Deep Human Learning
Machine learning is a hot topic in the media, promised to impact all industries. Essentially, we are trying to figure out how a machine can help us make decisions. Specifically to look at a large volume of data in a given area and make good decisions in a short period of time. This could be in areas as diverse as advertising, medicine or city planning. With this focus on getting machines to quickly learn new areas, it makes me wonder why we aren’t pursuing methods to enable expertise quickly for humans? A concept I’ll refer to in this article as deep human learning.
Human education still seems to be based on the idea that you spend a lot of time getting general skills (school and college), and often expect to spend your career in a single area. For example, most lawyers stay lawyers. They spend time in general education and then law school, and from there it is set. Similar for most other professional degrees: medicine, masters degrees, nursing, and so on.
However, the external world and its demands are changing at a rate that is much faster than a standard career. Even within the same job, the most important skills are constantly changing. For example, the top skill for CEOs has become vision setting in the past 7 years over other skills, like critical thinking or attracting top talent. Critical thinking was the most important in 2011, according to EDA’s Trends in Executive Development, and has been quickly superseded. LinkedIn’s ‘in demand’ skills seems to change on a yearly or even monthly basis. So why aren’t we figuring out how to make humans capable of fast, deep learning?
Even in places where training is offered to switch careers, like a coding bootcamp, the idea is to get someone a small set of specific skills to get junior roles in that subject area. Senior roles still expect several years of experience in a given subject area in order to get hired. These institutes are not aiming to create experts in a subject area. Likely because there isn’t a belief that you can develop expertise in an area in a few weeks.
I assert there is a way to train and learn so that humans can get relevant deep expertise quickly. In fact, I have seen it in action. I was lucky enough to spend a summer at a venture formation company that is essentially based on such a process. It was a VC firm that starts companies by having small teams of people quickly develop expertise in an area of biology, to see if there is an opportunity to start a company with some new insight. This process involves 1 or 2 people reading the relevant literature, and talking to experts to quickly find the connections in a given field, and the true unknowns. In this way the core fundamentals of that field are understood and the connections and key questions are clear. The state of the field is then known and hypotheses can be developed for how to answer the open questions.
What that experience taught me is that deep human learning is possible. Specifically, to develop expertise in a new subject area in a rather short period of time. Each of these ‘explorations’ of a new topic was phased. The first phase was 2 weeks. That’s a really short amount of time within which to learn a new topic, develop key hypotheses, and propose clear next steps. The next stage was de-risking the key open questions, which is a months time scale endeavor, at which point a company was formed, and funds were raised. That’s an impressively fast cycle. The whole cycle was essentially based on the idea that it is possible to become an expert in a new area quickly and know exactly what questions to ask.
Now, the key thing that the firm did not do, was train people to become fast learners, they simply hired those people in. The primary hiring profile was PhDs from Harvard and MIT, largely as a consequence of proximity to the firm’s office (it was literally next door to MIT’s campus). So I have been wondering why it is that they primarily hired PhDs for this process, while the rest of the ‘entrepreneurial’ world almost shuns education. In some extremes, like the Thiel fellowship, suggesting a college education is not necessary.
My belief is that the following skills are needed to learn a new area quickly and deeply:
- Understanding what the fundamental truths are for that field
- Thinking in first principles
- Understanding the ‘nth order’ effects of a given action.
Meaning, you need to get trained to think about what the true basic principle is that governs the system and to think about the 0, 1st and 2nd order effects of that basic principle.
Fundamental Truths
Getting really good at understanding the fundamental truth is probably the hardest skill to develop. It requires a great degree of humility and curiosity to make sure you are asking enough questions about what is true, and why. It’s not easy to determine if an observed behavior is indeed that truth for a given system. In fact, even among the large majority of scientists and researchers, people very often point to the fundamental truth of a system, and are entirely wrong. There is something more fundamental occuring, and they simply haven’t asked enough questions to find it.
That said, doing a PhD, your humility and curiosity are tested pretty quickly. The failure rate of experiments in most fields is extremely high. I must have produced tens of boxes, that contained thousands of vials, of chemicals that were the outputs of failed experiments. You don’t have a choice really, if you are being diligent, you will definitely gain humility quickly. It will become pretty clear pretty fast, that there is a lot that you don’t know. So then, what I have seen great PhDs do is to turn that humility and failure into intense curiosity. ‘OK, so that didn’t work, I wonder why. What if I try X instead? What if the governing principle is Y?’ and so on.
Don’t get me wrong, that amount of failure can often easily spiral the other direction as well, and lead to intense demotivation. However, if you are able to find a way to channel that failure into curiosity, you will have learned an incredibly important skill. In general, I found that the support structure around the student who is failing is the main variable that determines whether you get a great outcome, or just demotivation. I was fortunate that I worked with a postdoc who channeled my curiosity and didn’t allow our thousands of failures to lead to demotivation.
This leads me to the conclusion that there must be a way to enable learning of this deep curiosity among adults by allowing them to truly fail and learn as a team with strong support. This idea on its own is actually pretty widely discussed at this point. The Lean Startup, and other popular writings essentially espouse this same concept. In fact, the startup methodology is directly taken from the scientific method.
The important idea in this article is those attempts and failures should be in pursuit of understanding the fundamental truth of the system. Meaning, the experimentation and failure must occur to learn those truths that are fundamental to the system in question. In addition to being guided to specific learning, this activity must be enabled several times so that the person absorbing this skill can quickly get the core skill honed of learning to find the fundamental truth of any system.
