Andrew Jeavons

Persona Engineering

"How do you describe a person ? ". This is the essence of persona engineering, this is what persona engineering is distilling into a process. It’s a new concept that has come about by the use of AIs to simulate people.

Synthetic Respondents

AI personas , so called “synthetic respondents”, are becoming more and more popular. Either by use “en mass” to simulate sampling a real population or by using a few to represent segments or groups of people. There is a trend to define personas by a few demographic descriptors and some OCEAN, the big five personality framework. Is this enough ? What's the best use of the segmentation studies when developing your personas ? How do you generate the personas ? How do you keep personas consistent ? What variables do you include ? It’s a cliche that people are not simple and AI personas have to reflect this. To help with all these issues I developed the "Persona Engineering Module" to act as an integrated development environment (IDE) for AI personas, the first of its kind. Read more about this here.

Deep personas

Personas need to be complex because people are complex. I believe that "deep personas" have the best chance of modelling consumer behaviours. A deep persona may be hundreds of lines long and contain many different sorts of behavioural and psychological values and even that is not the sum of a person. Ten demographics and some OCEAN scores are definitely not a person. AI personas are models of segments or groups of people that are "run" by an AI, the probabilistic associative nature of AI makes them ideal to simulate consumers, we don't want deterministic models. We need "variation" and associative connections that only AI can provide.

Market Research meets Psychology

For decades "paper" personas have been developed and used widely yet there is little defining what a persona should contain. This could be partially due to personas being very "domain" specific. For example personas for hair shampoo may have little in common with the persona for lawn mower purchasers. It’s probably (hopefully) true that people who buy lawn mowers also wash their hair - at least sometimes.

The "basic" characteristics of people have to be similar in some ways. This is where Market Research runs into psychology. There is no surprise that psychology has a plethora of theories about how you can define a "person". You can go from the whimsical Lüscher Colour Test, my personal favorite, to the well known Myers-Briggs to the equally well known and widely used OCEAN personality model, so called Big Five model. Finally there is the HEXACO Personality Inventory which I think is the closest competitor to OCEAN. It shares many aspects of OCEAN on the surface but has some significant (and better to my mind) differences due to the increased number of personality traits.

You are what you do

It really is a question of "pick your poison" when it comes to personality assessment techniques. The default for a lot of AI persona vendors is a sprinkle (or more) of descriptive demographics with some psychographics (i.e. OCEAN scores) thrown in. There is a question as to if the demographics have been proved to be causal of behaviours, mostly I think this is ignored. It seems logical that behaviours should be emphasised - you are what you do in many respects. But deciding what these should be for each type of persona gets complicated. Just purchasing a lawn mower may not affect your hair shampoo purchases.

About

Andrew Jeavons is a consultant specialising in AI persona engineering , computational semiotics and image analysis. He has been in the technology world for more years than he cares to remember, across the USA, UK and Australia. He was a psychologist before being seduced by Unix, C, Quantime, and a long list of technology companies. Andrew is a well known commentator on technology and in particular the use of AI within market research. Andrew had become interested in neural network theory (the old name for deep learning) in the early 1980's. One of his more recent publications/talks is Quantization of Cognition, ASC Video: Quantization of Cognition.

Andrew developed the Aibods AI persona platform, and with it the concept of Persona Engineering: a structured approach to defining AI personas. A few paragraphs 'describing' a persona doesn't get you great results; people are complicated, and a handful of demographics and traits can't contain that complexity. Andrew has developed the Persona Engineering Module as part of his work at Mass Cognition. More about the PEM. With the advent of massive databases of personas the process of deep AI persona development is becoming both simpler but more complicated at the same time.

Andrew has been a lifelong fan of Charles Sanders Peirce — perhaps the greatest American philosopher no one has ever heard of — and has leveraged this slightly obscure interest into an AI-powered computational semiotics approach that brings Peirce's taxonomy to life. The ten semiotic mechanisms Examples of 10 sign analysis

It's all a sign - Computational Semiotics

Semiotics is a fascinating discipline. Peeling back the layers of meaning in images or text is always a surprising journey. But too often semiotic analysis produces a long document that re-describes the source material in different terms without resolution or direction. Semiotics should answer questions. It should not be an opportunity for obscure rumination.

A taxonomy of signs

Charles Sanders Peirce was a brilliant polymath and one who tried to bring structure to semiotics and known for adding the interpretant to semiotic analysis. Signs are not read the same way by all viewers, and the interpretant opens the door to cultural and personal effects of signs.

Peirce classified the forms signs take — icons, indexes, symbols — and built a taxonomy of signs. This consists of ten classes. These ten classes drive the AI-powered semiotic analysis of images and text I have developed. An explanation of the ten signs is here: The ten semiotic mechanisms.

Analysis is a start, not an end, this always has to be borne in mind. Examples of 10 sign analysis