The Proper Place of Humans and AI in Creative Work

The first time I witnessed a debate about the usefulness of machine translation and the human role in a world where computers do most translation was in 1995, at a summer school about computational linguistics in Edinburgh. The idea of artificial intelligence and intelligent computers replacing or displacing humans is not new, it has been in the air since the 1940s.

The "Proper Place" discussion

The debate in Edinburgh was led by the late great Martin Kay, who, as a researcher at Xerox, wrote an article called "The Proper Place of Man and Machine in Language Translation"... in 1982. Following a cue from the ALPAC report* in 1966, he argues that, rather than replacing the human translator, we should create an environment where human intention leads and computational intelligence is called upon when needed. In the second half of the article, Kay even presents the design of a translation environment that helps the user call on computer resources while writing. These principles later appeared in actual translation software, from the mid-80s to this day.

Even if I do say so myself – we developers of translation tools were, are, pretty good at respecting human intention. We have been taking extra effort to ensure that the human user gets the last word, and gets the last word early enough, if that makes sense. But even that respect and effort was rightfully questioned because the introduction of technology changes the nature of how creative work is controlled and how it exists in society.

This principle breaks when you're forced to use tools where you can't prevent the AI from running, or from interfering with your work at the end... and it's even worse if you can't even know if or when AI was run on your work. Or when you are a customer of a company and your only means to get in touch with them is an AI chatbot.

AI systems often have it backwards

My favorite example is the website hosting system that forces me to prompt an initial website and runs AI to create it, no matter whether I want that or not. That's already a lot of energy spent; and it takes even more to manually clean up the AI-generated website. I mentioned this elsewhere, but it merits repetition: cleaning up an AI-generated structure means weeding out pointless and irrelevant items the AI put there because it has no means to understand my intention.

In short, we lose energy, time, attention, and quality because an engineer decided to ignore human intention. Oh, and we damage the environment in the process.

If we are serious about keeping human control over creative processes, we must also work to uphold human judgment. And that means human knowledge, thinking, motivation, work ethic. My question to everyone who introduces AI-driven automation in their work is this: did you make sure you can still do everything without AI?

It seems that the indispensable ingredient in creative work is human intention. It simply has no counterpart in today's AI systems. But if we make humans outsource thinking and mental activity, we lose the intention and the judgment necessary to maintain human control.

Appeasing ignorance and laziness instead of stimulating thinking and ingenuity

It is more profitable to appease ignorance and laziness – which does abundantly exist in human society, despite the merits of the human race – than to design systems that incentivize human thinking and decisions. I believe that a good productivity tool will help me train my human faculties and make AI tools available when and how I want them. For example, I know I have a tendency to be verbose and redundant in my writing (although, hopefully, not the same way as an AI). If I could right-click this paragraph and ask the AI for tips of deleting words and expressions without changing the meaning, that would be great. I know you can do that in some tools.

But instead, for the most part, we need to fight temptations such as "press / for AI" so the tool will write another paragraph. So much so that I feel I'm out of line or I'm protesting when I choose to simply type my article... like now. I don't think this is the right kind of incentive and not the right kind of engineering.

If builders of technology want to agree on some discipline into AI development, it should also apply to the environment where the AI is used. I think engineers simply need to return to this principle: make sure human intention has priority and a human being can have the last word. Not only when necessary: everywhere. It is not up to the engineer or the AI to decide if human control is necessary. It is up to the human user, who must be able to decide to intervene independently of the AI and the engineer who built the environment.