The Rationale Behind the Empathy Gradient

Understanding the Impact of AI on Work

The future of work in an AI-driven world is often difficult to predict due to the speculative nature of the question. However, a significant barrier to understanding AI’s impact is the flawed foundation of many current studies being circulated in reputable business publications such as the McKinsey Quarterly, Financial Times (FT), and Harvard Business Review (HBR). These studies frequently present skewed findings because they rely on outdated frameworks.

Misleading AI Studies: A Flawed Foundation

Take, for example, the McKinsey studies published in 2021, which aim to shed light on AI’s impact. The problem lies in the fact that their analysis is built on the McKinsey Global Institute (MGI) research from 2017, which focuses on automation scenarios. These studies give undue weight to traditional optimization methods, arriving at conclusions similar to those of conventional cost-reduction models.

Relevant quote: "This research builds on MGI’s January 2017 report on automation and its impact on work activities. We assess the number and types of jobs that might be created under different scenarios through 2030 and compare that to work that could be displaced by automation."

Even the foundational studies quoted in such research can be problematic.

Task-Based Frameworks: Still Relevant, but in Need of Evolution

The foundational studies to all this work is the 2001 empirical study by Auto, Levy, and Murnane is based on the task-based framework that separates between “Routine” tasks and non-routine “Cognitive” tasks. Although this task categorisation is no longer relevant, it still serves as the foundation of the majority of works that are being quoted in the major publications that are read by senior leadership of large companies, like HBR and FT. The reason this task categorisation is no longer relevant should be obvious: AI is capable of performing cognitive tasks. The problem lies in finding an alternative for the model that can represent how tasks will get affected by AI.

Proposing a New Model: The Empathy Gradient

The model I propose brings us closer to solving that problem. I arrived at this model through an analysis of my work at a startup I founded where we were automating the optimisation of shopper marketing ads, usually a task reserved for creative talent and shopper marketing strategists. It was then evolved when trying to understand how to integrate data, tech, and then AI into the marketing supply chain and operating models.

The starting point is understanding that AI has the potential to impact all tasks, cognitive and non-cognitive, since it combines the traditional automation capabilities of algorithms with evolved analytical capabilities, like recommendation engines, with generative capabilities, like Large Language Models and image generation models. This means that AI is not only automating, it is also “thinking” and “creating”.

The next step is understanding that AI can replicate the majority of thinking and creative work, it just doesn’t do it as well as humans, for now. However, if the quality of the output is not important and the objective is e ciency of process with time or cost as the primary KPIs, then AI is either better than people or will become better than people over the coming years and decade. However, if we focus on the output, or more specifically, the outcome, then we find the source of the new model.

The Empathy Gradient: When Human Empathy Matters

There are tasks where the objective of the outcome is intrinsically human. Telling a joke, creating a work of art that resonates, identifying a human insight from data. Those outcome of tasks like those is improved by including a human because the outcome is dependant on human relatability. With that in mind, we take “Empathy” as a proxy for “being human” and can look at tasks in terms of whether empathy improves the outcomes or makes them worse.

For example, when including empathy in data collection, we find that the data is biased and the outcome is worse. However, when translating that data to a human insight, empathy improves the outcome. This enables us to draft an x-axis gradient where we can use “Impact of Empathy” as the measure, which goes from negative empathy impact to positive empathy impact. Giving us the graph shown in Figure A.

Figure A

The second step is laying out the best use of AI on that gradient from People Augmentation for the tasks that benefit from empathy to Task Automation for the ones where empathy has a negative impact on the ouctome. The gradient overlay is shown in Figure B.

Figure B

We are now able to map tasks to the axis. Current mapping is qualitative based on subject matter expert opinions, but the goal is to create a quantitative process for the mapping. An example of mapped tasks can be seen in Figure C.

Figure C

One current use of this mapping is the design of AI-based applications that apply to the work within advertising agencies. The mapping helps us understand whether to develop an application with the objective of automating the task or developing it to augment the person performing the task.

Example 1: Augmenting Insight Generation

I’ve mapped insight generation to the "High Empathy" side because when interviewing strategists and talking to data and insights companies, the current consensus is that AI can show patterns, but a person can then understand how that pattern is a useful insight. One company that is doing a good job at augmenting people is Waldo.fyi, a tool that helps strategists and researchers make sense of massive amounts of data while performing their research. This is a prime example of human augmentation capabilities.

Source: Waldo.fyi

Example 2: Automation of Ad Evaluation Through and AI-Enabled App

This example demonstrates a fully automated ad evaluation system. The system analyzes the ad content and provides recommendations on how to improve its effectiveness, particularly in terms of making a stronger cognitive impact on the audience. By automating this process, the system helps optimize the ad’s performance without requiring human intervention for the evaluation.

Contact me if you want to know more about this tool

Over time, I plan to evaluate a lot more of those tools and try to make sense of the insane number of tools out there that have the potential to augment us. I’ll keep you posted.