Thinking in First Principles
Intimately tied to finding the these truths for a given system, is thinking in first principles. This specifically means that the ‘rules’ governing that system are only those that are the most basic to that system. So any conclusions you reach should be based on some combination of the fundamental truths, not based on other observations, and especially not based on symptoms. One way I think about this is ‘anything is possible as long as you don’t break the laws of thermodynamics.’
The easiest thing for most of us to do, is to think based on symptoms. For example, it is really easy for someone to think ‘the last time I gave a task to John it took him really long to complete, so it’ll take long this time again.’ This type of ‘symptom’ thinking is easy because as humans we want to quickly associate the symptom to the fundamental truth. This is not correct. A first principles approach would have first discovered the actual fundamental truth. This may have been ‘the timeline was extended last time because it was during holiday season.’ Then you may learn ‘during holiday season, the person John depends on is always gone for two weeks.’ There you have it. John may actually be a great worker. However, you could easily assume he is isn’t if you don’t ask enough questions. In this case it was no person’s fault. If at all, it was was probably your fault for not knowing that there was a critical person who takes two weeks off during the holiday season.
So now, if you were to take actions based on first principles, what you would do is to understand all of the steps you needed to happen, and set the process up to succeed. For example, you would ask John to do the task at the time when the person he relies on is around. Further, you might ensure John has the bandwidth to do the task when assigned. Finally, understanding that John is motivated to complete tasks when given independence, you would assign the task, say you have confidence, and then let him run. Everything is being done based on first order truths and processes.
Understanding the nth order effects of a given action
Thinking in first principles is the right way to plan an activity to occur. Understanding the nth order effects takes that idea and asks what else will happen as a consequence of this decision? It is typically the case that the 0 and 1st order effects of an action are clear to most people. However the 2nd and 3rd order effects tend to be less obvious. Often because they are hidden behind some sort of assumption that you need to make in order to get to a second order effect.
In science, the order of the effects is often based on conditions of the experiment. For example, let’s say the core material property is that it is elastic (the fundamental truth). The first principle is then that you can stretch it and it will return to its shape, so you can design systems with the material based on this. However, the 2nd order effect is that with repeated stretches, and returns, the material heats up, and this reduces its elasticity. So you actually have a different property if you need to do this in a repeated fashion. The third order effect of this may be that once the material loses its elasticity, it actually becomes plastic (i.e. permanently deforms), which then means that when you next try to stretch it, it simply breaks. However, there are several things you need to know about the material before you can make this third order conclusion. Remember, we started with the simple idea that the material is elastic, so we can use it to build a system that needs to needs to return to a certain position, and ended up with something that would break if it needed to be used in repetition.
This same sort of nth order thinking can be immensely valuable in learning any system deeply. Whether it is a scientific, people, or organizational problems. I use this type of process when trying to understand and make organizational decisions. How a team operates is based on people’s strengths (a fundamental truth), and also based on who their boss is (a first principle). The interaction between those two people is then a second order phenomenon. As such, if you think through the problem in this way, you will understand the organization quickly, and be able to make recommendations quickly as well.
As you can see, if you seek to understand the truth, then apply first principles, and then start to understand nth order effect, you can learn a topic very quickly. This deep human learning is of course founded in a belief that heuristic decisions can be a great way for a human to learn a new field. I have proposed supplementing the core principle of heuristic decision making by forcing the learner to start by asking what the fundamental truths are, which in general is not a tenant of heuristic decision making.
This form of learning is also deeply human, It relies on the type of thinking humans are better at computer than. To prioritize certain aspects, and ignore the rest. Hence, this form of learning can enable humans to learn in a different ways than a machine.
How to enable deep human learning
So hopefully by now, you can see what the principles are of deep human learning, and why it is useful. So I’ll propose how I believe we can accomplish this. As stated early in this article, I have found that some PhDs end up being very talented at this skill. However, this is not something all PhDs end up learning, and similarly there are plenty of non-PhDs who are great at this. Hence, what is clear is that it isn’t unique to the PhD process. In fact, the most important aspects to enabling this is essentially 3 factors:
- A boss/teacher who enables you to fail safely, focussed on learning the truth of a system
- Repeated trials over an extended period of time
- Testing and honing your nth order thinking
As you can see, the first, and perhaps most important factor, of a boss that enables failure is more likely in a PhD. The only reason is that research is by nature something that is very difficult to get to ‘work.’ So most good PhD advisors help you learn how to fail, and learn from the failure. Typically moving to find the fundamental truth of a system. However, this is not exclusive to PhDs. In fact, I have been lucky to experience this in the tech world as well. Having had bosses that have allowed me to fail in some way or the other, and learning from those mistakes. This enabled me to quickly figure out the core truths of our organization and business.
The latter two factors from the above are also easily available to PhD students, again, just due to the nature of research. However, the core principle is easily accomplished in other fields as well. It just takes a really good boss to enable that form of experience and learning. Especially the third one. You need to either be challenged to think in this way or challenge yourself. Then you need to test your deductive skills to getting that form of thinking correct. Over time this form of thinking gets better as you get feedback from the previous times you applied this type of thinking.
I believe strongly that we can in fact train humans to become experts in a field quickly, and it requires only that a certain type of learning skill is honed. This learning skill is often found among those who have gone through a PhD, and can easily be replicated outside of that structure. In fact, it can be taught while in an industry job if you have a great boss that enables you to think in this way.
I hope that you spend some time thinking about the ideas here for how you can become an expert quickly, and try this type of learning. Deep human learning can enable you to gain knowledge in new areas quickly, and consequently grow your career and ability to add value in non linear ways!