During my work as a technology-business writer, I spend a lot of time reading articles about AI and machine learning. There are a lot of good journalists and publications out there. But in many cases, there’s too much hype and too little understanding of how the technology works.
Headline writers are repeat offenders. I’ve lost count of the number of click-bait titles announcing that millions of jobs will be lost in the coming AI revolution. Or that musicians, writers and painters will soon be replaced with robots.
But the biggest mistake, I think, is to assume that the AI revolution is just around the corner. In fact, it’s happening right now. The reason it doesn’t grab the headlines is that in most cases the public don’t notice it. That’s because it’s working quietly in the background, like a search engine. Or working so well that you don’t really notice, for instance the recommendation algorithms used by Spotify or Netflix.
In the past I’ve called this quiet AI or ambient AI. But in recent weeks I’ve settled on ‘competent AI’. The Cambridge Dictionary offers three definitions of competent and all of them apply here I think.
· Able to do something well
· Good at doing something because of practice
· Having the skills or knowledge to do something well enough to meet a basic standard
Any AI solution must pass this competence test if it is to be genuinely helpful at work or in the arts. This matters more than ever where the technology has the potential to do harm – witness the experimental healthcare chatbot that advised one of the test group (not a real patient) to take their own life. On a lighter note, AI incompetence is rife in the arts. If you’ve ever listened to AI attempt to imitate rock music you’ll know what I mean.
Let’s take a look at some recent examples of AI that are genuinely useful.
Robo-journalism: The Radar News Service, in the UK helps journalists to draft articles based on data sets released by the UK government. Using their investigative skills, the journalists identify data for the basis of their story and then build a template into which the data and standard phrases can be assembled. Unlike writing tools based on GPT-3, the template-based approach delivers structured, coherent articles that can be syndicated to national and local publishers.
Retail negotiations: Large retailers, such as Walmart, must negotiate contracts with thousands of suppliers. It’s a time-consuming and frequently adversarial process. But software from Pactum automates the negotiation process, taking into account dozens of parameters and determining an outcome that is to the benefit of both parties (the pareto outcome).
The most staggering statistic revealed by Pactum, is that the average Walmart supplier negotiation leaves $5,000 of unlocked value on the negotiation table. Imagine all those dollars unlocked and returned to the economy. Now imagine not just Walmart but every large retailer applying a similar approach to their negotiations.
Video conferencing: Right now, the most successful AI applications are quick, pragmatic fixes to specific problems that run behind the scenes without the end-user ever knowing.
Take video conferencing. While I’m participating, I’ll often move the computer to make space for a document or to type a message to other participants. All too often I fail to notice that I’m slipping out of view, especially when I minimize my own camera.
Microsoft fixed this problem in the latest generation of their Surface laptops by training the camera to follow the eyes of the user. The software ensures that you always appear to be looking directly at the camera, even when you aren’t. The beauty of this unspectacular solution is that it solves a widespread problem, that has affected millions of people since lockdown.
Marketing – A/B testing: To be fair to GPT-3, it works well when you limit its goals to short statements or even a paragraph or two of copy. So, it does have a useful role to play in digital marketing where the pressure to generate short form copy for advertising and email is greater than ever. This also enables you to run massive A/B testing events, constantly recombining copy and imagery until reaching a format that optimizes conversions.
In a recent study, carried out by Buzzstream and Fractl, 71% of respondents working in the media rated AI generated email subject lines as good or very good. Expect to see AI playing a greater tactical role as marketeers discover new ways for the technology to support their campaigns.
Spaceships and videos: Did you know that you can now add spaceships and planets to your outdoor videos? New AI software developed by researcher Zhengxia Zou at the University of Michigan, enables you to do just that by swapping out the sky in your video for something much more dramatic. I’ve used similar technology when editing photos, but this was very much a niche application for landscape photographers.
What Zou has done here, which is really smart, is to make this feature appealing to a much larger audience, as well as upgrading it to work for video. This to me is one of the finest examples of ‘competence’ you can find: Enhancing an existing technology for the benefit of a much wider audience.
If you’d like to get regular updates on the latest AI solutions, especially the ones that are genuinely useful to business, check out my newsletter, The Artificial Intelligencer, published every Tuesday on LinkedIn.
An archive of all the best AI stories can also be found on my Flipboard magazine which is updated daily.
Photo by Andrés Gómez on Unsplash
