the line between this and plagiarism boils down to now much originality has been added to what already existed.
One of the current fears of AI is that it will replace us in our jobs. One intellectual life preserver for many people in tech is that programming is a creative job that reflects how differently we think than AI ever possibly could. This argument boils down to the view of AI as a tool for automation and not one for actual creation. So if AI steals that means it isn’t creating and programming jobs along with whole categories of jobs are safe. The particularly scary scenario is that AI creates itself and destroys some jobs but does not create new types of jobs. But, AI is already starting to create AI such as this experiment at Google Brain. It is also predicting cancer better than doctors, fighting parking tickets, creating art and stories and yes, writing code. So one of the first things we need to do to prepare for the full impact of AI is to rethink how we think about creativity and learning.
Decide now how you feel about AI powered drones autonomously patrolling the skies. Imagine them learning from successes and failures and using new experiences to make decisions about how to apply rules of engagement to potential to targets below.
Drones have been one of the most controversial new developments in warfare. Your view about a drone in Afghanistan being operated by a military person sitting down the street from a suburban Red Lobster in America is something of a litmus for your views on the use of the military writ large. But autonomous flying vehicles are a separate issue altogether.
A Carnegie Mellon Robotics Institute professor recently let a drone learn how to fly indoors by letting it crash over 11000 times in 20 difference indoor spaces. Watch the video for more.
One of the big questions regarding the future impact and the likelihood of AI doing things we don't want it to is the extent to which its abilities can be extended. I understand the argument that people are prone to making decisions we don't like too. But that alone is not an argument for treading very carefully here.
Read more in this Digital Trends article and tell me what you think.
Microsoft CEO Satya Nadella Keynote at the Microsoft Build event on May 8th 2017 raises important questions about the role that the tech industry has in making sure AI doesn't generate deeply damaging unintended consequences but it also skirts the impact of intended consequences. It also highlights a scary truth, AI is long out of the academic research projects and in the hands of real product development teams. Satya says something I have not seen before. He points out that the future impact of AI depends upon the decisions of individual developers, program managers, business analysts, etc. who make the actual product design decisions behind the software that we use. In my time as a manager in Microsoft's product development organization in Redmond I saw the enormous scale of these software projects. Some are huge but many are small to medium sized teams with autonomy to do really whatever they want. I oversimplify a lot but the important point is that software isn't really a part of some large plan, it is the sum of huge number of individual decisions made by lots of people.
This is a real issue and the big AI houses - including Microsoft, Facebook, Apple, Amazon, IBM - should address this with internal standards and safeguards. Individual responsibility is crucial and creating a culture that cares about the impact of AI and not just the advancement and application of it is crucial as well.
The unintended consequences of AI gone wrong are troubling enough. But we need to really start addressing the intended ones. At the same presentation Satya spoke about the impact on blue collar jobs but that isn't the end of it all. the destruction of millions of driver and warehouse jobs is not the really big problems to solve (although the impact will be huge) but the what happens when large numbers of very broad ranges of high-paying information age and white collar jobs including lawyers, doctors, IT, programming, accounting are made unnecessary but not replaced by new categories of jobs as quickly.
I don't think that AI is going to kill us. I think for starters there are too many smart people worried about that and obvious disincentives in creating a self learning system with that potential. I understand the argument that creating AI is all about creating systems that can develop beyond its programming. The road to killing us is long and filled with enough mistakes that even an arguably fast maturing singularity won't get there.
What I do worry about is the social impact of the effect on the global economy of AI. I am in the camp that believes that few jobs cannot ultimately be eliminated. That won't happen overnight and by then we will have had to deal with a major overhaul of employment as a concept. The issue for the next few decades is that most jobs will be dramatically streamlined with major efficiency improvements. This impacted most fields but we survived that in the big PC, network and internet revolutions from the 1970s to today because it lowered effective costs and these revolutions created entire new categories of jobs, especially high paying ones with trickle down benefits and opportunity. These results didn't trickle all the way and too many people couldn't make the transition leading to a rise in low pay services jobs and underemployment at the same time companies struggled to source enough new tech workers.
AI and Robotics will accelerate with advances from quantum computing. Then, paired with improved technologies in VR and IoT, this will reduce the demand for jobs greater than it creates new ones.
Take legal for example, AI doesn't have to replace lawyers but if it streamlines contracts and advisory services enough as a tool for lawyers than the demand for lawyers is reduced. More formulaic things will automate completely as evidenced by a Stanford student automating traffic ticket defenses. Lawyers are smart but those skills are only so transferable. That makes the supply of legal services relatively inelastic. When the same thing is happening in computer hardware and software, medicine and other major white collar and information age fields, where do these people go?
In this blog I am going to explore these issues and also touch on emerging thinking on how we address these things. This includes the periodic waves of attempts to suppress it like regulations or protectionism but will focus more on solution like reinventing education and guaranteed minimum income solutions.
Mike Zammuto is the CEO of Completed.com and Cloud Commerce and a strategic adviser to several startups. Mike's background is in SaaS services, B2C sites and B2B firms and has worked extensively in online reputation, digital marketing and branding.