Gradient Ascent for "Career" Optimization
I’ve always been daunted by the concept of a “career”, something that I have found confusing at best and misleading at worst. When people think of a career, it usually means a post-graduation life focused on a specific skill and gaining experience in a specific field, to eventually get to a position of mastery and leadership. It makes sense why. Oxford defines a career as an
occupation undertaken for a significant period of a person's life and with opportunities for progress
Take note of the singular “occupation” (not occupations) that you do for a “significant period” of your life. This is in line with a very traditional view of a career. For some, this passion-driven way of approaching a singular outcome is optimal, and society bolsters these specialists, who we are taught to revere.
Jim Collins says that we should do what we are genetically encoded to do. But if there are millions of unique and high potential people, with extremely different genetic encodings, why can we count the number of “good careers” on one hand? Why do we conflate genetic encoding and societal conditioning?
For many, the traditionally defined career is not sufficient to creating the optimal post-grad life. Taking a computer science perspective, building your career is like being given a solution and then reverse engineering to find the optimal algorithm to get to that solution. This approach is great for when you have the solution in hand, and the path to that solution is a very clear application of effort. For example, if you wanted to be an Engineering Manager at a tech company, the path is relatively straightforward. Study computer science in school, study for programming interviews, get hired at a tech company, work diligently for a few years, build the expertise in the software you work on, and slowly take on more leadership until you can begin to manage people. Using David Epstein’s analogy in Range, this approach is like getting really good at golf. Practice and perfect the strokes and play hundreds of holes, and there is a structured path to being the best.
On the other hand, life is full of uncertainty and twists and turns. We aren’t usually given a solution. We often figure things out earlier, or later than others. There is no playbook for how to be successful. According to Robin Hogarth, life is more like “martian tennis” than golf; you are aware there is a court and there are rackets and balls, but no one has told you the rules. In such a seemingly sub-optimal setup, how do we find optimality? How do we extract clarity from this mess?
For inspiration, we can again turn to computer science to find an algorithm that satisfies our needs. Enter the gradient ascent algorithm (gradient descent but finding a maximum instead of minimum for this analogy). According to Wikipedia:
Gradient ascent is a first-order iterative optimization algorithm for finding a local maximum of a differentiable function. To find a local maximum of a function using gradient ascent, we take steps proportional to the positive of the gradient (or approximate gradient) of the function at the current point.
Essentially, this algorithm starts at some arbitrary point and then looks at locally available information to figure out the next step, which at any point is simply the steepest grade of ascent possible (the highest positive gradient or slope available). It then chooses this local maximum and then repeats the algorithm until it gets as close as possible to an unknown “global maximum” (the red peaks in the graph above). The interesting takeaway here is that sometimes the steepest current option might not lead to the overall best option, and sometimes we might have to first descend before we can climb back to the optimum. But these aspects of the algorithm make it more human than anything, because it can give us a framework to decide our own optimum when the available information may be scarce.
So I think the question of how to build a good career in something is a bit misguided as it assumes we know the solution. I think its better we use gradient ascent as a way to find the most optimal path we can take in the short term, using the heuristic of learning and development. At every point in your life, evaluate at which opportunity you can learn and develop the most at that moment in time, and then take advantage of it. Rinse and repeat. As counterintuitive as it sounds, I think this potentially meandering path could be very exciting and impactful for those who may not know their genetic encoding, or may feel uncomfortable that they are not “specialized” in any one thing. Furthermore, this could lead to a wealth of interdisciplinary and innovative thinking that can take you very far, as you learn to apply core principles across fields, maximizing your ability to create and innovate.
Credit for inspiration goes to David Epstein. I’d (and Bill Gurley) highly recommend his newest book, Range.