No Bot BFF

We are told that travel broadens the mind. So I traveled this week to a new place: the American Sociological Association’s conference on Reimagining the Future of Work. One of the sessions I attended discussed the realities of human-AI collaboration.

The Promise of Human-AI Collaboration

“Collaboration between humans and artificial intelligence (AI) could unlock up to $15.7 trillion in economic value by 2030,” according to the World Economic Forum’s recent statement on human-AI collaboration in the workplace. That’s a dollar amount that grabs the attention. But how close are we to achieving this dream?

At this week’s conference, Jonathan Winter (European University Institute) reported on work he and his colleagues have done comparing the performance of human teams versus AI (autonomous agent / bot) teams versus hybrid human-AI teams. Which team do you think did best?

Unfortunately for the World Economic Forum, the dream team did the worst.

The Reality of Human-AI Collaboration

The tech optimists have been telling us for years that AI will make everything better. The tech skeptics warn us that AI will put us out of work. So we oscillate between excited curiosity about these powerful new tools on one hand, and existential dread of displacement, on the other hand. In the midst of all this, we’ve been assured that the lucky ones among us will find a way to work harmoniously with AI in hybrid teams that are stronger than their constituent parts.

Unfortunately, that bit of tech optimism has been disproven for the time being. When Jonathan Winter and his colleagues tested the performance of hybrid human-AI teams against teams entirely comprised of humans or autonomous agents, they found that the hybrid human-AI teams underperformed both the entirely human teams and the entirely AI teams.

So much for the dream team that benefits from the best of human and AI capability working together.

But why?

Operational Constraints?

According to Winter, some people explain this underperformance by saying there likely were operational constraints on hybrid teams:

• they might have reduced communication, lacking in both the quantity and quality of communication enjoyed by entirely human teams
• they might have impaired coordination between the humans and autonomous agents in the team
• there might be fundamental mistrust because the humans were not confident that the bot would get the job done properly

Human Team Super Powers

Winter and his colleagues thought there might be a different explanation for why entirely human teams outperformed hybrid teams: human teams have a super power, an extra social motivator in the form of teammate commitment that spurs better performance. This commitment is based on three factors:

• a shared vulnerability to the consequences of poor performance, which meant that everyone had skin in the game
• an expectation of reciprocity, which meant that an offer to help a teammate now would result in future help from that teammate
• the formation of personal bonds, which led to a genuine concern for the welfare of other members of the team

By contrast, AI is not vulnerable to the consequences of poor performance. It won’t get demoted or fired if the project does not meet expectations. Further, AI doesn’t recognize a general obligation of reciprocity so it can neither appreciate help or reciprocate later. Finally, AI doesn’t engage in interpersonal care. So it cannot strengthen teammate commitment. In other words, AI makes a lousy BFF at work.

No BOT BFF – Yet

While you may not be best friends with your human colleagues, you stand a greater chance of developing teammate commitment with them than with a bot. This is your super power.

But this may be a temporary advantage for human teams. Winter suggested that the more anthropomorphic AI becomes, the more it may be able to discern and reflect human emotion.

Will that create a bot BFF with social motivation to improve hybrid team performance? Only time will tell.

[Photo credit: Andy Kelly]