Thomas Frey has a thought provoking article “78 Skills that are Difficult to Automate”.
Frey breaks down the categories of jobs that he believes will remain “safe” from automation:
- Complex systems too expensive to automate
- Creative endeavors that only humans can appreciate
- Human to human interactions that produce an emotional response
- Decisions that need human-based reasoning
- Complicated outputs that demand a human translator
- Situations that require the human touch
- Settings where the loyalty of hacker-proof humans is preferable over digital machines
- Human to human valuations
- Positions where humans control robots
I will attempt to address each one from the perspective of the “Deep Learning Canvas” that we’ve developed. In our approach, we fuse together the ideas of “Jobs to be Done” (JTBD) approach for identifying tasks that need to be addressed and an understanding of the cognitive limitations that can be enhanced through Deep Learning.
In Frey’s article, he builds up his argument that the jobs most likely to be safe are those that take into account the irrationality of humans. The JTBD approach also goes beyond pure functionality, but also addresses human needs such as emotion and social currency. Whatever the Deep Learning system will be automating, it will need to target any combination of these three needs.
Now, we don’t use a blunt instrument and try to replace tasks wholesale. Rather, we do so in a surgical manner, identifying first the capabilities required to address the customers needs and then identifying which specific cognitive tasks can be enhanced through automation.
What people seem to miss is that the replacement of jobs is performed piecemeal and incrementally. This kind of detail is missed by many prognosticators. The capabilities of those that remain employed become more powerful. The lesser skilled folk who are first to lose their jobs become less adept at using automation tools. It is a vicious cycle where those that have the skills, gain more advanced skills. While those that don’t have the skills are forced to pay premium to gain the skills. In many cases, there aren’t any educational institutions that exist to teach them these skills. Furthermore, less skills imply more commoditization.
People will be forced into the ‘gig economy’, relegated to endlessly competing in markets with margins pushed towards zero.
However, let’s examine Frey’s list because it is an informative base to perform more detailed analysis.
Complex systems too expensive to automate
Complex systems will always require people with advanced skills to orchestrate. However, that does not imply that there will less automation that will be enabling this activity. In fact, one should expect bleeding edge companies to leverage as much automation as needed to accelerate work. Tesla and SpaceX are likely highly automated companies as compared to their competitors.
Company’s that find it too expensive to automate processes are likely the ones that don’t know how to exploit automation to reduce development costs.
Creative endeavors that only humans can appreciate
This is similar to the previous but relates to more artistic endeavors. In a lot of the Hollywood Science Fiction blockbusters, we continue to use increasingly advanced automation to reduce costs. Film makers no longer need to hire armies of extras to film massive battle scenes. These are now all done through CGI simulation. There was a period in time were Epics were too expensive to make, but today that’s no longer a problem.
Certainly we’ll have the creative folks continue to drive development, but the human resources required will continue to diminish. In the future, we may not even need actors for films and just use CGI renditions of famous actors. In fact, if we ever get bored with seeing the same faces, we can always generate arbitrary faces!
Human to human interactions that produce an emotional response
Most of what Frey describes, (i.e. a smile, a hug, a kiss, a massage etc.) usually aren’t paid for. In fact, in many societies, paying for these “human interactions” would be considered illegal!
Decisions that need human-based reasoning
The human as a watchdog against runaway automation. I think this kind of job will not go away if governments enact regulation that requires this. It is just like humans who serve as gas attendants. Certain states require it, although it isn’t absolutely necessary. However, this kind of legislation can keep a lot of people on the job.
Complicated outputs that demand a human translator
Examples: Doctors, Data analysts, judges, business executives, privacy advocates, relationship building strategies, birthing processes, genealogical mapping.
This is not very different from the previous class or even the first class. It is just supervision at the top rather than at the bottom. It however pertains to the need for humans to interpret machines so that other humans can understand (and accept) a machine’s conclusions.
