My latest field of endeavour is to learn as much as I can about AI and Deep Learning and how these disciplines are going to impact the field of strategy and marketing.
In my explorations I have come across the P versus NP problem, a major unsolved problem in computer science.
Without getting into the weeds let me try and state the problem in simple terms and set it up for the thesis of this post.
A class of problems can be solved within a finite amount of time (polynomial time — time that grow polynomially with the size of the problem and not exponentially). Let’s call them P class of problems. There are other problems whose answer if known can be verified in polynomial time — such problems are known as NP class of problems.
Take the example of Sudoku -an incomplete grid of numbers that must be completed to fulfill a condition say all the columns, rows and diagonals adding up to same number. Given a solution verifying whether it is correct or not can be done fairly quickly. This remains true even as the Sudoku grid grows bigger. That is the Sudoku problem is an NP class of problem.
The big issue that is at the frontier of computer science and Deep Learning is whether a NP type problem — a class of problems wherein a given solution can be verified quickly (in polynomial time) is also a problem that can be solved quickly (in polynomial time). That is whether NP=P.
While currently the intuitive answer among large section of mathematicians, computer scientists and thinkers is that NP is not equal to P. For example, a NP type problem like Sudoku there is a perception that there is no algorithm that can solve a Sudoku in finite time as the Sudoku grid grows in size and complexity. NP is not equal to P.
However at the cutting edge of Deep Learning this notion is being challenged. Advances in Deep Learning like AlphaGo in 2016 seem to be on the way to proving to NP=P. AlphaGo could crack the code of games like Chess and Go and beat champion humans based on just a knowledge of the rules and a finite amount of playing against itself. This is in contrast to Deep Blue — the IBM chess computer that beat Gary Kasparov in 1996 — which worked on the basis of analysing every move against a database of millions of games — a rather inelegant and inefficient process compared to the AlphaGo process.
It is expected that the next 20 years could see the emergence of General Intelligence (GI) systems driven by Deep Learning algorithms that have proven the NP=P hypothesis across a large universe of problems.
General Intelligence that can match human intelligence — intelligence that can respond to real world conditions with messy boundary conditions and a shifting, ill-defined set of objectives.
In other words creative intelligence.
Let me now state the thesis of this post: Advertising is a perfect encapsulation of the P versus NP problem!
Anyone who has worked in advertising knows that there is both a science and an art to it.
The science works through the research and strategy planning process and the art resides in the creative development process.
If one thinks about it the research and strategy planning process addresses a P class problem. In that a competent researcher or planner can build a research or communication strategy plan for any given situation in a reasonable amount of time as long as the situation is well-defined and all the required data is at hand.
The creative process, on the other hand, is quite different. Most good advertising and marketing people will tell you that it is easy to judge an advertising campaign as good or bad. That is creative judgement is a NP class of problem. However the same people will also confirm that there is no guarantee when and if a creative team will deliver a good advertising campaign.
Currently as things stand it seems advertising is proof that NP is not equal to P!
That is why, perhaps there are as many cases, if not more, where strategists are forced to write strategy to good advertising to cases where creatives develop campaigns to strategy.
Some creative types in advertising are prone to dismiss strategists to the status of critics who write reviews for art after it is produced! While that is debatable I have tried to fend away such criticism (with some success) by positioning myself as working at the creative end of strategy and the strategy end of creative!
Ribbing aside, the day Deep Learning proceeds to the stage where NP=P is the day tasks and professions that require creativity will come under the ambit of AI (or more accurately GI — Machines with General Intelligence).
Ian McEwan’s new novel “Machines Like Me” is a masterly exposition of what happens when the NP=P theorem is proven. And does so without getting all “science-fictiony” and with that particular McEwan brand of wry humor. Worth a read.