Bringing out the worst in employees with AI

I'm seeing adverts on multiple forums (fora would be pedantic) for AI "solutions" to improve employee efficiency. Most of these approaches seem to be based on spotting patterns in complex data - which is not surprising as that's about all today's AI is good for - and flagging these for managers to take action.

Unfortunately, the data is generally about employee behaviours at the micro level - time away from desk, time not looking at screen, time not typing - and the patterns seem to focus on things which a micro-manager would want to know, not things which necessarily impact employee performance.

Even more unfortunately, the output of these AI solutions seems to concentrate on criticising employee behaviour and penalising perceived slacking, rather than measuring real efficiency and rewarding good behaviours.

All this is understandable given the poor ability of current AI to "understand" processes and causal relationships, and the obsession of some managers with making their employees sweat, metaphorically or otherwise. But would you really want to invest in such systems, and would you get any return on such investment? I think not.

There is a huge amount of evidence to show that even in the most mundane jobs, employee engagement and enthusiasm is much more important for efficiency than time spent hands-on. If you search the internet, you will find thousands of blogs and guides, and thousands of different opinions. Academic research is rather more reliable, but equally extensive: here's an example.

In common sense terms, getting the best out of employees basically comes down to carrot and stick. A good employee should get a lot of carrot, and a bad one should get a lot of stick - but not for long as they should learn. If stick is the only option, good employees will not be rewarded and bad employees will not be motivated to change.

How does AI fit in? Well, very badly I expect. In order to identify good behaviours, an AI system would need a huge amount of specific data on what good employees do, which means that organisations need to know who their good employees are and what makes them high performers. If you already know this, why do you need an AI solution? On the other hand, if you don't know this, the temptation is to generalise, to look at data you can easily measure (time at desk, time at keyboard, number of emails sent, etc.) and assume that these data correlate with efficiency or effectiveness. This just isn't the case - and again, if it were, why would you need an AI solution?

For a large organisation, or a business which has a lot of complex data on employee performance, an AI solution might still seem to have some benefit, assuming the results were intelligently filtered by management. However, the well-publicised downsides of current AI solutions need to be borne in mind:

• - bias is often intensified by AI - ethnic minority faces and even hands have caused serious issues recently because they were poorly represented in training data
• - understanding is a problem - for example the customer buying a baseball bat who was also offered knuckle-dusters and a balaclava
• - garbage in still means garbage out - and most organisations do not have clean data

Quite apart from the potential reactions to AI solutions by your employees - is the company spying on me, what is happening to my data, where can I hide from the robots? - the impact of any statistical analysis is likely to be felt most by the outliers. Take supermarket staff as an example. The man who collects the trolleys in the car park in a very methodical way, leaving him time to stand and smile at customers, may not seem to be as busy as the one who runs around randomly or chases every stray cart. The woman who stacks the shelves carefully, ordering products by expiry date rather than just front-loading new stock, may seem less efficient if measured by square footage stacked. The person who interrupts their current task to help customers, making sure they are satisfied with the service, rather than trying to avoid eye-contact or just pointing vaguely, may be much more valuable to the business than a typical employee in their role. How would an AI solution spot this behaviour, and how should a company reward it?

If it were my business, I would consider investing in better managers, or better IT equipment, or even better working conditions for employees, before I started investing in current AI solutions. I would also look into my favourite piece of industrial psychology - the Hawthorne Effect.

#ai #aiethics #bias #wfm