software robot

Software robots should be judged by how they enhance their users

Last week, the Financial Times published an article, ‘Businesses turn to software robots for office work’, that outlined how, “Software robots have become one of the hottest fads in business automation, as a new wave of AI is poised to sweep through the back-office functions of large corporations.”.

Various numbers are thrown around in the article—a $153m investment in UiPath and a valuation of $1.1bn for the same firm, Blue Prism valued at GBP 1bn with annual sales of GBP 24.5m, subscription revenue growth of 146 per cent for Automation Anywhere, an overall market valuation of $2.9bn in 2022, and four million workers in the US likely to see their roles automated by 2021.

Firstly, as we’ve written about before, the replacement of workers with robots will not be a blanket development; the impact mostly keenly felt in specific sectors, industries, and roles. The article does correctly point out that most of the workers will be kept on and retrained to do different roles. After all, there is always a market for the quality work that only humans can do.

A striking parallel that can be drawn is between fine dining and McDonald’s. The first McDonald’s restaurant opened in San Bernardino, California in 1940 but only began to grow into the monolith it is today from 1955 when its first franchise was opened by Ray Kroc in Des Plaines, Illinois. The advent of cheaply-produced food available at low prices did not spell the end of fine dining—high-end-restaurants still exist and do good business. Nor did people’s tastes change—in recent years, after decades, the public began to turn towards ethically-produced, high-quality food.

McDonald’s responded to this by developing and opening McCafe’s, aping the other end of the market. Automation will play out the same way—the public will prize and value high-quality, individualised work even more eventually. And that’s the remit only of humans.

But perhaps the most-problematic statistic is also the article’s most-anodyne. “Each bot,” the report say, “can handle the work it would take three or four full-time workers to perform.”

That number, which originates for Forrester Research, seems incredibly low and makes one question if those bots are being run on a Commodore 64. At Retresco, we can generate individual texts, originating from extremely-complex data, of around 12 sentences at a rate of 24 texts per second. For one client, we routinely generate over 200,000 texts per day.

Forrester Research’s estimate of three-to-four full-time workers also forgets that robots do not take breaks, or get sick, or go home at 18:00. Robots can work twenty-four hours a day, seven days a week, three-hundred-and-sixty-five days a year. And while robots do require some maintenance and upgrades, it is no more than what be expected in any large organisation.

If you compare the productivity of robots to humans, humans will always lose. A more-useful comparison would be to not look at which of the two sides is better, but how best they work together. The relationship between the two should be complementary, not adversarial.

* * * *

In a few days, we will post the second part of this ongoing series where we will examine practical, real-world uses of NLG and Natural Language Understanding (NLU) applications.