However, this likely won’t change because legislation is already in place that humans are required for these kinds of job. We can’t have non-human judges sending people off to lifetime incarceration or death row.
Situations that require the human touch
Examples: Teaching
The examples Frey provides in this category all seem to revolve around education. However, I would argue that education is moving massively online and that education will only get better with Virtual Reality, Augmented Reality and AI. That’s because we can build many more interactive environments that can serve many more students at a quality level that is many times better than the average teacher.
The problem with our current education practices is that the lecture model is all passive listening. We need to strive for greater participation of the student in the model of Active Learning. People learn best by doing and not by lecture. The expertise to create highly engaging education will be in demand. This will require a deep understanding of the interactive technology, an understanding of human behavior and teaching methods. Where Deep Learning comes into play is its ability to react to human behavior. This is essential to effective teaching and is not outside the realm of what’s possible.
Settings where the loyalty of hacker-proof humans is preferable over digital machines
Examples: Guarding VIPs, holding secrets, personal confidants, safeguarding corporate knowledge, consultants, lobbyists, leader of robot resistance groups(?)
On the contrary, Blockchain systems have been shown to be hacker-proof as opposed to systems that have human elements that can be “socially engineered”.
Human to human valuations
Examples: Stock market, voting, government policies, consequences on policy violations, buying, purchasing agents, rating agencies, surveys and poll.
We already have systems that make all sorts of ‘valuations’ on what we should focus our attention on. Facebook manages our reading list. Amazon recommends products that we might prefer. Google filters our search results. More and more, AI is making our decisions for us. We’ve become so used to machines like GPS giving us instructions that we’ve lost all sense of direction.
What we will likely see in this domain is that Blockchain technologies will ensure transparency and integrity of many of the interactions that assume a fair market or a democratic process. Today, many of these processes are gamed to the benefit of the very few. Human to human valuations should not be entrusted exclusively to humans, but rather it should be done through a transparent collective manner. The reason here is because humans have motivations to game the system.
In an AI economy, it will become harder and harder to game one’s credentials. Machine intelligence will become sophisticated enough to expose many in the workplace as imposters. Let’s all be perfectly honest, the reason why people get paid more than that they deserve is due to a gross inefficiency in how humans assess an employee’s worth. 20% of employees do most of the work, while 80% are all pretenders.
Positions where humans control robots
Examples: owner and managers, software developers, system engineers, product designer, robot maintainers, auctioneers that sell robots.
This is the same theme as 3 of the previous themes. That is, automation as a tool to enhance human work.
In summary, the list can be simplified even further:
- Jobs that use automation as a tool.
- Jobs that that use humans as safety valves against automation failure.
- Jobs that interpret the decisions of machines.
- Jobs that design human-machine interfaces.
- Jobs that design automation to manipulate human behavior.
That’s five classes of jobs that will exist in the future that appears to be safe. On the other hand, with the exception of the “human safety valve”, all these other jobs require high level skills. Jobs of the future need to have an deep understanding of humans as well as machines, and it is in this interaction of man with machine where jobs will exist.
I think what few seem to appreciate is that Deep Learning AI is technology that is like human intuition. It is an opposite technology from more classical AI technologies that focused on reasoning. At this time there remains a Semantic Gap. However, humans capabilities are stuck between a rock and a hard place. Between Artificial Intuition and Artificial Reasoning. This is were many people seem to be getting it all wrong about what jobs are safe and what is not. Let us not fall into this fantasy that our unique human intuition is safe from being replaced by automation.
There aren’t many jobs that are safe with the emergence of Deep Learning. We have to come to grips with so we can get a head start in examining the core of our economic system. AI will likely break capitalism and we unfortunately are not starting serious discussions on what will replace it. The stark reality is that the emerging AI economy is reserved for the highly skilled. Everyone else need not apply.
Originally posted: https://medium.com/intuitionmachine/jobs-that-deep-learning-can-automate-208b273be644